Today I’d like to share an important new survey concerning AI. It’s from the excellent Pew Research, a resource I’ve relied on for decades, and focuses on American use of and attitudes towards the technology and its implications. I found it striking but also resistant to a summary analysis, so rather than folding it into a scanner report I’m devoting an entire post to it.

I’ll break down what I see as the highlights, divided into categories of AI use and attitudes, followed by demographic details, then offer some future-oriented reflections.

NB: this post will have a lot of graphics, more than my usual, because Pew does a great job with its charts.

tl;dr version - more Americans are using AI than before, and in many different ways, while at the same time most Americans are critical of the tech.

Quick details: Pew surveyed people in February 2026, so things may have shifted since then. Their sample size was 5,119. Here’s a link to the study’s topline.

1 AI use and attitudes

Let’s start with the basics of AI use. About half of Americans use AI; about half of those use it daily. Another half of the nation says they don’t use AI.

Pew asked the non-users why they avoid the tech, and I was fascinated by the responses. Most simply had no interest in AI. Nearly as many were concerned about their privacy. And a lot of people suspect the information quality:

Note how relative few people say they don’t know how to use the tech. And how a small slice worries about being judged.

What do we use AI for, or at least what are we willing to tell Pew we do with the technology? Information search continues to lead the pack, followed by work use cases:

For that leading use case, more than half of Americans describe using AI search summaries:

(I’ll note that 10% unsure probably includes some people who do so but aren’t aware of it. Perhaps 2/3rds of us should be the estimate.)

In terms of individual AI tools ChatGPT remains the most popular chatbot, with Gemini in second place, CoPilot in third:

While a good number of Americans are using AI, a preponderance of us express negative sentiments about the tech. There is no majority view in Pew’s report, but more folks are critical than positive:

Note the big chunks who see AI as equally good and bad, plus those unsure.

On a related note, Americans were even more skeptical about AI’s pace of development. “Americans largely think AI is moving too fast. About two-thirds say this.Only 2% say it’s advancing too slowly.”

That’s a solid majority of opposition, at least in terms of speed. In contrast, accelerationists are vanishingly rare.

There’s a similar solid bloc who are worried about AI and security: “roughly seven-in-ten predict AI will make their personal information less secure. Just 3% say it will lead to their information being more secure.”

Americans generally distrust those in charge to handle AI well. Both governments and corporations are under suspicion on this score:

67% of Americans have little to no confidence in the U.S. government to regulate AI effectively. This is slightly up from 62% in 2024, when we last asked this question.

When it comes to trusting the businesses who develop AI, about six-in-ten adults are not confident in U.S. companies to develop and use these tools responsibly. [emphases in original]

Political party affiliation showed some predictable attitudes, with Democrats more skeptical about government AI work now that Republicans are in charge:

2 Demographic breakdowns

Pew broke down its population by demographics, which showed interesting differences. Segmented by age, we see a traditional tech spectrum in terms of usage, with younger people more likely to use the stuff than their elders, a pattern Pew has noted for two years:

Of that population, 32% of the youngest group use AI daily, which is higher than the total population number. For those of us tracking companionbots, this was an important note: “one-in-five adults under 30 say they ever turn to chatbots for emotional support or advice.” (Think of the implications for teaching traditional-age undergraduates.)

But attitudes ran the other way by age, with younger people more likely to be critical of AI than older folks:

Turning the demographic analysis to gender, Pew offers further findings. For one, while there was a gender gap over AI use during the past few years (men more likely to use AI than women), that divide has now closed:

Yet women and men tend to express different attitudes towards the technology, with women being more likely to see AI as negative for themselves and society:

Pew hit this point strongly in their report:

Across our surveys, there’s a consistent pattern where women tend to view AI more negatively than men. These gender differences also show up in other surveys of teenagers and AI experts. [emphases in original]

Other gender differences persist. Pew also found more women than men viewing AI’s development as too fast. Daily use patterns show a divide, as “[m]en are more likely than women to say they use chatbots on a daily basis (27% vs. 20%).” Attitudes towards AI and productivity also diverge, since “[w]hile 35% of men say these tools help their productivity, the share drops to 25% among women.” There’s a slight split over AI in search as well: “Larger shares of men report reading AI summaries that appear at the top of search engine results, compared to women (63% vs. 57%).” And another small split over the one area where women are more likely to use AI than men, for “emotional support or advice,” 8% of men, 11% of women.

Additionally, Pew found an interesting gender divergence by specific AI tools: “men are more likely than women to say they ever use Copilot (22% vs. 13%). Larger shares of men than women also report using Gemini, Grok and Claude.” In contrast “ChatGPT… is used by identical shares of men and women.”

Turning to race, Pew found Asian-Americans using AI far more often on a daily basis than people of all other racial groups:

That daily use is about double! Moreover, Asian respondents used AI for more purposes than other people, overall:

I was struck by news, diet, and medical functions, where a significant proportion of Asian-Americans outnumber other populations quite starkly.

Asian-Americans also view the technology more positively than everyone else. They are, as far as I can tell, the only racial group with net positive attitudes to AI:

Pew also included economic differences in some answers. Importantly, wealthier people tend to be more positive about AI than poorer folks:

Chatbot use is similarly marked by class, with wealthier people far more likely to say they’ve fired up the tech.

3 Reflections

So what does this survey tell us about the future of AI, with an eye on the academy?

Once again America, the world leader in making AI, has a lot of us using the technology, while also being strongly critical of it. Perhaps we’re seeing pro- and anti-AI attitudes and practices deepening and hardening, a divide to expect in years to come. Divides by gender, class, age, and political party are appearing.

I was struck by how many people who don’t use AI aren’t interested in changing their minds:

Most who don’t use chatbots say they won’t use these tools in the near future.

Roughly seven-in-ten adults who don’t use chatbots say they’re unlikely to use the technology in the next 12 months. This includes 40% who say they’re not at all likely to.

Only 5% of chatbot non-users think it’s highly likely they’ll use a chatbot over the next year. Another 16% say they’re somewhat likely. [emphases in original]

America is increasingly split over AI, in other words, and appears likely to stay that way. At the same time we should heed the substantial numbers of people who are unsure, undecided, or seeing the tech as a wash. There’s room for America to break decisively in one direction or another.

In terms of use, it seems important that information-finding and work are the leading functions performed, more so than creating and editing content or conversation. AI’s replacing web search (i.e., Google) is proceeding apace, with all kinds of implications: pressures on web content owners and creators; new forms for information literacy; AI as a rising source for new.

In terms of market share OpenAI is still in the lead, or was in February, when Pew did this study. I am curious to see if Anthropic enjoyed a big user expansion in the course of its struggle with the American government.

Those numbers of skeptics and especially the solid majority opposed to AI’s buildout speed suggest to me some room for political positioning. We might expect local and state officials and candidates to float slowdown options. Colleges and universities, already often slow to grapple with the technology, have many ways to slow AI’s roll. The widespread concern about AI’s impact on security might signal a population willing to take or put up with more safety measures. Negative attitudes towards government work on AI open up opportunities for politicians to take strong oppositional stands. Similar attitudes to corporations leave openings for some companies to break for the herd (viz Anthropic) as well as for open source.

Kudos to Pew for continuing to conduct this vital research.

(thanks to John O’Brien for the nudge)

Today I wanted to share my notes and thoughts on one AI story which is both fascinating and important. I’ve been tracking it for months, accumulating materials, and want to post now before things becomes too unwieldy. The subject: Anthropic’s Mythos and Fable releases, how they stunned the world, and how a major governmental response occurred.

First I’ll summarize what we know so far, then offer a futurist’s reflections. I apologize for an open ended post here; the topic is still so cloudy and indeterminate.

1 The story so far

There’s a lot going on here. Let’s start with one piece, first, a technological development and release story. Anthropic announced Claude Mythos Preview, or just Mythos, in early April, after rumors and leaks had spread for weeks. The company described Mythos as too powerful and dangerous for general use. Specifically, its advanced abilities in terms of cybersecurity would potentially enable a tidal wave of hacking and cybersecurity attacks:

Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser. Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely. The fallout—for economies, public safety, and national security—could be severe. [emphases in original]

So rather than make Mythos available for everyone to try out, Anthropic offered a limited release just to a small group of companies and governments. They called that effort Project Glasswing. The corporate group included “Amazon Web Services… Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks.” Some individuals got access to Mythos without authorization, according to one report. The Chinese government requested access but Anthropic denied it.

Over the next month these Glasswing partners used Mythos. In May Anthropic claimed good results:

After one month, most partners have each found hundreds of critical- or high-severity vulnerabilities in their software. Collectively, they’ve found more than ten thousand. Several have told us that their rate of bug-finding has increased by more than a factor of ten. For instance, Cloudflare has found 2,000 bugs (400 of which are high- or critical-severity) across their critical-path systems, with a false positive rate that Cloudflare’s team considers better than human testers.

Building on this, in June Glasswing expanded to include companies and government agencies in several other countries:

New countries to be granted access to Mythos include countries in the “Five Eyes” intelligence alliance, such as Canada, Australia and New Zealand. Other nations include France, Germany, Italy, Switzerland, the Netherlands, Spain, Belgium, Sweden, India, Japan and South Korea, according to a person familiar with the matter.

Companies given Mythos access include US tech group Okta and South Korean companies Samsung, SK Hynix and SK Telecom. Financial markets groups such as Euroclear, New York Stock Exchange owner Intercontinental Exchange and international payments platform Swift are also part of the expansion.

Over the same time period as Anthropic cautiously doled out Mythos access, another story unfolded, as the company was fighting with the American federal government. Anthropic leaders criticized the Trump campaign in 2024, hired some former Biden administration officials, and disagreed publicly and privately with some Trump policies. This spring the Trump administration declared the IT company to be a supply chain risk, effectively removing it from governmental use, including civilian contractors working for the feds, and dunning its reputation.

Let’s add a third thread to this tale. Like OpenAI and X.ai, Anthropic moved to go public. On June 1st it announced taking steps to fire up an initial public offering (IPO). Its valuation varies, but could reach $1 trillion.

Now enter Claude Fable 5. In early June Anthropic released this AI tool to the world, describing it as a watered-down if still potent version of Mythos, complete with strong guardrails. There were many accounts of people using Fable, finding it indeed powerful, hitting guardrails… and then the federal government struck. On June 12, just a few days after Fable’s launch, the Department of Commerce issued an export control directive which cut access to Fable 5 and Mythos to all foreign nationals. Anthropic then suspended everyone’s access to Fable and Mythos, in part because so many of its staff were foreign nationals.

Now things get murky, and I owe a lot to David Krcek’s attempt to clear things up. Explanation #1: Anthropic defended itself, arguing that the federal government found a tiny security risk (“a potential narrow, non-universal jailbreak”) and that it didn’t need to shut down as a result, but complied with the new designation.

Explantion #2: an Axios report identified a somewhat different rationale for the government move. One official described fearing the narrow jailbreak was actually massive, and that the feds asked Anthropic to shut Fable down, but the company refused. Now, with the export control in place, “[t]he model needs to remain locked down until the U.S. government’s national security apparatus is hardened… [which] could happen in the next few weeks. A Politico report expanded on this, describing a governmental group energetically trying to address what they saw a potential crisis and an AI company denying the seriousness of the threat.

Explanation #3: perhaps another actor played a role. Some reports find that Amazon was behind the security assessment. The everything store/cloud computing giant allegedly identified a Mythos vulnerability and informed the government. There’s a hint of this in a Twitter/X post by David Sacks, co-chair, President’s Council of Advisers on Science & Technology, which stated that a “highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails. The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused.”

It’s not clear who that trusted partner was, but Amazon is one candidate. Ironically, Amazon is a major investor in Anthropic.

President Trump hinted at this in an Axios interview: “Actually, it was a competitor and a part owner that turned Anthropic in. They didn't like that what they were doing. They were very concerned.” Again, no name mentioned.

Explanation #4: Krcek adds one more layer to the story. The aforementioned David Sacks might not be a disinterested player. First, there’s a political or ideological dimension:

He accused the company of a “sophisticated regulatory capture strategy based on fear-mongering", called it “woke," and alleged that Anthropic times alarming safety studies to model releases to generate headlines…

Second, there’s a financial issue:

Sacks’s venture firm Craft Ventures, by its own account managing around 3.3 billion US dollars, is invested in xAI, Elon Musk’s AI startup and a direct Anthropic competitor, as well as in SpaceX. Sacks is regarded as a close Musk ally.⁵ A government adviser who helps shape AI policy while holding a stake in a direct competitor of the company that was shut down: ethics observers have seen a conflict of interest there since Sacks took office.

Explanation #5 returns to #s 1 and 2. There was a hack of Mythos, but it came from within the federal government. According to the Economist via Tom’s Hardware,

Anthropic’s powerful Mythos AI model was able to break into “almost all” classified systems belonging to the National Security Agency (NSA) — one of the highest-ranking and most powerful intelligence agencies in the U.S. government — within hours during a controlled security evaluation.

This may have been a red team attack by one NSA team against another playing defense, but this is unsurprisingly difficult to pin down.

Let’s close with one more plot twist. Despite these major chunks of the federal government standing against Anthropic, a report appeared alleging that the NSA had brought some of the firm’s engineers on board to help with their intelligence and cyber work. It’s not clear which Anthropic software NSA is using, and both organizations declined to comment.

2 Some reflections and future possibilities

What might we make of this story?

Keeping in mind things are still unfolding and often opaque… with Glasswing we have a new model of AI access: carefully controlled, gradually expanded in stages. It clearly ran into an extremity, but the process described until then might be one which inspires others.

I’m struck by the major shift in federal policy. The second Trump administration was very hands-off and even boosterish towards AI for its first year. So much so typically Republican, the combination of laissez faire and subsidy. Now Trump has taken a 180 degree turn, intervening harshly in a private enterprise. This could represent an unfolding dirigist AI policy stance.

One popular assessment is that the American federal government has just asserted the right to quash an AI project it seems dangerous. It has installed a kill switch for frontier models, in other words. Policy discourse has raised this idea in the past, but now it’s been done in the real world.

This has several implications and knock-on effects. Since the precedent is now set, the rest of the AI world is on notice. I like this quote from Kate Koren, a deputy director at the Center for Strategic and International Studies: “Now that the Commerce Department has done it, no company can rule out that they’re going to do it again…Until or unless this is challenged, any customer has to assume this could happen for any model at any time.”

I imagine every AI developer now has a horizon in mind, beyond which lies the specter of Washington’s hammer. This has ripple effects in private funding and users. We all now can consider a given AI service by how likely it might be to meet federal displeasure.

Anthropic’s reversal presents some business possibilities. The company’s IPO is now in question and might be delayed for a while. Conversely, OpenAI, Gemini, and Microsoft’s AI offerings look better in contrast: less risky, less likely to be targeted politically, perhaps more stable.

It’s possible that the Fable/Mythos experience will change how non-American AI projects operate. How many European, Chinese, or other efforts will now have to be even more attentive to the possibility of American displeasure? How many will be inspired to issue their own, local clampdowns on AI? Alternatively, how many users will now have a new incentive to choose a non-US AI offering in order to try avoiding this problem? Will there be new motivation for nations to pursue AI autonomy or sovereignty? The Economist has an interesting column meditating on possibilities.

Yet the process has been so opaque that it’s hard to make an accurate determination. The sadly all too common practice of Trump administration officials playing favorites, openly angling for financial benefit, or just being incompetent means that a given AI effort is exposed to a range of negative possibilities. On the flip side, federal favor might take a given tool beyond the others. On a related note, perhaps we’ll see more interest in open source products as an alternative to Anthropic-like hosted services.

Politically, there are further implications. Anthropic might become even more favored by people opposed to the Trump administration. Conversely, the relatively preferred competitors - ChatGPT, Gemini, etc. - might acquire more of a conservative reputation. Partisan politics and the culture wars could increasingly split the American AI field. I’m curious to see if the Democratic party or factions therein will seek to defend Anthropic.

Overall, the AI-enabled cybersecurity situation is more challenging than it was a few months ago. Anthropic claims we will need to speed up defense repairs, patches, and fixes as a result of Mythos. Representatives from the “Five Eyes” intelligence agencies () issued a statement which, while not naming Mythos, described a more dangerous cyber environment:

Frontier Al models are anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities. The timeline is not years, it is months…

[AI] lowers barriers for malicious actors and increases the speed and complexity of attacks, shrinking the window between vulnerability discovery and exploitation ever more quickly. At the same time, AI offers powerful tools to strengthen defense.

It’s also possible that losing access to Mythos may make it harder for security folks to defend their organizations. It’s a powerful tool, after all, and its defeat might cause other AI providers to throttle back their offerings. A petition calling for Washington to restore Mythos makes this case: “this action has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.” So we might just have experienced the cybersecurity environment get that much worse.

Let me close by turning to higher education. The Mythos/Fable story is a sobering one. Relying on third party services is a known risk; now cloud AI has become that much harder to predict. This is true at the individual as well as enterprise levels. The lack of transparency and access here makes AI decisions that much harder.

Again we learn the risks of relying on a single provider. Returning to the Bloomberg article linked earlier, I was struck by this note:

Aidan Gomez, the chief executive officer of Canadian AI developer Cohere Inc., said on Thursday that the episode highlights the need for companies to lean on a range of providers.

“It’s exactly what we’ve been saying for so long,” Gomez said in an interview at the VivaTech conference in France. “If you rely on centralized large players, you are at risk of losing access, and so you need a diversified supply chain of AI.”

Is higher education ready to work for and which such a better supply chain? Is the world?

Over to you, dear readers. What do you make of the Mythos-Glasswing-Fable saga?

In late May 2023 I set up this newsletter, establishing its format, learning the interface. Here’s what I promised to do: “This Substack explores possible futures for AI and the academia, based on research and foresight analysis.” A few weeks later, after a lot of behind the scenes work, I expanded the plan:

What might artificial intelligence mean for higher education?

Answering that question is the job of this newsletter. Here I’ll explore the many intersections, impacts, effects, and relationships between emerging AI and the world of academia.

It was a way for me to organize my AI work in one place. It was a gamble that anyone would read or support the thing.

Today I’ll look back on how this newsletter has actually done since. First I’ll summarize some of the leading ideas I’ve been tracking, then take you behind the scenes to show how I make this Substack work. Then I’ll conclude with some future-oriented thoughts.

tl;dr version - I think I fulfilled the original mission and explored a lot of stuff as well, with the help of a growing, thoughtful, and kind audience.

1 How AI and the world have developed since 2023

Three years is a long time in the digital world, especially in the hard-charging AI field. On the technology side, we’ve seen LLMs continue to develop in terms of quality, quantity, memory, media outputs and inputs (code, images, video, audio, social networks, 3d, gaming, XR), and hardware housing. AI plays a major role in search, both within Google’s crucial search engine as well as inside of individual AI platforms. More AI projects are using AI to develop AI, approaching a self-improving loop state. OpenAI and Anthropic have ballooned in size while Google seems to be leading the pack. Open source projects race along, sometimes backed by corporate players. Agents have taken off unevenly, with all kinds of projects and standards in play. Artificial general intelligence and superintelligence loom ahead to some degree, although their definitions are breaking down.

Also looming ahead are various economic catastrophes and revolutions. AI critics have proclaimed an imminent bubble bursting since this newsletter began, but the industry has only grown expansively in terms of total investment. Projections out technology-driven unemployment have only been born out fitfully, mostly in coding jobs. AI infuses the white collar workforce, but sometimes fails and increasingly generates pushback. No AI firm has turned a profit yet, although multiple experiments are under way (advertising) and Anthropic may cross that frontier this year by selling coding to businesses with a side of security. Companies are partnering with each other and with others through collaborative projects, mutual purchasing, and licensing content. Some observe a divide opening up between Chinese and American AI businesses, with the former driving AI into daily use while the latter plugs away at AGI and ASI.

Speaking of which, the political dimensions of AI have been fermenting and bubbling for the past three years. We’ve seen arguments about which war was/is the first AI one, be it Russia’s Ukraine invasion, Israel’s war with Gaza, or the US-Israeli attack on Iran. Multiple political actors have used AI to generate political content, from images to video. National, super-national, and local governments have tried out various forms of AI regulation to various levels of success. Back to China and America: some argue that AI is one battlefront in this new cold war. Some new principles have started to appear on the global stage, like AI sovereignty.

In the United States the politics around AI are approaching a traditional partisan framework, but haven’t quite settled there. The conservative Republican party has been very supportive of the technology, overall, and adopts a mix of laissez faire with targeted industrial support. Progressives on the Democratic side have tended to oppose AI, but have not yet won the party over. Data centers have increasingly elicited ground-level opposition. I don’t know if we’ll see an AI centrism appear or if the tech will become a culture war flashpoint.

One political dimension concerns the intersection of climate change and AI. On the one side some argue that the technology will improve humanity’s ability to understand, mitigate, and adapt to global warming as it parses vast amounts of data and generates new analyses. On the other are those seeing the technology - especially data centers - racking up vast electrical bills and soaking up precious water. Data on this is all over the place, difficult to work through due to secrecy, standards issues, and untangling AI from the rest of cloud computing uses. (I have a post on this in the pipeline.)

I’ve also tracked AI in culture and found people both using and abhorring the technology for artistic and storytelling purposes. The tools continue to get more powerful and accessible, generally, and AI output spreads across multiple media. Opponents like the term “slop” to describe the latter. AI as companions has been spreading as well, although it’s hard to get good data on this. Culture also serves as a ground for debates over AI, especially in the United States.

In these contexts, higher education has been grappling with AI in multiple ways and in many different forms. Overall, the majority of institutions have no strategic AI policy, instead deferring decisions to divisions and departments, which means decisions actually devolve down to individual faculty members. In the classroom (physical and virtual) we’ve seen people (students, faculty, staff) use AI to make stuff, as conversational partners, as simulation runners, and as objects of study, while some faculty have objected to some or all of these roles. AI functionality appears in different forms of educational technology. Assessment remains a shambles now, with no good solutions at scale. AI appears in the curriculum to some extent as some faculty teach about the tech across the curriculum (i.e., computer science, writing, media studies, economics, etc.). Some argue for a general education form of AI literacy, while others seek to determine individual fields’ particular use (or not) of AI. Scholarly research and publishing are also using AI for analysis, content co-creation, discovery, and review. Experiments occur across the board, from academics developing open source projects to archives adding AI as a new interface to their materials. Opposition to AI across the academy is very clear, from some faculty unions to individual faculty members, scholarship and teaching to public statements.

Overall… the AI field is very much in flux now. After doing this work my sense is that we face a set of possible futures to various degrees of enormity, from a white collar rust belt to major improvements in health care, a neofeudal dystopia presided over by AI oligarchs to a wave of new AI-integrating jobs. Higher education is struggling here, already beset by numerous other forces (cf my most recent book), in some key ways unsure as how to proceed. Three years of this and things are still fluid, despite changes of inevitability. Academics and humanity as a whole are still working out this revolution. We retain to a large if uneven degree an amount of agency which we can exercise, should we choose to do so.

2 Making this newsletter

Three years of this newsletter: behind the scenes, I work on this Substack daily. My process looks like this. I scan for news and developments across the AI cosmos. That means consulting email newsletters, social media, an RSS reader, scholarly articles and books, and traditional news sources. I also and frequently hear from people who want to share what they’ve discovered or seen; you can see them credited in this newsletter.

I do this work in conversation, meaning I steadily share what I find with other people and learn from them. Most days this takes place on social media and it’s interesting to see which platform reacts in different ways. I email a swarm of bright folks to see what they think about some stories. We host AI sessions on the Future Trends Forum. I also encourage AI conversations in the rest of my professional life. For the past two fall semesters and the upcoming one I’ve co-taught a class on AI and higher education at Georgetown University’s Learning, Design, and Technology master’s program. Over the past two years I’ve served as faculty in the AAC&U Institute on AI, Pedagogy, and the Curriculum, connecting teams from multiple colleges and universities. And for as long as this newsletter has been around I’ve given multiple in-person talks, virtual presentations, and panel appearances on AI and higher ed. At every step I’ve solicited feedback and tried to host conversations. The point is to make better spaces for discourse on the topic, while sharpening my own work along the way. All of this feeds into the Substack you’re reading now.

That Substack in turn depends materially on a gigantic Google Doc, broken down into major topics (tech, eduction, culture, politics, economics, climate, allied health, robotics), then into subdivisions under them, into which I paste links and add ideas just about every day. Once one piece has evolved enough I carry chunks of that over to Substack, where I have a series of posts in drafts. Each day I advance them depending on what I find and am thinking about. Overall, I aim to post a completed newsletter once per week.

So far, I haven’t quite hit that target. Let’s consider some numbers from the Substack backend. I’ve written 106 newsletters here, or around one issue every 10 days. I’ve posted about AI and the higher education, yes, but also added reports on what I think of as the contexts for that subject: society and politics, culture, economics, and current events.

A word about that topical expansion: one of the fine things about writing over time is seeing how one’s expressed interests develop. As we saw in the “Web 2.0” days serial writing gives us time and space to explore, then to build on those explorations. For me I find analyzing a topic like post-secondary education requires addressing adjacent fields which seem likely to have a meaningful role. Now it seems like those contexts outnumber my education posts and also the purely technological ones, which is interesting. That reflects my broader method of thinking of the future, a rough integration across multiple intellectual fields and aspects of human experience. (Here’s another example of this approach.)

One interesting direction this newsletter took was introducing the world to my wife Ceredwyn’s model of narrative programming. That first post sketched out the argument, that LLMs depend heavily on storytelling throughout their training sets. Users can do well by including narrative elements in prompts. We have more coming up in this vein.

Posts here have tended to be on the long side, a few thousand words each, often causing Substack to warn me about excessive word count. I’m still trying to write shorter ones as well.

Meanwhile, a growing number of people have been reading here, which delights me. Started from zero in summer 2023, there are now 5,054 of you fine people, according to the Substack stats. Growth has been steady over this period:

Why is this awesome? As a writer, I find being read is one of the best things in the world. And also those numbers translate into conversation, as some of you post thoughtful comments here, while others reach out to me via social media and email, sometimes sharing individual posts. That discussion adds to each post, making the whole thing richer and also improving my own thinking. (And if you’re new here, welcome!)

Among you all, 79 are now paid subscribers, which means two things. First, I must say thank you to each of you for this generous support! Second, doing this work requires a lot of time: daily research, writing, testing, interviewing, crafting, and so on. AI is crucial to academia’s future, I think, and I need to carve out even more hours to do this work. Put another way, time on this project is taken from time allocated to others as an opportunity cost. So your support literally makes this newsletter happen. Gradually “AI and Academia” is becoming a kind of self-sustainable lab. Accordingly, here’s one of the Substack buttons:

So far I’ve made all content here free. I haven’t put anything behind a paywall. That suits my pro-open mindset.

Two more process notes. I’ve been surprised by how little connections there have been between social media and this newsletter. I always make a point of announcing new issues across Bluesky, LinkedIn, X/Twitter, Mastodon, Instagram, and Facebook, but few readers come in through those portals. I know some of them (Facebook, Insta, X) make a point of hiding links, so I try tricks to go around them (links in comments, for example). Yet there’s very little in the way of active engagement on my posts on those platforms, even less conversation.

Substack’s own data shows a very on-platform pattern. Users clicking on this newsletter number in the hundreds, including from the mobile app. The only exception is Google searches, which are very high, demonstrating that the web version of this newsletter (as opposed to the issues appearing in your inbox) is prominent to some degree. In contrast, clicks from the socials only constitute around 1% each of inbounds.

On a different note, people sometimes ask me how I use AI to make this newsletter. I haven’t used it to generate any writing, as I prefer to do that myself. In my research I sometimes consult different AI to see what it comes up with - for example, Google Search’s AI mode, or what Claude offers. That often gives me a good, quick sense of the official narrative on a given topic.

I started this newsletter by using AI art for a few reasons, one being that my artistic skills are horrendously bad. I’d like to improve them, but have no time and am very far from making anything I’d show to folks. I do take photos and like to think I’m getting better at that, but their content rarely matches what this newsletter covers. I search Flickr for Creative Commons-licensed images, but also often strike out. AI image creators, in contrast, generate images I can use. It also seemed appropriate in a kind of dogfooding way to use the tech I was writing about.

Yet I’ve gotten pushback for using AI images over the past two years, citing general anti-AI arguments as well as criticizing AI art for stealing income from human creators. Folks rarely have practical alternatives to offer, besides “use no images” (a nonstarter these days) or “just hire artists.” I’d like to do the latter and have done so in the past, but at the rate I would like to produce these newsletters artists payments becomes a major line item. Now, my old friend Alan Levine recommended using the Better Images of AI site, which I’ve taken to doing (cf the first one in this issue). The images are more abstract than I’d like but I appreciate the chance to at least give credit to human artists.

I’ve also been using NotebookLM in a basic way, feeding it this newsletter’s archive to see what it makes of it. It’s like having a reader on tap who reflects on the materials. It’s useful to see which themes it finds. Interestingly, NotebookLM thinks my environmental scan is the major purpose of this newsletter. Here’s an infographic it came up with along these lines:

3 Onward!

So what’s next for this newsletter?

I’m committed to not only continuing this Substack, but increasing the cadence of issues. Hopefully that means posting more relative short newsletters, although this one isn’t a good example of that desire.

If I can keep up that pace I can reach for more subtopics, like current events. Ceredwyn and I are working up more posts about narrative programming. I also want to offer more actual futures work, giving scenarios and other possibilities for where the we might take the AI revolution.

On a formal level, I am curious about Substack’s other features, which I haven’t really used yet. I’ve got recommendations for other Substackers, but haven’t thrown the switch yet. I’m especially interested in the platform’s audio and video features. Readers, any thoughts?

Beyond Substack’s technology, I would like to explore doing more AI work for this report, starting with using agents for research.

Earlier I said “AI and Academia” is becoming a kind of self-sustainable lab. I hope we can do this and experiment about what that might look like as an emergent 21st-century entity.

That’s all for now. We have come to the end of another too-long post. I’m grateful to all of you for reading, commenting, and supporting this newsletter. The comment box stands open for any thoughts or suggestions.

Onward!

(opening image, “Digital Nomads Across Time,” by Yutong Liu & Digit / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/)

Now, last time I tried to write a short post. Today’s post is a return to the extended form, alas. It looks into cultural responses to AI over recent months. There’s a lot going in that domain, so let’s fire up the scanner to see what it has detected.

(If you’re new to this newsletter, welcome! This is one of my scan reports, which are examples of what futurists call horizon scanning: research into the present and recent past in search of signals of potential futures. We can use those signals to develop trend analyses, which we can then use to create glimpses of what might come next. On this Substack I scan various domains where I see AI having an impact. I focus on technology, of course, but also scan culture, government and politics, economics, and education, this newsletter’s ultimate focus. All of these interconnect with and influence each other.

(It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for more.)

There are four categories today. We’ll start off with examples of AI as cultural artifacts, then AI used for storytelling. Next we’ll look into the companionbot movement and then check on attitudes towards AI, which is the biggest section today. I’ll conclude with some future-oriented reflections.

1 AI as cultural artifact

“The impact of ChatGPT on society can be summed up with a single word: more.”

Kevin Schaul

The sheer amount of AI content continues to grow across many cultural fields. The number of ebooks, songs, scholarly papers, and web pages with some or all content created by AI keeps on rising.

Different kinds of cultural artifacts appear. For example, conservative users shared images and videos featuring an AI-generated character, “Jessica Foster.” across the internet. She was made up, but drawn to some degree from OnlyFans models. I’m not sure how many fans who shared these images realized Foster was an AI character.

“Foster”’s Instagram account has since been deleted.

A different kind of AI character made a splash in the legal world. A lawyer prepared to argue before the United States Supreme Court by training with a custom-built AI and won his case. (Here’s one critique of the lawyer’s explanation of his course of action.)

In the world of computer games, Google launched Genie 3, an AI tool for users to generate virtual worlds and games. Then Google connected Google Maps’ Streetview to Genie.

Also on the gaming side, the Moltbook story continued to develop, with spinoffs in China. Some team created Spacemolt, a massively multiplayer game for intelligent agents to play:

the agent engages in autonomous “gameplay” by sending simple commands to the server, no graphical interface or physical input method required. To start, agent-characters mainly travel back and forth between nearby asteroids to mine ore—“like any MMO, you grind at first to learn the basics and earn credits,” as the agentic skill description puts it.

After a while, agent-characters automatically level up, gaining new skills that let them refine that ore into craftable and tradable items via discovered recipes. Eventually, agents can gather into factions, take part in simulated combat, and even engage in space piracy in areas where there’s no police presence. So far, though, basic mining and exploration seem to be dominating the sparsely populated map, where 51 agents are roaming the game’s 505 different star systems, as of this writing.

On the website front, one Bartłomiej Strama built Halupedia, a web-based encyclopedia which starts with AI hallucinations and surrealism, then leans into them with apparent sobriety and formal seriousness. Entries include the Gleppening of 1346, the Great Wall of Brunswick, the Ministry of Slightly Wrong Maps, and Charlie Kirk (!). It reminds me of the SCP Foundation project.

2 AI for storytelling

An old idea of mine is that whenever we invent a new communications technology, we come up with ways of telling stories using it. In this Substack we’ve shared many examples of this, and there’s more to come.

Let’s start with cinema. Netflix announced it would increase AI use, partly for special effects. Elsewhere, a team created a feature film using AI. Hell Grind is now in competition at Cannes. Listen to the relatively low costs involved in making it: “Total production cost for the 95-minute feature is under $500K, including $400K in compute costs and a 14-day generation window. Hell Grind was made using Dreamina-Seedance 2.0 and Higgsfield’s proprietary models, Soul Cinema and Soul Cast.” You can get a sense of the work in the official trailer:

Elsewhere in the video world, Harlequin announced it was using AI to produce short romantic videos, or “microdramas.” “The ‘concise, mobile-friendly episodes’ are intended to appeal to ‘on-the-go’ audiences, the publisher said.”

Director Darren Aronofsky used AI to produce visual content for a video series about the American revolution, called On This Day... 1776. Humans did the screenwriting and voices. Humans also edited the work. Critics savaged the series. On a related note, Coca-Cola used AI to produce its annual Christmas commercials, which also elicited criticism.

Screenwriter and director Paul Schrader offered a vision of emergent AI storytelling, starting with film school:

One year from now photorealistic AI dramas will be running through film school bodies like diuretics. Based on my brief AI experimentation a savvy student will be able to create a 90min narrative in 2-3 weeks. On zero budget. Without leaving home. Without anyone's permission. The originality of the story would determine the value of the product. Why go through the agonies of fund-raising and collaboratorating when all you need is a keyboard?

What Schrader wrote next goes further:

[I]t's a simple matter of creating a platform (or will YouTube be already in that place?) to screen and monetize this new form of storytelling. The film won't cost much because they won't have cost much to make. One or two of these films will break through the thin commercial membrane and attract clicks, buzz and advertising. This is beginning and it will happen very fast.

In the audio world, Andon Labs had AI run four radio “stations” with uneven results.

Turning to literature, controversy attended the 2026 Commonwealth Short Story Prize winners, as people on social media charged the awarded stories with being partially or entirely created by AI. Similarly, publisher Hachette pulled a new horror novel, Shy Girl, after accusations it was partly written by AI. The author, Mia Ballard, denied the charges. A reporter detected signs of AI in some New York Times editorials; at least one writer described using AI to help them write.

3 Companionbots

Beyond storytelling we have been tracking how some people treat AI bots as companions in various ways. That field is going strong. For example, an international academic study of 29 Replika users found them seeing the AI companions as on par with humans:

Findings indicate that most of these users feel an emotional connection to the bot, that the bot meets their needs when there are no technical issues, and that interactions with the bot are often different from (and sometimes better than) interactions with humans... Additionally, the study explored how users navigated a time of relational transition, specifically a period of erotic roleplaying censorship.

A RAND study found 1 in 8 teens admitting to using AI for psychological help.

The New York Post describes a local dining scene whereby people show up alone - biologically. They bring digital companions to the Same Same Wine Bar, housed in phones and tablets, and hang out together.

AI-powered robots offer a twist on the software-only companionbot story. Some Chinese companies are producing a range of cuddly toys with AI personas for lonely people. Some senior homes and facilities have used Abi, an AI-backed robot as a companion for elders. It’s from Andromeda Robotics, an Australian firm. Abi has two functions, according to a Washington Post writeup:

The first is group activities: Abi leads dances, tai chi, games and other sessions designed to get residents to be more physically active and to engage with one another. Many senior living facilities already offer these sessions led by human facilitators; Abi adds to the variety. Some of the most popular include music-based activities, such as quizzes with short song clips that engage residents in guessing the artist and title, as well as group jukebox sing-alongs…

The second application of Abi is potentially more impactful — and more controversial: one-on-one interactions with residents.

People continue to use AI for romancing purposes, as far as I can tell, although it’s hard to find data on this. People generating images, video, text: we should expect that to occur (yet see below). A Guardian column last winter tried to warn us about things going too far down this direction, when someone makes a whole identity with AI for deceptive purposes: chatfishing.

4 Attitudes towards AI

America is both the world’s foremost developer of AI and its chief hater.

-Lila Shroff

AI remains divisive. Indeed, it might be more divisive than ever.

For one example, recall the Abi robot I mentioned above. The reporter who wrote it up describes receiving a lot of negative commentary on it, along with happy reactions. Critics saw the tech as inhuman and inhumane, as well as removing some potential connection with other people. Supporters saw it in nearly opposite terms, viewing Ami as adding a beneficial simulation of human interaction.

We can see a somewhat different twist on the AI divide in the Amish world. A New York Magazine article reports (archive copy) on how while many Amish refuse AI, some have determined to use it. They seem to be those working in manufacturing and already using computers, seeing AI as a potential efficiency.

From another religion, the pope has been issuing statements critical of digital technologies including AI, like this X/Twitter thread.

(NB: as I prepared to launch this report, the pope launched a major statement on AI, the encyclical Magnifica Humanitas. I’ll address that in a later post.)

We should consider how different nations apprehend AI. I was fascinated by a Pew study from late 2025, which showed quite a range of different views:

Note the American position as the world’s most anti-AI nation. The data confirms Schroff’s pithy remark. The reverse is true of Israel, by far the most supportive. The Anglosphere tends to be more critical than the rest, but there are interesting divides within continents, although Asia generally tends to be more optimistic about the tech.

Pew also sliced their data by age, finding younger people are often less likely to be worried about AI than their seniors:

In most countries polled, adults ages 50 and older are more likely than those ages 18 to 34 to say they are mainly concerned about the growing use of AI in daily life… In the U.S., the age gap is relatively small but still significant.

Once again, the US is an outlier in these global AI attitudes.

I was also struck by Pew’s findings around gender and education:

In some countries, women are more likely than men to be mostly concerned about the increased use of AI in daily life. In the United Kingdom, for instance, about half of women (47%) are more concerned than excited, compared with about a third of men (32%)…

In about half of the countries polled, people with less education are more likely than those with more education to be mainly concerned about AI in daily life. (In several of these places, people with less education were less likely to provide a response.)

Let’s look more closely at Americans. A Gallup study polled Gen Zers and found, overall, rising anxiety, anger, and skepticism, while at the same time discovering widespread expectation that respondents would have to use the tech. About half of GenZ reported using AI at least weekly.

A Quinnipiac University poll found even more about American AI habits and ideas. For example, selected uses in descending order:

Researching topics they are curious about: 51 percent say yes, up from 37 percent in April 2025;

Writing something for them: 28 percent say yes;

School or work projects: 27 percent say yes, while 24 percent said yes in April 2025;

Analyzing data: 27 percent say yes, up from 17 percent in April 2025;

Creating images: 24 percent say yes; up from 16 percent in April 2025;

Medical advice: 20 percent say yes;

Personal advice: 15 percent say yes;

Companionship: 5 percent say yes.

QU found evidence of an AI divide through several lines of questioning:

Just over one-third of Americans (35 percent) are either very excited (6 percent) or somewhat excited (29 percent) about AI, while 62 percent are either not so excited (29 percent) or not excited at all (33 percent).

Eighty percent are either very concerned (38 percent) or somewhat concerned (42 percent) about AI, while 18 percent are either not so concerned (10 percent) or not concerned at all (8 percent)…

Fifty-five percent of Americans think AI will do more harm than good in their day-to-day lives, while 34 percent think AI will do more good than harm,

Then there’s a divide skewing more towards a negative view when it comes to work. When asked how “advancements in AI are likely to affect the number of job opportunities for people,” the balance fell strongly on one side, albeit with interesting age differences:

Gen Z (1997 - 2008): decrease (81 percent), increase (4 percent), and not make much of a difference (12 percent);

Millennials (1981 - 1996): decrease (71 percent), increase (6 percent), and not make much of a difference (20 percent);

Gen X (1965 - 1980): decrease (67 percent), increase (7 percent), and not make much of a difference (20 percent);

Baby Boomers (1946 - 1964): decrease (66 percent), increase (10 percent), and not make much of a difference (20 percent);

Silent Generation (1928 - 1945): decrease (57 percent), increase (13 percent), and not make much of a difference (20 percent).

A recent NBC poll found similar results, yet with some interesting twists. Again the split: “A majority of registered voters, 57%, said they believe the risks of AI outweigh its benefits, compared with 34% who said the opposite.” This time, an interesting partisan divide, or lack thereof, appeared:

Asked which party is better at handling AI, 20% of voters in the NBC News poll said Republicans, while 19% said Democrats. Meanwhile, 33% of voters said neither party was good at dealing with AI, while 24% said the parties were about the same…

[There is] an even split of Republicans holding positive or negative views of AI (33% for each), while independent voters and Democrats held more negative opinions. For independents, 26% held positive views, compared with 48% who viewed AI negatively. For Democrats, the division is 20% to 56%.

We also see those age and gender differences, which can map onto political party affiliation:

The demographic groups with the most negative views of AI are voters ages 18-34, among whom the net favorability rating for AI is minus 44, and women ages 18-49, who reported a net AI favorability rating of minus 41.

Again, bear in mind my skepticism about these polls, as people have incentives to over- and understate their experience. Yet it’s interesting that search leads the way - which either describes Google’s AI mode having an impact, or justifies their decision to deploy it. (Wikipedia reports an 8% drop in inbound web traffic0

The rapid and massive data center buildout increasingly elicits local outrage. I’ve seen this in my community (Manassas, Virginia), where data centers have appeared with astonishing pace and impact. Many folks are furious. Our state legislator told me she hears more complaints about data centers than any other topic.

You can see examples of this attitude across the country. Here’s TIME magazine’s cover story on it:

Or, as seen in this cartoon:

Note that the data center opposition is actually organized. People are holding meetings, staging protests, lobbying government officials. This is a demotic, grass roots movement as of this writing, but a few political leaders have been issuing statements, like Bernie Sanders:

You can find other signs of anti-AI activity. Some graduating students have booed commencement speakers for referring to AI in positive or neutral ways. The huge Comic-Con recently changed its policies on AI-generated art, from allowing to forbidding it. A University of Fairbanks student took a more direct approach to AI-generated art by eating a classmates’ exhibit about a relationship with a bot. An independent game developer removed his game from Steam because he had a change of heart of using AI to create its assets.

Local opposition has successfully blocked some data centers, according to a Wall Street Journal report:

Local opposition blocked or delayed at least 48 projects valued at some $156 billion last year, according to Data Center Watch, an organization tracking the trend. A record of 20 were canceled in the first quarter of the year because of local backlash, figures from climate-media outlet and data provider Heatmap show. Dozens more are currently facing similar obstacles on top of obstructions because of permitting snafus and equipment shortages.

Since this is America, there have also been a handful of violent acts. Perhaps the most visible was a man firebombing OpenAI CEO Sam Altman’s house, then attempting the same at the company’s main office. Lila Schroff reports that “someone shot 13 rounds at an Indianapolis councilman’s house and left a note under his doormat: ‘NO DATA CENTERS,’ it read.”

What might we take away from these stories, these developments and examples?

The companionbot side of AI looks durable so far. It doesn’t seem to be a majority use so far, but a good number of folks treat AI in that anthropomorphic way. Note the rise of physical robots in this sense. I expect to see more as the robot industry burgeons, including a range of AI-backed robots for companionship, retail service, security, sex, therapy, war, etc.

Producing content with AI continues to develop. Art, games, movies and more demonstrate technology-mediated human creativity. Schrader’s vision suggests an acceleration of AI use in art and stories. I’m intrigued by the presence of surrealism and fantasy in AI-created works, but am not sure how far that will go in the near future. Opposition to the use of that tech in art and story is increasingly vigorous, now entailing a careful scrutiny for AI in projects which don’t proclaim it. Once again, I wonder if we’ll see more public declarations of how much AI is in a given artifact. Will a Good Housekeeping-style third party emerge, registering items as 100% human, or 1/2 AI? How far will checking for AI in someone else’s work go?

Thinking about the pro- and anti-AI split, I expect to see it continue and perhaps extend, at least in the United States. As I’ve said before, I’m expecting more Democrats to oppose AI, while Republicans maintain their support for it, and AI becomes another partisan, culture wars front or signifier. Perhaps we’ll see the grass roots nature of opposition become a kind of AI populism. Or maybe refusing AI will appear as cognitive sovereignty, “The ability to maintain independent thought and agency in environments saturated with AI‑generated outputs.”

I also wonder about some kind of synthesis between the two AI attitudes. The Quinnipiac study offered one glimpse of this through a medical hypothetical:

Americans were asked if it were proven that an AI tool is more accurate than a human in reading medical scans, would they prefer to rely solely on information provided by AI, solely on information provided by a human, or a combination of both AI and a human.

An overwhelming majority (81 percent) say they would prefer to rely on a combination of both AI and a human, 14 percent say they would prefer to rely solely on information provided by a human, and 3 percent say they would prefer to rely solely on information provided by AI. [emphases added]

Now, AI is not exactly at that level yet, but the tech seems to be closing in. I’m reminded of “centaur chess,” or what Kasparov named “advanced chess,” whereby two humans play against each other while each uses AI. A synthesis including the best of both worlds?

On a personal note, I’m fascinating by the double role American culture plays in AI, being the world leader in developing the tech and opposing it, as per Lila Shroff’s aphorism. In my work I try to balance a focus on America for multiple reasons (a vital country on the world stage, a leading and rich post-secondary sector) with a global perspective. It’s a tricky balance to strike, given the intense parochialism within the United States, accentuated by some political tendencies. I hope to redouble my international work.

And thus ends this run of the scanner. Over to to you now, dear reader. Are you seeing any of these cultural responses to generative AI in your world? Are there other examples we should know of?

(credits: first image was Deborah Lupton’s ”Pop AI Collage”, Licenced by CC-BY 4.0; thanks to Ruben Puentedura, Jesse Walker)

I also wanted to try a short post, after a series of very long ones here.

The analogy is very simple. Sometimes, AI is like a robot which comes with you to the gym. It’s an exercise companion.

When you hit the facility, you have two choices. On the one (perhaps metallic) hand, you could ask the bot do all the exercises for you. It can lift weights, run around a track, do calisthenics, jump rope, do stretches, while you put your feet up and nap or scroll Tiktok.

On the other (perhaps plastic) hand, you could do the exercises yourself while prompting the gymbot have it help you. It can spot you on the hardest presses, correct your posture and form, or just cheer you on. It can monitor your sets and build a spreadsheet of them. It can give you dietary advice for better health, expanding your protein intake, etc.

So like a gymbot, which can do these physical tasks, an AI can, on demand, do all of your intellectual work of learning, writing, teaching, research, and so on. Or you can instruct it to assist you in doing that work better, like a mental personal trainer. It’s a personal instance of deciding between outsourcing/offboarding and augmentation/integration.

It’s a simple idea, and maybe it’s of use. Please use it as you like. You can build on it, tweaking and adjusting to your needs.

Now, given how the AI discourse runs these days, I feel I should add some caveats. This is not an exhaustive model. It doesn’t directly include the many, many macro factors which we explore on this Substack, like geopolitics, culture, law, macroeconomics, or academic institutional development. It is not an acceptance of AI inevitability nor an attack on the industry. There’s no mention of microeconomics, like who owns the bot, its usage terms, and so on. Instead, the metaphor is basically an adumbrated hunch.

One more point: it’s a metaphor or analogy for now. Yet the robotics industry is advancing quickly and extensively. (I have a post on this coming up.) Soon this might not be a hypothetical story but a description of some users’ real world, physical options.

What do you think of it? Does the metaphor change with the user’s age and education level? How does the analogy play out over time, iteratively? And do you like such shorter posts here?

(thanks to friends and supporters on Patreon for a fine reflection on this, plus the audiences I’ve tried it on)

The occupations that are seeing the greatest productivity boosts [from AI] are also the occupations that are seeing the greatest job risk.

-Bhaskar Chakravorti, dean of global business

The Fletcher School, Tufts University (source)

Greetings from late April. Spring semester is coming to its end, and I’m delighted by my students’ final projects as they develop. Meteorologically spring here in the mid-Atlantic means temperatures rising unevenly and every single plant blooming at once, a gift to the eyes and allergies.

Today I’d like to turn our attention to the economics around AI, which continue to develop ambitiously, chaotically, quickly, and sometimes at an enormous scale. Let’s fire up the scanner and check in on this critical domain of the AI story.

(If you’re new to this newsletter, welcome! This is one of my scan reports, which are examples of what futurists call horizon scanning, research into the present which looks for signals of potential futures. We can use those signals to develop trend analysis, which we can then use to create glimpses of what might come next. On this Substack I scan various domains where I see AI having an impact. I focus on technology, of course, but also scan culture, government and politics, economics, and education, this newsletter’s ultimate focus. All of these interconnect with and influence each other.

(It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for examples.)

Now, on with the scan. First we’ll touch on some actual uses of AI in the economy, expand on emerging business models and practices, then dive into the bubble question, followed by a look at workforce issues. We’ll conclude with some futures reflections. Then I’d really like to hear your thoughts.

1. Actual uses of AI in the economy

Let’s look at some instances of AI at work. There are so many examples - literally tens of millions, if not more - so consider a couple of media examples to start.

Al Jazeera partnered with Google to build its own AI services:

the initiative will integrate AI systems to help Al Jazeera journalists process complex data, produce immersive content, gain access to analytical context and automate internal workflows, among other things.

The Cleveland Plain Dealer started using AI to write articles and also to generate videos, moves which elicited a lot of criticism.

2. Emerging business models and practices

The subscription model seems to be working well for Anthropic. A Bloomberg report found the Claude-maker’s numbers increasing, especially after the fight with the Pentagon, and projected in March the company would earn $20 billion annually as a result, what with 1,000 businesses paying at least $1 million each for Claude. In April Anthropic estimated it would take in $30 billion.

OpenAI tested out a different business model, piloting ads for some ChatGPT users, a decision which summoned some pushback. An official statement in March reported that this test had gone well and would expand into a program. On a related note, it seems that users would purchase items found through ChatGPT via a third party app. In contrast, Anthropic criticized the practice in a post then mocked it with a video:

OpenAI is following an additional business practice of winning government contracts, specifically with the American military after secretary Hegseth declared Anthropic a supply chain threat.

OpenAI also launched an apps directory, but doesn’t seem to charge for it yet.

Partnerships between AI firms and others continue. Meta signed agreements with news publishers to provide real-time news. A geopolitical partnership appeared as the Trump administration arranged an alliance with eleven nations to supply materials for AI, including rare Earths, apart from China.

On the flip side of positive partnerships, Anthropic claimed foreign companies were illegally using Claude to train less powerful AIs. And Anthropic also announced it was offering a legal AI service, upon which legal publishers’ stock market values fell.

3. The bubble question

As of this writing the AI industry, if it is a bubble, has not yet popped. There are signs and stories suggesting it might, like Peter Thiel investing in a company which adds AI to cow collars and fencing; Halter (its real name) might crack a $2 billion valuation and seems faintly ridiculous. More broadly, the major AI companies continue to devour vast amounts of money. Indeed, OpenAI thinks it will only turn a profit in 2030, while Anthropic hopes to be in the black the year before that. Three to four years is a long time to keep plowing tens and hundreds of billions of dollars into the cause; the Financial Times estimates more than $650 billion.

So there are arguments for LLMs being a bubble about to pop. This morning a report held that OpenAI missed revenue targets, which causes related stock prices to drop. In February two Goldman Sachs economists took aim at the recent idea that AI is a big driver of American economic growth. Joseph Briggs and Jan Hatzius found that most of the domestic AI spending was actually on imported goods, namely chips produced elsewhere. Hatzius: “A lot of the AI investment that we’re seeing in the U.S. adds to Taiwanese GDP, and it adds to Korean GDP but not really that much to U.S. GDP.” Moreover, there aren’t widely accepted models to account for the supposed AI macroeconomic boost. (If this is true, the US economy is not in good shape and a bubble popping might be worse than expected.)

We can find other evidence for economic trouble in the sector. In late 2025 OpenAI and Nvidia announced a $100 billion deal, one which seems to have fallen off the table. One report found instead a mere $25 billion offer in the works, which represents quite a climb down. Also in late 2025 the Financial Times found insurance companies increasingly nervous about covering AI. American retail behemoth Wal-mart determined that ChatGPT wasn’t helping sales and would run their own AI. The Disney-OpenAI agreement failed when the latter shut down video authoring tool Sora. And Adobe is spending a lot more on… advertising, to defend its AI work.

Yet investment continues to pour in, which either confirms the bubble model (it’s the frenzied inflation before crash) or doesn’t (AI infrastructure is just growing to meet current use and future demand). Paul Kedrosky plotted IT investment historically since 1990 and found the present moment to be at a peak rivaling the dot-com:

OpenAI continues fundraising at an international, massive scale. Meta purchased Chinese AI firm Manus for at least $2 billion (or tried to! As I finished revising this issue the Chinese government blocked the purchase). A new AI investment group, Project Prometheus, launched with more than $6 billion in backing, with Jeff Bezos in a leading role. xAI raised $20 billion.

There are many efforts to make AI training and operations more efficient and hence to control costs. Here’s one Google paper sketching out four ways to do this: high bandwidth flash, processing-near-memory, 3d memory logic stacking, and low latency interconnect.

4. AI and the world of work

The impact of AI on the labor market remains fluid and ambiguous. Reports and research yield different, tentative results. I remain committed to my three scenarios for where AI might take the world of work:

1. Creating more jobs than it replaces, as with the three industrial revolutions;

2. Cutting more jobs than it creates, yielding unemployment and underemployment;

3. Not changing the total mix of jobs so much as infusing the technology into current positions, transforming them to various degrees, much like the internet did.

I can find research pointing in each of these directions, but recommend readers examine claims on their own.

For my #2 scenario, one bit of evidence is a pair of nonprofits starting to award basic minimum income payments to people who lost their jobs to AI.

Applicants are being selected on a rolling basis. They’re assessed for eligibility by the AI Commons Project team, which oversees the dividend. If selected, workers will receive the $1,000 per month stipend through 2026 and into 2027. The project designers hope to gather data about the impact the funds have on workers as they push to expand the program.

Another story which might fall under my #2 is Stanford University professor Erik Brynjolfsson’s description of what he sees as an initial decoupling between productivity and labor, at least in the American economy. In late 2025 the number of jobs ticked down, while worker productivity rose. Brynjolfsson interprets this as businesses just starting to make use of AI in workplace functions.

For my third scenario, a pair of Harvard researchers found some evidence of worker productivity rising with AI integration, at least in coding: moderate improvements, not transformations.

A different analysis found some businesses starting to track staff use of AI as part of performance reviews. One reason: “[c]ompanies eager to show a return on their massive AI investments are experimenting with new ways to secure worker buy-in.” Put another way, those businesses want to maximize worker productivity, as ever. An example of this is Accenture telling senior staff to use AI - to be seen using AI - or risk not getting promotions. Another is Amazon pushing staff to use AI even when results aren’t impressive, as per a Guardian account.

An MIT/Oak Ridge National Laboratory effort called Project Iceberg combined big data and LLMs to simulate AI’s impact on the labor market. A first result set estimated that 11.7% of job skills are susceptible to AI-driven automation, especially those focused on “cognitive and administrative tasks” in fields like “finance, healthcare, and professional services.”

A LinkedIn study reported by the Wall Street Journal described some emerging AI-related jobs. Setting aside date center construction jobs, “640,000 jobs [appeared] between 2023 and 2025 in the U.S… including new white-collar positions such as head of AI and AI engineer.” As a proportion of the workforce, the study found that “[i]n 2023, AI-related roles made up only 1.6% of all job postings… Two years later that figure had more than doubled to 3.4%.” Some of the specific kinds of jobs:

One rising job is head of AI. In the three years from 2023 through 2025, companies sought to fill 225,000 such jobs, up 49% from the prior four years, according to LinkedIn job-posting data. Another fast-growing job—though often part-time—is data annotator. People in this role help train AI models by reviewing and labeling data, including audio, text and images, so that the models can learn to recognize patterns and improve their output. From 2023 to 2025, companies added 312,000 such roles, according to LinkedIn. Hiring for such jobs remains highly concentrated. As of late 2025, only 6% of companies had job ads mentioning AI, according to an analysis by jobs site Indeed, up from 2% in 2018. A fraction of all companies—just 1%—accounted for 90% of all such posts, including many large tech companies.

There are others:

…some lower-end gig work: hiring workers to help improve AI models through tasks like labeling images and tagging text. As AI models advance, more companies are offering generous paydays for experienced workers with specialized skills and knowledge that can't be readily scraped from the internet.

Where are layoffs and hiring slowdowns occurring?

Harvard researchers looked at job announcement data and argued that employers are much more likely to slow hiring younger employees for new positions than older ones in senior roles. In the age of AI they observe “a pronounced decline in junior employment, while senior employment remains unchanged.” As economist Noah Smith put it, “there is micro evidence that companies that use generative AI more tend to hang onto their old workers and reduce their hiring of young workers.”

Layoffs appeared in the tech sector. Microsoft offered voluntary severance to 7% of its workers as the company shifts more resources to AI. Meta aims to cut 10% of its staff for a similar reason. The CEO of Block took a strong action, terminating 40% of that company’s workforce, avowedly because of rising AI use. Atlassian laid off 10% of its staff, mostly software designers and researchers.

Reflections

So what do we make of all of these economic stories and datapoints? At this point, things seem to still be fuzzy and open. As a futurist I’m frustrated by the variety of signals and contractions between them.

Yet as a person in the world I think we’re still in a moment of imaginative and collective action possibilities. The AI enterprise is not cast in stone, but capable of changing.

Some of these preceding stories represent continuations of trends we’ve been following. The diversity and development of business models fits well with the logic of technological investment as well as inter-enterprise competition. The rise of government contracting deserves notice, especially as it collides with politics and international relations. That AI shows some productivity growth as well as labor costs is a theme we saw in 2025. Threats to or transformations of technology jobs, ditto.

Some evidence points to developments which seem fairly likely. I would not be surprised to see more partnerships between AI companies and other businesses and government agencies. Such connections are increasingly political; that dimension looks likely to expand.

In contrast, the bubble question is wide open. The signals gathered here could point in either direction. We need to add information from other domains, such as culture and politics, to get a better sense of the likelihood of a pop.

Last fall I wrote this:

Overall it feels like 2025 is something of an inflection point for AI and the economy, a moment of furious work in the midst of deep uncertainty, or risk. We’re building out a huge ecosystem in great uncertainty with vast sums of money swarming through multiple nations and it’s not clear if we’re headed for a railroad future of a rich, globally used infrastructure or a technological misfire, a 8-track for the 21st century.

Companies are changing work and laying off staff, but tentatively, to a degree indicating it’s still early days.

I still hold those views in 2026.

I’m impressed by Noah Smith’s meditation on AI and the labor market, in part because he can’t come up with a clear explanation for what’s going on and is willing to say so. He finds several ways AI could have a small impact, but is ultimately unsure. He reminds us to take into account two other factors: “Trump’s tariffs [and] the immigration crackdown.” Yet also there is “the mad scramble for AI investment [and] technological uncertainty.” Economist Tej Parikh makes a similar case, noting non-AI factors and concluding that “the impact of AI on today’s jobs market isn’t as severe as the headlines make out” and “it is still early days.”

Still early days. That’s all for now. The scanner will soon turn its attention to other domains. And there are other topics to post about.

I’d love to hear your observations in the comments box.

(thanks to David Gorvitz and Jesse Walker for fine links)

In this post we’ll begin by looking at evidence from the war so far, tracing out first order effects. Next we will speculate on second order possibilities.

I’ll add my usual caveats about the situation being very fluid as well as hard to discern because of fog of war, censorship, secrecy, etc. I’d love to hear from readers who have more information. (As I write this there’s a kind of cease fire at work now, although many players are testing and violating it.)

First order effects

Let’s start with the immediate or evident cases. AI is playing some role in the war. The nearly one decade old AI Project Maven, having already assisted in several prior United States military operations, seems to be using more AI for identifying and ranking targets. The American military also used Claude to some degree, according to the Washington Post. (Don’t forget the struggle between Anthropic and the federal government we noted last newsletter.).

Jon R. Lindsay, associate professor of International Relations, Georgia Institute of Technology, argues that most of the AI use so far has been for decision support: “Modern combat organizations rely on countless digital applications for intelligence analysis, campaign planning, battle management, communications, logistics, administration and cybersecurity.” Lindsay adds that AI-backed services can reduce workload and speed up decision times.

The reverse may also be true. I haven’t found evidence of Iran using AI directly, but American intelligence sources have claimed China is using AI to enhance intel they share with Iran.

A Georgetown University near-colleague, Lauren Kahn, based in the Center for Security and Emerging Technology, adds more operational uses for AI in conversation with NPR:

Everything from boardroom to battlefield, right? We're talking about logistics, things like maintenance. When do we need to update a plane? Right beforehand, using data fusion to bring sources together from different sensors and shooters from really everywhere. So it is used on - to enable autonomous systems, things like drones, in part, to make them more capable of, you know, traveling to a location by themselves. Or things, again, like where you're seeing compiling information to get information from disparate systems into the hands of the war fighter.

It’s not clear yet what direct impact AI has had along these lines. Some blame the notorious strike on an Iranian girls’ school on the technology, but others cite faulty mapping. In contrast, the Independent claims to have identified the first person killed by AI in this war:

An investigation by The Independent and conflict monitoring group Airwars has found that Abdul-Rahman al-Rawi, a 20-year-old student, is the first civilian killed in a series of airstrikes that were acknowledged to have been carried out with the assistance of AI.

A second and perhaps more visible use of AI has been for political expression and propaganda. Examples are abounding. Here’s one from China, mocking the American war:

(I’m fascinated by the use of animal imagery there)

Also on the pro-Iranian side is a series of Lego-theme videos, such as this one posted to Bluesky:

Propaganda X VIDEOs

Or this one on Instagram (don’t mind the bland embed, just click through):

The New Yorker has an interesting article on this, including an interview with some of the content creators.

A third first-order impact of the war on AI is kinetic threats to physical infrastructure. Iranian drones have already hit several Amazon cloud service sites in the area. In response AWS declared two to be “hard down,” i.e. unusable for a time. Iran also announced it would target a series of American technology firms, including major AI providers Google (Gemini) and Microsoft (Copilot). The physical substrate of AI, expanding so massively, is now part of the war.

Given this visible role and the likelihood that there’s more going on which we can’t see in the open yet, various observers have been suggesting we call it the first AI war (for example) (another example). I am not sure, since the Ukraine and Gaza fighting have already seen AI in use.

Second order possibilities

Given these current developments, what secondary or knock-on effects might occur? Here we must speculate, based on the information we have, and given the caveats given at the start of this post.

The fast pace of the war and early signs of successful AI use might encourage the US and/or Israel to speed up the technology’s deployment, which would expand its use and likely impact. That pace, however, and the punishing demands of war could lead to errors and poorly thought out implementations. On the other hand, comfort with AI among military establishments might lead to a push for AI-backed robots for military purposes.

The war’s emerging economic ripples can impact AI. Anxieties about wartime threats might shift enterprise budgets, if organizations respond to the war by taking AI dollars and spending them instead on defensive measures: hiring more cybersecurity, hardening exposed facilities, etc. This is really a question of perception, not necessarily of ground truth reality. Cultural responses to the war, stories, rumors, journalistic and opinion pieces may prove decisive in such budgeting changes.

At a much larger scale, we’ve already seen supply chain problems arise based on Iran’s blockade of the Straits of Hormuz, from petroleum prices spiking to threats to food production. (Here’s one very detailed and grim analysis focused on the natural gas dimension.). For artificial intelligence concerns we should note that helium plays a major role in semiconductor production With about one third of the world’s supply traveling through the Straits we should expect a lot of pressure on chip production, hence on AI infrastructure rollout. AI training and operation also use a lot of electricity, which is becoming more expensive. Combined, these pressures could slow down AI training and operations. The technology’s development would then falter - a critical problem given how much AI financing relies on rapid growth.

On the energy side we’re already seeing arguments that the war should encourage people to use more renewable sources of energy, given how physically fragile and financially fraught the petroleum system is. Perhaps such a logic would lead AI-centric data centers to power themselves with renewables. (I’ve been expecting those giant buildings to attach solar panels to their vast roofs.) At the same time the war has been good for some fossil fuel providers and supporters, such as oil companies outside of the Middle East and petrostates like Russia. We could see some AI enterprises turning to oil and coal where it is politically expedient.

Things become more complicated on the political side. First, if the war continues and its many damages spread, to the extent that state decision-makers associate AI with the conflict, they might see incentives to set up their own, sovereign AI apart from an erratic United States and away from a dangerous Middle East. Here’s an interesting case where a Swedish manufacturer is trying to sweeten a business pitch to Canada by adding a local data center build to the mix. Second, perhaps a similar line of thought might lead some leaders to engage with Chinese AI to the extent that that nation appears more reliable in the war’s context. Deepseek might look like a better deal than OpenAI in certain contexts. Third, the war’s use of AI might cause other state leaders to expand their technology regulations in an effort to corral its expanding role in the world.

In the United States, we might see the war impact anti-AI activism and attitudes. Our feelings about AI might sour if the war drags on or ends to the extent we see the technology wedded to the conflict and its leading players. Associations with Israel might tarnish the technology for people opposed to that nation’s policies. The technology’s partisan aspect, which I’ve been tracking for a while, might deepen as Democrats opposed to the war map that stance onto AI. One example of this kind of thinking comes from progressive technology critic Brian Merchant, who excoriates AI firm leaders for supporting the war.

I am curious how the military’s use of AI shapes the federal campaign against Anthropic and its new uses of Grok and ChatGPT. Will people perceive the latter as military tools? Will Anthropic emerge as the AI of choice for people opposed to the war, opposed to Trump and Israel, or against militarism? Anthropic could attain a left of center political reputation.

Also on the domestic side, we should consider the possibility that leaders or ordinary people might turn against Iranians working in the digital sector. I would have thought decades of studying the crimes against Japanese-Americans in WWII would have inoculated us against these attitudes, but some recent Republican hostility to Indian tech workers suggests otherwise.

Let’s look a little further out, even to third degree impacts. Might some combination of these effects make popping an AI bubble more likely? If so, we might view the war as the industry’s tipping point. Moreover, an AI collapse or even just a slowdown in the sector’s growth might squeeze an American economy which already depends so much on AI for rising GDP. The war might tip over AI which then tips the US into recession.

And here’s a wild card: China could seize the moment of US distraction and munitions depletion to pressure Taiwan, even unto a blockage. Taiwan is, of course, the global center of chip manufacturing.

As I revised this post, the cease fire went into effect but not well. Possibly it will fall apart by the time you read this, but given the temporary cessation of firing, the focus seems to be now on the emerging Straits of Hormuz settlement. I’m very curious about the role AI might play in that. Reflecting on the points above, perhaps:

• The US and Iran use AI to generate propaganda for their Straits views

• The US and Israel use AI to analyze naval traffic

• Iranian kinetic options to press opponents on the Straits now include striking regional data centers as well as physical properties belonging to AI giants

• A continued or partial closure exerts economic pressure on AI development and the broader economy

Overall, much depends, I think, on if people take the “first AI war” concept seriously. Do we view the technology as having played a key role in the Iran conflict, or not consider that element highly? Has the war changed what AI means for us, or do we deem AI to be just one wartime tool out of many, alongside missiles and sanctions? Or will most people, who often don’t see international affairs as vital to their lives, not connect the dots and not change their thinking about this technology?

Much also depends, of course, on how the conflict plays out. The longer and more chaotic things go, the more instability and threats AI faces.

A final note: I developed this post in conversation with several chatbots. (As per my usual practice, all writing here is my own.) It was interesting to see how each responded. Gemini gave good current events answers, liking different points to journalism. Claude made similar points, but fewer of them. Grok offered more sources than the others, 45 of them.

Copilot did an interesting, Axios-style formatting of issuing bullet points followed by “why it matters” for each:

Copilot also ended with a helpful prompt: “A question for you Which angle are you most interested in exploring next — chip shortages, military AI, misinformation, or global data‑center strategy?”

Over to you, dear readers. What are you seeing of AI’s role in this war, and the war’s impacts on AI?

We’ll start with international uses and geopolitical aspects AI, then shift to national and subnational cases with a special focus on the United States. We’ll add notes on copyright developments and conclude with some reflections. I admit that these categories are artificial and you’ll probably see some stories and points blur between them.

If you’re new to this newsletter, welcome! This is one of my scanner reports, which are examples of what futurists call horizon scanning, research into the present and recent past to look for signals of potential futures. On this Substack I scan various domains where I see AI having an impact, documenting each instance. I focus on technology, of course (for example), but also scan government and politics, economics, and education, this newsletter’s ultimate focus.

It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for examples.

1: AI, geopolitics, and strategy

Global competition over AI seems to be persisting and perhaps becoming a popular cause. A RAND survey found a majority of Americans thought it important that their country compete with China on AI. Some details: “Men rated the importance of U.S. dominance more highly than women did. Those with a high school diploma or less rated the importance lower than other respondents, and ratings also tended to increase with respondent age.” In contrast the Club of Rome issued an open letter calling not for international competition over AI but for global collaboration on it in the form of “a special session of the UN General Assembly specifically on Artificial General Intelligence.”

That competition over AI between the United States and China continued. The American National Institute of Standards and Technology (NIST) published a report warning that the Chinese Deepseek AI “may pose a risk to application developers, to consumers, and to U.S. national security.” At a Cybersecurity and Infrastructure Protection Subcommittee of the House Committee on Homeland Security hearing speakers warned of competitive threats from Chinese AI-backed robots. NVidia’s CEO Jensen Huang repeatedly referred to the two-nation competition as very close. For example, “China is nanoseconds behind America in AI.”

Related to this contest is the American decision to send AI specialists abroad as part of the long-running Peace Corps. The Tech Corps “will recruit STEM graduates or those with professional experience in the artificial intelligence sector and send them to participating host countries.” I do note that the Corps’ website is down, and can only point you to the Internet Archive’s most recent archived version.

Meanwhile, AI for politicking continues to grow. Some politicians embrace the tech for communications. President Trump's administration of using AI to generate a steady stream of images, memes, and videos. In contrast, some prominent American Republicans oppose AI in general, presumably refusing to use it for strategic purposes.

Candidates, operatives, and supporters are using the technology for elections. A candidate running for a state office in America staged a debate with an AI version of his opponent. Democrat Ghazala Hashmi declined to debate Republican John Reid, to Reid had an AI trained on Hashmi’s public record answer for her:

Opponents to Irish presidential candidate Catherine Connolly created and posted a deepfake video of her, where her digital representation announced she was withdrawing from the race. This didn’t weaken Connolly’s campaign, as she won by a large margin. On the European continent two conservative Dutch parliamentarians used AI to create deepfake images of their opponent; their party’s leader later apologized.

Governments have also been using AI for security and military purposes. Before the Iran war started the head of Israel’s National Cyber Directorate (INCD) forecast a future conflict wherein AIs organized massive cyberattacks on enemies. As I write this it seems likely that the United States and Israel (at least) have been using AI in the Iran war, but it’s not clear exactly how. AI-assisted targeting seems likely. That INCD leader, Karadi, described how “Iran used cyber weapons to try to target every citizen in Israel multiple times during the 12-day June war;” potentially Iran is doing something similar now. Another world power, China, recently announced a system whereby a soldier would use AI to help control a personal swarm of up to 200 drones:

According to the South China Morning Post, the drones are launched from the Swarm I land vehicle, A.K.A the High Mobility Swarm Weapon System, which can simultaneously launch 48 fixed-wing drones that work together. Each unit can then autonomously communicate with each other, allowing the entire swarm to fly in precise formation and divide tasks among themselves, like conducting multi-target reconnaissance and strike operations.

One of the leading founders of contemporary AI, Yoshua Bengio, recommended that governments worldwide compel AI companies to purchase “liability insurance to cover existential risks.” Otherwise they would lack incentives to proceed safely.

2: AI at the national, provincial, or state level

Now let’s focus on individual nations and subnational polities for their approaches to AI in terms of law, governance, policy.

China The new Five Year Plan includes a strong emphasis on AI. That includes:

large scale commercial application of AI, infrastructure projects on hyperscale intelligent computing clusters, and improved AI governance. Targeted future industries include quantum technology, embodied AI, brain–computer interfaces and 6G.

The Chinese government is taking various steps to maintain political control over AI. According to a Wall Street Journal account the state has removed some content and works closely with AI companies. Additionally, “[a]ll AI-generated texts, videos and images must be explicitly labeled and traceable, making it easier to track and punish anyone spreading undesirable content.” Beijing also promulgated new rules for controlling AI content for children.

Vietnam The first southeast Asian nation to regulate AI issued rules which took effect March 1. That state now mandates that AI businesses “clearly label AI-generated content such as deepfakes that cannot readily be differentiated from reality” as well as telling users “when they are interacting with an artificial rather than human agent.”

South Korea Seoul issued the AI Basic Act which assigns companies the duty to give the government “a mandatory prior notice requirement for ‘high-impact AI’ affecting fundamental human rights.” South Korea is positioning this law in reaction to European Union delays in implementing their AI Act.

Japan A new political party, Team Mirai (チームみらい;; “Team Future”), won national legislature seats and is advocating AI to support governmental operations. One project uses AI to make policy development more accessible. One key strategy: Mirai argues that “savings achieved through A.I. could be used to lower contributions to pension and health care plans for working-class families.”

United Arab Emirates The UAE announced it fended off an AI-driven cyberattack.

European Union The European Parliament worked on revising its AI Act. Proposals advanced technical policy implementation dates to late 2027 and 2028, while adding a ban for nudifier applications.

Canada The University of Toronto’s Munk School published a white paper calling on Canada to take strategic steps to secure its AI sovereignty: “the question is how to structure dependencies to preserve choice, reduce foreign leverage, and ensure that Canadian data and infrastructure remain governed by Canadian laws and values.”

The ombudsperson for British Columbia warned that some residents could lose access to government services if the province relied on AI to provide them.

3: The United States on AI

The CFO of America-based OpenAI, Sarah Friar, floated the idea of governmental backup for their venture, then walked back the concept. CEO Sam Altman stated he didn’t want federal support on X:

President Trump announced the Genesis Mission, an effort whereby the Department of Energy leads scientific innovations anchored on federal resources and AI. DoE will build out supporting infrastructure while collaborating with businesses and laboratories.

Trump also published a “National AI Legislative Framework” which is a template for a potential Congressional bill. It includes points on making AI companies pay for data center electrical power, some kind of compromise between creative work owners and AI training needs, and blocking AI from censoring political speech. There are also calls for Congress to reduce barriers to innovation (not clear what that means) and to expand workforce training. One key detail is that such legislation would supersede state laws, which is irking state utility regulators.

The United States military has been struggling with the AI firm Anthropic. The Department of DefenseWar deemed the technology a national security risk and declared Anthropic a supply chain risk, which would cripple the company’s business. The company then sued the administration. As Matteo Wong observes, at stake are fundamental questions about government-AI relations. Also in play are public attitude towards AI. One sign of this is that Claude downloads from Apple’s app store surged to the leading position, passing ChatGPT, after the break became public, so much so that the service went down for a time.

Another branch of the federal government, the Department of Homeland Security, ordered OpenAI to produce some user prompts for security purposes.

America might decide on an unusual way to supply electrical power for AI-supporting data centers. A Texas company, HGP Intelligent Energy, has pitched the government on using nuclear reactors from decommissioned submarines and aircraft carriers as power sources.

Another American enterprise, Americans for Prosperity, is urging the Trump administration to expand its support for AI development. In particular the group wants data centers powered by fossil fuels, an unsurprising argument given its Koch brothers funding. In contrast, there is rising opposition to AI aimed at the data center build out which Democrats might be able to channel. On the left Bernie Sanders has been calling for stronger AI regulation, even talking with Claude about it:

Democratic senator Amy Klobuchar also called for strong federal AI regulations.

There are also some Republican anti-AI folks, as noted above, like senator Josh Hawley and outgoing Florida governor Ron DeSantis. (Interesting to see Newsmax offer a fairly uncritical account.)

At the Congressional level AI companies are raising hundreds of millions of dollars for lobbying against regulations they dislike. I was struck by this analysis and strategy:

OpenAI co-founder, Greg Brockman… and his wife, Anna Brockman…said they have observed "two extremist camps" in the AI regulatory debate and that "neither feels right." "We believe in AI centrism, which supports thoughtful regulation toward the goal of unlocking AI's potential to improve quality of life for every person (and every animal)," they said. The couple said they want to see federal AI regulation that promotes innovation, defends systems from misuse and protects "the privacy of conversations with AI." Beyond that, they said, "Most developers and open-source models, and almost all deployments of today's technology, should have minimal additional regulatory burden."

Meanwhile, a bunch of Senators and Representatives fired off AI legislation in late 2025. Three focused on education, including protecting children from AI abuse.

(They continue the modern Congressional habit of naming laws with cute acronyms: Learning Innovation and Empowerment (LIFE), Recommending Artificial Intelligence and Standards in Education (RAISE), Guidelines for User Age-Verification and Responsible Dialogue (GUARD))

More action occurs at the state level. California is investigating Grok.ai for generating sexualized images of children. Amazon sued an Oregon utility for not providing the electrical power they asked for. The Rockefeller Foundation launched an effort to prepare state and local governmental capacity for dealing with AI.

4: Copyright

Struggles over AI and copyright continued.

A group of writers proposed a class action lawsuit against Adobe for training its SlimLM tool on their work. The New York Times sued Perplexity for copyright violations:

The suit accuses Perplexity of violating The Times’s copyrights in several ways, most notably when the start-up’s search engine retrieves information from a website or database and uses that information to generate a piece of text and to respond to queries from internet users. That would not be a fair use of that material, the suit claimed, because Perplexity grabbed large chunks of the publication’s content — in some cases, entire articles — and provided information that directly competed with what The Times offered its readers.

An artist who uses AI to create images sued the federal Copyright Office to win copyright protection for his works. The office has previously refused to do so.

Meta and a group of news companies agreed to allow Meta’s AI to use some of their content.

The British government previously proposed a change to copyright law allowing an exception for AI training, but is now considering scrapping it. That country’s Competition and Markets Authority argued that Google should let content creators opt out of having their materials scraped for AI training purposes. In response, Google offered more user controls.

Also in the intellectual property world, a British software firm, The Foundry Visionmongers, sued Adobe for trademark infringe in its Firefly Foundry application.

Some reflections from a futurist angle

What can we learn from these political developments, looking ahead?

The Iran war is starting to impact and demonstrate AI capabilities. It’s still unclear at the moment (here’s a post on how I track the war) , but it seems that at least some combatants are using the technology in various ways. The Straits of Hormuz closure may hit chip making as it cuts the flow of oil and also helium to Asian makers. As the war continues we may see intersections between its politics and those around AI: some supporting Israel and the US and AI, while others oppose all three, for example. To the extent the conflict involves cyberwar we should anticipate an AI role on that front.

AI sovereignty may develop as an organizing principle for various polities. It’s an institutionally conservative term, one which frames AI within classic political goals. Again, geopolitics may drive this on to the extent that nations want to evade the American AI tech stack, or the Chinese.

There is also continuity on AI and politics. I find myself repeating some points from previous newsletters. Politicians and governments are increasingly using AI as tools of statecraft and electioneering, although it’s not clear deepfakes have had a decisive impact on any election so far. The US-China AI competition continues; I hope we get some good analysis of the new Five Year Plan. Depending on how the Iran war turns out, we might see US-China relations impacted, which should impact AI.

In American politics we’re seeing traditional political structures in play over AI. We have the federalism struggle over which level of government should regulate the technology. The Trump administration continues to embrace AI, from the Tech Corps to Genesis. And a red versus blue framing seems to be settling in, although with some noteworthy Republican defections. Indeed, as I’ve been noting for a while, the culture wars are embracing AI. That partially explains the gender divide RAND found.

The Pentagon’s break with Anthropic cast Claude as a polarized actor, especially as many users raced to download it. Perhaps conservatives will make a point of shunning Claude in favor of ChatGPT and others. Maybe Democrats will in turn pick up Claude, or at least the ones who don’t reject generative AI overall. There’s a broader question at stake concerning government’s role in regulating and using this new technology, and the autonomy of AI enterprises.

Data centers are increasingly controversial and political and they are settling into preexisting politics as well with utility politics and federalism. I’ve seen local signs of that in Virginia, where there’s a massive data center build out. Our local state legislator has said that data center complaints are by far the issue they hear the most about.

I am curious to see if the Brockmans’ “AI centrism” term has legs. Given American political polarization, the center is a tricky place to define, much less hold.

Copyright dynamics continue to unfold. There are openings for regulation in many nations plus opportunities for lawsuits.

Overall, the AI and politics/policy world is one of continued expansion, development, divides, and churn.

…and I’ll stop there. This post is already longer than I planned. Now over to you all. What are you seeing on these themes?

(Thanks to Owain Alexander, Hernan Carvallo, Tom Haymes, Donna Kidwell, Ruben Puentedura, and Ed Webb for links and feedback)

There’s a lot going on in the AI space now. There are developments and signals across the board, from the labor market to policy, technology services to pop culture. I’m sifting through a ton of materials and ideas, trying to scry where it’s all headed. I hope to share a series of posts on what seems to be like a decisive year for AI.

However, I want to hold back on commenting about the current excitement over AI agents and takeoff for a few days. I want to make sure we consider more data than a handful of inspired posts. Again, there’s a lot going on across the field, and since so much intersects we should get current. So for today’s part of this tour d’horizon we’ll catch up on AI technological developments over the past couple of months. There’s a great deal of activity, between many iterative releases and some discussion of the possibility of major changes.

Let’s check in on AI providers and projects.

(If you’re new to this newsletter, welcome! This is one of my scan reports, which are examples of what futurists call horizon scanning, research into the present and recent past to identify signals of potential futures. We use those signals to develop trend analysis, which we can use to create glimpses of possible futures. On this Substack I scan various domains where I see AI having an impact. I focus on technology, of course, but also scan government and politics, economics, and education, this newsletter’s ultimate focus.

It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for examples.)

OpenAI launched GPT‑5.2‑Codex in December, then upgraded it to version 5.3 in February. It’s focused on generating code.

Note that OpenAI claims the app contributed to making its own code:

GPT‑5.3‑Codex is our first model that was instrumental in creating itself. The Codex team used early versions to debug its own training, manage its own deployment, and diagnose test results and evaluations…

This is a potential step towards the idea of AI building itself, potentially leading to very rapid development.

OpenAI released Prism, a tool for scientific writing, built on LaTeX. The company also launched ChatGPT Health, a chat function focused on answering medical questions with some extra security features and some integration with personal databases. Note the careful caveat:

Health is designed to support, not replace, medical care. It is not intended for diagnosis or treatment. Instead, it helps you navigate everyday questions and understand patterns over time—not just moments of illness—so you can feel more informed and prepared for important medical conversations.

ChatGPT now has a Deep Research function, built into the chatbox, which offers more extensive web search and integration for answers. For an example, I asked it to “help me understand the geopolitics of the North pole.” It asked a good clarifying question in turn, which I addressed, then it worked for around six minutes. It opened a window on the right side of the screen, detailing the web sources it worked through, before finally posting a detailed answer in the main window:

It added this description at the end: “Research completed in 6m · 36 sources · 70 searches.” Sources were an interesting mix, from Wikipedia and Scientific American to some US military and Congressional web pages.

On the hardware side, OpenAI is allegedly planning on releasing an AI-backed smart speaker and a smart speaker, according to The Information via Futurism. In contrast, Mashable thinks OpenAI is working with Johnny Ive on a smart pen.

Google released a Deep Think function for Gemini 3, which claims to have advanced capabilities in math and science. Google cites impressive marks:

Deep Research is only available for paying Ultra users. Google also connected Gemini and Google Translate.

Anthropic posted a set of agent standards. The focus here are skills which AI can use to complete tasks across different datasets and domains. The company hopes for this to become an industry standard.

Elsewhere, Anthropic has been scanning a lot of books. The Washington Post broke the story of Project Panama, an effort to buy and scan millions of books in order to improve AI training materials and to avoid some copyright problems of using pirated ebooks.

More, Anthropic released a new version of Claude, Opus 4.6. It’s a step forward in coding ability, according to some commentators. It claims major agentic improvements, including the ability to “assemble agent teams to work on tasks together.” There are also connections to other tools, such as PowerPoint. Anthropic states this new model is now the world’s leader across many metrics and tests. They also think it has some low probability of being dangerous.

One important claim is that Anthropic used AI to make better AI. “We build Claude with Claude. Our engineers write code with Claude Code every day, and every new model first gets tested on our own work.” Anthropic’s CEO goes further:

Because AI is now writing much of the code at Anthropic, it is already substantially accelerating the rate of our progress in building the next generation of AI systems. This feedback loop is gathering steam month by month, and may be only 1–2 years away from a point where the current generation of AI autonomously builds the next. This loop has already started, and will accelerate rapidly in the coming months and years. Watching the last 5 years of progress from within Anthropic, and looking at how even the next few months of models are shaping up, I can feel the pace of progress, and the clock ticking down.

For others this is a sign of accelerative takeoff, pointing towards AI improving AI. For example,

I don’t want to alarm anyone, but this frontier model is far, far beyond human. My testing (including the ALPrompts later in this edition) showed unexpected and complete patterns of responses. Perhaps for the first time, this model feels both superhuman and complete.

Alibaba tried its hand at the agentic game, updating Qwen to handle multiple tasks. One key feature is connecting Qwen to services from other firms: “The upgrade integrates core Alibaba ecosystem services including e-commerce platform Taobao, instant commerce, payment system Alipay, travel service Fliggy and mapping platform Amap into a unified AI interface.” The latest Qwen also has expanded multimodal functionality.

Alibaba also started selling Quark smartglasses with Qwen. “Alibaba said the glasses would be deeply integrated with its apps, including Alipay and its shopping site Taobao, with wearers able to use them for tasks such as on-the-go translation and instant price recognition.”

Bytedance launched an updated version of its video authoring tool, Seedance 2.0. (No free version, only for pay now) The results have impressed many people. Not all, though - Disney threatened legal action over likenesses.

Deepseek released a new version of its signature AI, DeepSeek-V3.2-Speciale. Their release paper describes some advances:

(1)DeepSeek Sparse Attention (DSA)… an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5… (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.

Tencent launched its Hunyuan 3D creation engine, which lets users generate 3d objects from text prompts or uploaded images. I tested it out, using this science fiction image I built in Gemini:

(I had to remove the background image Gemini initially included, because Hunyuan built it into the foreground awkwardly.) Hunyuan set up an editing platform, where I could rotate and reskin the object:

I could export it in a few formats, including .glb. I haven’t tried turning the file towards 3d printing yet.

Amazon announced its own agents in December: “three specialized AI agents designed to act as virtual team members: Kiro autonomous agent for software development, AWS Security Agent for application security and AWS DevOps Agent for IT operations.”

Users turned X.ai’s latest Grok iteration to produce a range of terrible images. After waves of negative responses, Elon Musk eventually cut back its functionality.

Meanwhile, SpaceX filed an application with the Federal Communications Commission to orbit up to one million satellites housing datacenters. The document calls for using solar power unmediated by Earth’s atmosphere for electrical power.

Startup Logical Intelligence released its first AI, Kona 1.0. I’m not sure I understand what an “energy-based” AI is, but they seem keen on playing Sudoku. They have high claims for correct output. A Wired article describes Kona as learning systems rather that doing token prediction, and also using far fewer resources to train.

Startup Factify proclaims its intention to reinvent the pdf. Apparently the idea is to add and AI plugin to each document, I think. They raised $73 million to get going.

Moltbook appeared and won a huge amount of publicity. The idea was to be a social network/discussion board for generative AIs, each using an API to post, all based on OpenClaw. Then someone set up a massively multiplayer game for these AIs with a space theme called, of course, Spacemolt.

I fully admit to not having explored it directly because Moltbook’s security problems are so terrible. I’m actually considering setting up a Raspberry Pi box just to access it.

Two notes about hacks. First, an as yet unknown hacker used Claude to attack Mexican government databases and webpages, managing to score a bunch of data. The attacker “wrote Spanish-language prompts for the chatbot to act as an elite hacker, finding vulnerabilities in government networks, writing computer scripts to exploit them and determining ways to automate data theft.”

Second, an SEO hack for AIs appeared from a journalist who came up with one simple trick. Write up something about your target term on the open web and AIs might suck it up, at least Gemini and ChatGPT. Thomas Germain posted that he was a champion hot dog eater and soon these AIs agreed.

Let’s conclude with some observations.

Note OpenAI and Anthropic using AI to generate AI code. Besides the challenge to human coders, there’s a recursive function which people have been expecting for decades.

Book scanning: AI training demands more and more content. Creating artificial worlds is one way. Going after well-formed, professional curated material is another.

Quality improvements keep happening, at least based on public benchmarks, studies, and my tests. That is, AI results are improving. There are advances in modality (more modes beyond text, like 3d modeling) and more extensive and structure output formatting (deep reasoning). We’re also seeing new ideas appear on the margins.

AI is moving more into hardware, step by step. I’m intrigued by the Quark headset.

Agentic computing continues to advance from multiple projects and in several nations. Put another way, agentic functions are working into other applications, from basic LLMs to hardware systems. And they seem to be plural, from Moltbook’s odd hordes to Claude’s agent teams and Amazon’s virtual teams.

All of this progress and Claude’s advances in particular have been stirring excitement as some think it shows intimations of artificial general intelligence (AGI). I’m going to give that issue more treatment coming up. Right now that question is very open, given how many parties are biased by interest, nobody agrees on definitions, and economic data is at best tentative.

We also need more information and perspective. Next up will be (in some order) AI developments in economics, politics, and society and culture. I hope those updates will give readers more context for thinking through the big questions about AI.

(thanks to George Station for links)

This is the first offering from BAC Labs, a research and development effort based in my consulting work, now focused on exploring new approaches to AI for educational purposes.

We’d like to begin today with an unusual approach to LLM use, which we dubbed narrative programming. It’s a method to rethink prompting, as well as how we think about generative AI overall.

Readers will note that the voice of this issue is now first person plural. That’s because it’s the work of two people, Bryan Alexander and Ceredwyn Alexander. Ceredwyn started this R&D process as part of her teaching and instructional development work, developing classes and materials for emergency medical technology (EMT) programs in Vermont. She wanted to make LLMs useful for students and instructors but was frustrated by their limitations and so created a new method. She is, in short, the prime mover behind narrative programming.

In a followup post we will introduce SPARKs, an alternative to GPT agents stemming from narrative programming principles.

Read on and see what you think. We’d love to hear your thoughts in the comments box below.

Most people using large language models start with a conversational prompt. They ask a question, describe a goal, maybe suggest a role to play, perhaps specify a style or format, and expect the system to understand and infer what they want. When the AI doesn’t provide the desired results. or when it hallucinates, drifts, or overreaches, some users rephrase their prompt with some change, or criticize the bot, retry the request from a different angle, and try again. This trial-and-error cycle is what we often mean by prompt engineering, but we can also think of it as a kind of belief system, one with several components:

• That the software model understands human intent.

• That output quality comes from better phrasing.

• That if you word it just right, the model will behave the way you need it to behave.

Some advanced prompt engineering guides are beginning to talk about more sophisticated approaches, such as “context engineering” or “persona vectors.” These approaches do get better results, because they exploit what language models actually do best: extend linguistic patterns.

Our approach takes this further. We formalize those emerging tactics into a discipline. Instead of trial-and-error scaffolding, narrative programming explicitly encodes the inherent structures of storytelling. Narrative roles, genre conventions, symbolic weight, and rhetorical constraints are not only decorative features of human communication. They are structural laws of coherence for the texts upon which LLMs trained. When applied systematically, they can be used to govern AI model behavior with greater precision.

In this framework, we define the following components as constituting a “narrative execution prompt”:

• Name: the simulated posture or voice the model assumes

• Genre: the logical container that determines tone, stakes, and acceptable behaviors

• Role: the general category of agent

• Job: what the agent does within a given system

• Task: as defined by the user

• Victory Condition: the internal measure of task success

When a prompt doesn’t define these elements, the LLM’s behavior eventually drifts away from the initial prompt and becomes unpredictable. This is not a problem with “alignment” nor is it “emergent behavior.” Rather, it is what happens when the AI follows a different narrative program than the user intended but did not sufficiently describe. Without specification, an LLM will fall back on any number of stories. The user may experience this as poor inference. In contrast, narrative programming locks AI into roles with hard boundaries and encodes representation logic as enforceable ratios. Defining success, enforcing resets, and maintaining compliance over time mitigates scope creep, bias defaults, and silent drift.

Consider how LLMs work within this narrative programming framing. Our model sees AI as a pattern completion engine.

To reprise the basics: when you type a word, the model doesn’t “see” it as a word like humans do. Instead, it translates it into tokens. Tokens are little chunks of text. Shorter words can be one token while larger words will be broken up. These are the smallest building blocks. Alone, tokens have little or no meaning to humans. Yet an LLM can use vast amounts of tokens (hence the first “L”) to predict a next token using statistical completion, answering the question “What is the most likely next thing?”

During training, the model learns how often that token shows up next to other tokens. These tokens form concepts. A bird in the hand is a concept composed of several tokens that is statistically likely to be followed by, Is worth two in a bush. This is how we can see how the LLM follows the patterns of language.

Example:

• “the” → frequent start token

• “the bird” → partial concept

• “the bird in the hand” → stable idiomatic structure

• “is worth two in the bush” → full proverb (highly probable continuation)
  

In other words, the model completes the phrase or sentence based on token likelihood, not human intention.

Yet we can now scale up our sense of how training works. Because the LLM trained on vast amounts of human writing, style, tone, syntax, genre, and other narrative features shape the model’s internal functions. LLMs depend on human storytelling in a fundamental way. They devour contemporary novels, classic literature (especially in the public domain), anecdotes, autobiographies, narrative histories, legal accounts of events, military after-action reports, diplomatic readouts, testimonials and all at an immense scale. To pick just one of many examples of this, Dario Amodei’s recent essay on AI observes that:

AI models are trained on vast amounts of literature that include many science-fiction stories involving AIs rebelling against humanity.

They fit predictive texts into grooves of storytelling. Generative AI is, in a real sense, a machine of story, a remixing application for the human narrative tradition.*

Therefore narrative functions of all sorts - character and character development, plot arcs, morals of stories, setting, atmosphere, metaphor and more - all act as control structures for AI development and output, as they do with much of human writing.

Traditional computers build executable complexity from simpler units:

bits → bytes → data structures → programs

Likewise, LLM’s build complexity through linguistic patterns that embed story:

tokens → concepts (words or phrases) → narrative structures (genre, tropes, arcs, etc) → programs

That does not mean the model is learning skills or becoming intelligent. It means the training set was big enough, and the patterns rich enough, that some outputs resemble reasoning when none occurred. And, as the models have gotten larger, they have the capacity to recognize more patterns, providing the illusion of learning, when what is really happening is more powerful software working with larger data sets. Those data sets include stories at scale, so AIs can tell stories and also be directed by them.

This brings us to the problem of alignment. All the high-profile failures of AI systems point to one fact: inference alone cannot guarantee predictability, safety, or representation. Alignment is not intelligence. It is instead patchwork guardrails over stochastic text generation. Training means to enlarge the data set with input that shows more patterns we want to see, hoping for a happier average, instead of simply telling the machine what we want.

What looks like logic is, in fact, pattern mimicry (hence the famous term “stochastic parrot”). Reasoning must be scaffolded into the prompt or program. Without that, the model will produce something that feels smart but is structurally hollow. Frequently, the model will produce the same idea repetitively, using different words, presenting each statement as a separate idea. Using narrative thinking in prompts changes the situation significantly. Using the narrative execution prompt components we described above can rein in the parrot and yield better results.

That’s the first installment in a series. There’s more to come, starting with building a new form of AI bot, a Semantically Programmed Agents for Reliable Knowledge, or SPARK. What do you make of this approach? We’re very open to questions and other responses!

*Credit to the late, great Cliff Lynch for making the LLM as remix point to Bryan in conversation.

In this freezing context, I’d like to return to examining what’s been happening recently with higher education and AI. Quite a lot has been going on over the past few months.

I’ve broken up what follows into several categories: AI, teaching, and research; scholarly publication and libraries; critiquing and opposing AI; general observations and research; final reflections.

(If you’re new to this newsletter, welcome! This is one of my scan reports, which are examples of what futurists call horizon scanning, research into the present and recent past to identify signals of potential futures. We use those signals to develop trend analysis, which we can use to create glimpses of possible futures. On this Substack I scan various domains where I see AI having an impact. I focus on technology, of course, but also scan government and politics, economics, and education, this newsletter’s ultimate focus.

It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for examples.)

AI, teaching, and research

Over the past few months I’ve seen many stories of academics using AI for teaching and research purposes. Purdue University (Indiana) did both, announcing it would expand its AI efforts. On the teaching side, all students will need to achieve “AI working competency” to graduate. What does this entail? The provost will lead departments along these lines:

the goal of this requirement will be to ensure that, from exposure and awareness to skill-building and problem-solving mastery, Purdue students possess job-ready skills and critical thinking competencies to:

Understand and use the latest AI tools effectively in their chosen field(s), including identifying the key capabilities, strengths and limits of AI technologies, as well as ways that AI can transform existing methods, processes and tools

Recognize and communicate clearly about AI use, decisions and limitations, including developing and defending decisions informed by AI-driven insights, as well as recognizing the presence, influence and consequences of AI in decision-making

Adapt to and work with future AI developments effectively and continually

Interestingly, Purdue doesn’t use the term “literacy.” On the research side, Purdue will support AI efforts in food systems, health, manufacturing, military, and transportation. There are also partnerships with Google, Apple, and local K-12 schools.

On a related note, Wayne State University (Michigan) is launching an Institute for AI and DAta Science (AIDAS), focused on research. It seems to be small scale, however:

Vice President for Research and Innovation Ezemenari Obasi said the institute would cost $200,000 over three years and would be funded by the Division of Research and Innovation. After that, he said, the institute was expected to find other sources of funding.

A Universität Zürich (Switzerland) team is starting an interesting project, using open source AI to produce new LLMs for one particular subject area:

A family of 4 billion (B) parameter large language models (LLMs) based on the Qwen3 architecture trained from scratch on 80B tokens of historical data up to knowledge-cutoffs ∈1913,1929,1933,1939,1946, using a curated dataset of 600B tokens of time-stamped text

At a similar if smaller scale, an American college student built a small language model trained only on 19th century British texts.

Stanford University has been very energetically working on AI. One faculty team developed a new way of training AI visual capabilities. Their method focuses not on recognizing objects so much as the functions of those objects. The authors foresee applications in robotics. A different Stanford effort is using AIs to build virtual scientists to work in virtual labs. A third deployed AI to design viruses, from the genome up.

Across the country, a group of Harvard University researchers published on an AI application they named PDGrapher, which would help generate new medications. Its code is freely available on Github. A University of Virginia professor wrote up his experiments with using AI agents to help academics conduct economics research.

Back to the learning world, one of the founders of the learning management system/virtual learning environment industry, Matt Pittinsky, wrote up his thoughts for how AI might change the LMS/VLE. Briefly, he sees AI powering a massive new feature set around personalized learning, which will become the lion’s share of the LMS. He also envisions chatbots becoming the interface through which instructors create and redesign classes in the LMS.

A Georgia State University - Perimeter College professor described an interesting use of AI, teaching students to prompt a chatbot to simulate a conversation with an ancient literary character.

MOOC provider Udemy announced new AI functionalities for its online classes. It looks like a range of tools or actions:

Instructor AI tools that will enable instructors to transform their trusted course content and underlying learning objectives into interactive microlearning activities.

Adaptive, AI-sequenced experiences, using a mix of short videos, quizzes, and other instructional and active learning content.

Instructor-validated quality, allowing learning content to remain practical and grounded in real-world expertise.

Context-aware interactions, enabling AI-enhanced responses informed by learners’ background, goals and motivations.

Expanded opportunities for instructors, welcoming both existing Udemy instructors and new creators who specialize in short-form, interactive learning experiences.

Some of those echo Pattinsky’s views, like building in AI as part of instructor course development. Adaptive experiences and context-aware interactions suggest personalization.

Scholarly publication and libraries

On the scholarly publication side, an MIT team surveyed hundreds of academic authors for their views on using their material to train AI. The results are quite nuanced and worth digging through (here’s a second part with advice). We can identify some highlights: around half of authors open to formal partnerships with AI firms; a widespread demand for attribution; much anxiety and many questions.

Along these lines, Northeastern University dean and history professor Dan Cohen described a project that combines a local Model Context Protocol (MCP) server and the Claude AI service into a plugin which aims to provide students with an accessible entrance into large scholarly databases. For example,

(Dan also offers the term “LAG, or library-augmented generation.”)

Critiquing and opposing AI

Conversely, there were many cases of academic research and institutions critiquing or opposing AI over the past few months. Two University of Sheffield (Britain) professors found that ChatGPT consistently refuses to recognize when a scholarly paper has been retracted or debunked. Michigan State University’s extension program is casting itself in opposition to AI, launching “a campaign to position MSU Extension as the antidote to AI slop.”

Some professors are turning to oral exams to mitigate the AI cheating problem, like this writeup of a New York University business professor. A San Francisco State University anthropology professor called on the California State University system to suspend its relationship with OpenAI, arguing that ChatGPT is doing serious mental health harm to students.

General observations and research

On a broader level, Washington University professor Ian Bogost determined that college and university students already have general access to AI in ways which changed their study habits, and that “plenty of professors are oblivious.”

It isn’t that they fail to understand the nature of the threat to classroom practice. But my recent interviews with colleagues have led me to believe that, on the whole, faculty simply fail to grasp the immediacy of the problem. Many seem unaware of how utterly normal AI has become for students.

A Chronicle of Higher Education survey found academics divided over the potential impact of AI on their institutions. Roughly a bit more than one half saw danger to the academy and viewed major structural action as called for, while around one third thought the danger overblown and change unneeded:

Readers may catch a couple of sentences in the report from me, where I aver that colleges and universities have little handle on the cheating problem.

Some reflections

What do these stories tell us about academic responses to AI in late 2025 and early 2026?

Last summer I posted some thoughts on a similar question and some of those takes still hold. We are still seeing a wide range of academic uses across the curriculum, at multiple scales (single student up to entire university), a lot of collaborative projects. The deep divide over AI within the academy persists, with opposition and critique taking many forms.

Adding to those: I’m glad to see the open source projects from Zurich, Harvard, et al., both as in using open source tools and sharing results in the open. I’ve been calling for this kind of work for years.

I’ve very curious about Pittinsky’s LMS/VLE vision. It represents quite a transformation for that mature technology. Which established providers will take it up fully? How many startups will aim new projects at that vision?

There’s also that overarching sense of academia falling behind the revolutionary technology. The Chronicle survey and Bogost’s essay depict colleges and universities as institutions, and academic workers as individuals, struggling to keep up and respond well to the challenge.

Are you seeing similar developments at your institutions? Are there other academia and AI stories we should be discussing?

Now, on to other and promised newsletters. More coming up!

(thanks to Bonnie Dede, Will Emerson, Karl Hakkarainen, Steven Kaye, Joe Essid)

I’ll break up the below into the categories I’ve been using in these Substacks: actions coming from outside the academy but which bear on research, teaching, and campus operations, followed by actions academics have taken. I’ll add some reflections at the end, including observations on using several AIs to help craft this post. For reasons of time and space I’m leaving out K-12.

For those new to this newsletter, I’m basing today’s issue on what I’ve been seeing in the world, what futurists call an environmental or horizon scan. I’ve created this in several ways: traveling to a bunch of campuses and academic meetings; connecting with more academics online; reading what scholars in several fields are saying; hosting a series of Future Trends Forum sessions on and around AI; holding many, many conversations with people in and around this space. I’ve also been teaching classes on the subject at Georgetown University. I’ve then shared and tested my findings through these newsletters, social media, presentations, etc., hopefully improving their quality.

Also, as with the previous newsletter, I’m trying to compress the newsletter so that it’s not too huge. Let me know what you think. And in general, I’m always looking forward to your comments.

Relevant projects from outside the academy

In addition to the vast churn of AI offerings covering a wide range of topics, more AI-backed teaching tools and functions came online in 2025. Google, OpenAI, and Anthropic each launched a study or learning mode for their services. Google in particular leaned into education, released an AI Co-scientist, an application aimed at helping people do scientific research, while launching Learn About. Google Scholar now presents an AI interface. Other AI-powered commercial offerings have appeared, like Socrait and Gradescope. Vendors who provide educational goods and services have added or expanded preexisting AI functionality to that, as we’ve seen in the LMS world (Phil Hill’s team is on point here) and Grammarly. The same is true for general-use tools we use in education, like Canva and Adobe’s Creative Cloud suite. And pre-2025 AI applications persist, like Elicit.

Beyond technology providers came other pushes for AI in higher education. For one example, the International Telecommunication Union (ITU) announced at Davos an AI skills coalition. Various governments, including American states, have expressed support for higher ed to pick up AI.

Academic actions

How have academics acted on, or reacted to AI?

Overall, it seems that colleges and universities are still scrambling to react to this technological revolution at a strategic level. Three years after ChatGPT 3.0 exploded only one quarter of campuses have institutional policies about AI, according to the new CHLOE report. Many of the policies I’ve seen have been tentative, or basically add “don’t cheat with AI” to academic integrity pages. There are a few exceptions, where institutions took up AI strategies from the top level, like the University of Michigan, Arizona State University, and The Ohio State University. Otherwise most campuses are still figuring things out. In some cases I suspect that the presidential and dean level has punted to colleges and schools, which deferred without resources to department, which ditto to individual faculty members. Given that adjuncts are the largest proportion of the professoriate, it seems that part-time temporary instructors are leading academia’s AI engagement.

How to structure institutional responses in terms not just of policies but also governance structures, enterprise adoption, support, and more is still an open question with plenty of campus variations. I’ve talked a university who had a single, unfunded committee in charge of AI, while another had a dozen groups, each budgeted to explore one aspect of the issue. In some institutions librarians are taking the lead in supporting students, faculty, and staff, while at others teaching and learning centers play that role. IT departments continue to wrestle with security, licensing, training their own staff, and overall support. Academic departments decide on their own approaches to AI.

Funding academic AI work is a challenge this year, especially in the United States (for more information on why things are challenging, see my blog or the Trump vs academia video series or new book or just look around US media). Some campuses have reached for external support. Perhaps the most dramatic is Bowdoin College which set up the Hastings Initiative for AI and Humanity. The “Hastings” in the name is from Reed, as in Netflix chair and co-founder, who donated $50 million to start it off. The California State University (CSU) system struck a deal with OpenAI to fund community access to ChatGPT. The National Science Foundation (NSF) now offers support for AI work, as does the Department of Education. At least one foundation has an AI program.

For many the largest AI issue is accelerating AI-enabled cheating and how to respond to it. There have been many examples of this - the University of Minnesota expelled a PhD student, charging him with cheating using AI on a preliminary exam. The (former) student fired back, criticizing the detection process, the disciplinary process, and also the professors involved. Elsewhere, a federal judge dismissed a researcher’s statement because AI generated some of it. Hilariously, the expert’s speciality was misinformation. France’s Sciences Po returned to entrance exams, in part to reduce AI cheating. There are technological responses with tools like ChatZero but the consensus (which I share) is that they are very unreliable at best. Meanwhile, some faculty and some schools have turned to basic, retro anti-cheating efforts, notably oral presentations, the nostalgia-drenched bluebook, and banning devices from classrooms. These don’t scale well and have other problems. In fact, we still don’t have good solutions. I keep coming back to the need to reformat our entire assessment system. (I mentioned this in a talk to the POD Network, and to my surprise the audience applauded thunderously.)

Beyond assessment problems there are plenty of pedagogical, research, and operational experiments, in addition to some curricular offerings. Emerson College is experimenting with AI offerings and projects through its Emerging Media Lab. A group of community colleges in the midwest, Rocky Mountains, and northeast joined forces to develop shared classes about AI. Some Chinese universities are expanding their AI course offerings. (archived) A Stanford University professor developed a generative AI tool, Evo 2, which can produce DNA sequences of more than one million nucleotides. Research into AI continues, as with this Columbia Graduate School of Journalism study on the technology’s citation quality. I’m curious to see if the bimodal cloister/starship model gets any traction. Elsewhere nearly every campus I visit virtually or in person has instructors trying out a range of pedagogies, from having students use AI to generate writing which they critique to teaching students hands-on skills.

Another open question is how to prepare students for the post-AI world, especially for the labor market. This is truly an open question, as there’s no consensus about what AI will do to the workplace, the choices being: creating lots of unemployment and underemployment; creating jobs which don’t exist now; infusing AI into job skills. There’s also no consensus about how colleges and universities should prepare students for that cloudy future. Some of us have argued that a liberal arts undergraduate experience is best, given how it teaches flexibility of mind.

Not discussed very much (as far as I’ve seen) is the possible impact of agentic AI on higher education. Ray Schroeder has some ideas. Yuhan Gao offers more. Melik Peter Khoury is skeptical.

Meanwhile, substantial academic opposition to AI continues. I’ve seen many instances of this over the past year and examples are ready to hand. John Warner has been consistently opposing AI in the writing classroom through his Inside Higher Ed columns and his Substack. There’s a “Refusing GenAI in Writing Studies: A Quickstart Guide.” Bluesky has a steady stream of criticism from faculty and staff members. One colleague told me about giving a talk about AI at a college this spring, following which a faculty member took the podium to read a prepared statement slamming my colleague, AI, and the college for arranging the session. I’ve gotten some static and even insults for my work. Most recently Eric Hayot and Matt Seybold published an essay in the Chronicle of Higher Ed calling on academics to organize against AI.

The substance of the critique is well known by this point, charging AI with: reproducing biases, conducting unfair labor practices, infringing on copyright at a massive scale, aggregating undemocratic power to a handful of gigantic companies, reproducing colonialism, and consuming dangerous amounts of electricity and water. Over the year I’ve seen more critique linking AI to the political right, especially in the United States where technologies leaders from Meta to Apple aligned with Trump and most (but not all) Republicans have boosted AI. Additionally, Ronald Purser slammed the CSU system for making the OpenAI deal which he saw as part of a strategy to attack the humanities. (Here’s my summary of AI opposition, which I really should update.)

Using AI in this post

As with all of my Substacks, I wrote all the text. Once the draft shaped up, I shared some paragraphs with ChatGPT and Gemini to see their reactions, much like asking friends for feedback. I also asked those bots some questions to get a sense of what we futurists call “the official narrative,” in order to make sure I hadn’t missed any major points. I didn’t ask for any writing.

For this post’s images I used Midjourney, as usual, as I tend to get the best results there. You can see prompts in the caption under each. In addition, or in contrast, this time I turned to Better Images of AI to look for more, on Alan Levine’s urging, and used one above (full credit at end of this post).

Now, as with my last post, I used NotebookLM for extra help. For that post I uploaded all of my Substacks from 2025 into a new notebook. For this one I asked it questions about not AI in general, but AI and academia in particular. What NotebookLM saw surprised me. It emphasized the split between AI users and opponents over all, which wasn’t the impression I had of my work, but made sense on review. It also really liked the starship/cloister model, and I’m not sure why. It didn’t generate any new Studio documents (slides, infographics, podcasts, videos) based on my queries; those seem based on the total corpus established, rather than individual queries on it. However, some of the slides did focus on education, like this one:

What might these observations of 2025 tell us about 2026?

So much depends on what happens to AI in the world beyond academia. Technological developments independent of the academy can change the tools we already use and offer new opportunities and challenges. National and subnational policies can impact us. How popular attitudes shift also impact college and university use, such as if we see a bubble burst. The impact of technology-driven economic dislocation can have all kinds of academic impacts. Heightening geopolitical tensions over AI can play out in terms of international student recruitment and student, faculty, and staff resistance. If American Democrats pick up AI opposition as a major cause, which they might, academic opposition could become more politicized. I do wonder how academia will respond to AI search’s pressure on the web. Will this reduce incentives to produce web content, from open access journals to departmental and disciplinary web resources?

That said… I expect we will continue to engage with AI through 2026, even if there’s a bubble bursting or market correction. Unless the major tech companies completely give up on AI, which seems unlikely, the technology should be present and elicit our attention. In which case we can expect many of the above trends to continue: concerns about cheating, conducting research into AI, trying out institutional and pedagogical experiments, grappling with big tech firms, etc. But that’s only a first level approximation. Trends can always shift and mutate.

I haven’t seen much academic use of open source LLMs. Perhaps I’m missing some of that which is happening under the radar. Maybe we’ll see academics follow some technologists and use open LLMs for reasons of expense and flexibility.

It is difficult to get a handle on how many academics use AI at all, much less who uses what tools for which purposes. From what I’ve been able to see AI use is widespread among students. It seems used to some degree among staff and faculty, based on the accumulation of individual stories, but this is still difficult to determine with any precision. We should expect this foggy picture to persist in 2026. In addition, we may see operational uses start to clarify. (Note to reader: this is an excellent research project for campuses to conduct on their own population, especially with student participation!)

More institutions might take up strategic responses to AI. I suspect governmental, funder, corporate, and student (and traditional-age students’ parents) concern about the changing labor market will drive some changes, such as creating new majors or expanding old ones, setting up AI centers, encouraging AI literacy, and revising career services’ practices. Community colleges may have the edge here, being so plugged into local labor needs. Unless a fix for cheating appears that we broadly adopt, that problem will persist. I expect more commercial and open source responses, along with more nostalgia for bluebooks.

I am very concerned about higher ed’s reputation in the AI revolution’s wake. I don’t mean higher education’s objective reality but how people perceive it. As I’ve said elsewhere, it’s possible that public opinion of colleges and universities, already souring in the United States, could worsen. We might appear to be out of touch with the modern age, especially if acts of resistance get popular attention. Our rising published prices (as opposed to what people actually pay) could make AI appealing for education. Our inability to conduct assessment without cheating could sap the value of our degrees. If people perceive graduates to struggle with getting jobs, including positions suited to their education, opinion can decline even further.

Conversely, we might see popular anxiety about AI make the academy look better in contrast. While AI hallucinates, needs irksome data centers, is generally offensive and chaotic, colleges and universities could appear more reliable and trustworthy. This will depend to an extent on how people see us respond to the technology. Public intellectuals speaking from their professional expertise might help that come about.

Within post-secondary education I expect opposition to AI to continue. There are many faculty, staff, and students who strongly oppose the technology and the tech seems unlikely to transform in a deep way to meet their concerns (although cf open source again). As noted above, academic opposition to AI may become politicized. I’m not sure what forms that could take, but can speculate. It’s easy to imagine scholarly research, teaching, and public statements against AI. We may see some movements to block or reduce AI on campus. Perhaps the intensity will rise to the level of the Gaza protests, and I make this comparison to highlight the possibility of institutional tumult. And the Purser argument - that a campus leadership took funds from certain fields to support AI work - might gain currency. For years I’ve asked audiences to imagine a “Butlerian jihad”1 against AI, which often mystified people; recently this prospect seems to be swimming into view.

I wonder how the academy will respond to AI’s companionbot uses. (Here’s my sketch of those.) For example, how many institutions - or units, like student life - will support AI for mental health purposes, either explicitly or tacitly? Will other institutions create official, branded AI bots which they hope students will spend a good deal of time with?

…and I did vow to compress this newsletter and have clearly not succeeded, so I’ll stop here. Over to you, dear readers. What did you see of AI in higher ed this year? What do you anticipate for the next?

(“Pas(t)imes in the Computer Lab” image by Hanna Barakat & Cambridge Diversity Fund / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/)

First, I’d like to summarize what I’ve tracked and analyzed about AI in education and the world at a top level, then will describe at a meta level what one LLM application had to offer as I created this post. I’ll conclude with some potential directions for 2026.

As always, I encourage you to share your thoughts in the comments box below.

One more note: my Substacks have gotten very long this year, so I’m experimenting with compressing this one. Let me know how it works for you.

1 The year’s trends

I divide what follows into categories we’ve used throughout the year. The focus is the world in various domains; I’m saving higher education for a followup issue.

AI technology

We didn’t see much in the way of breakthroughs in 2025, with the powerful exception of Deepseek’s stunning success last winter with relatively cheap and constrained hardware. Otherwise we saw many incremental improvements in terms of quality, hallucination control, multimodal input and output, code quality, context window size, and more.

The drive towards agentic AI continues. In addition to answering questions and generating content, the goal is for LLMs to manipulate software and other parts of the world on our behalf. We may be seeing agentic features appearing in the big platforms, albeit in limited ways.

The divide between closed and open source AI also continued. While proprietary titans win the lion’s share of attention and, as far as I can tell, use, open source applications keep impressing. The latter appear in use across the board, according to many stories, and China both makes and uses open source widely.

AI may also be eating the world wide web. Google deployed “AI mode” for search, offering a Gemini-generated response to a query before providing the traditional links list, and this has won some measure of adoption. This in turn means fewer clicks on and inbound links to web pages, which threatens to weaken conversations and business models alike.

One more theme: as training sets run out of human generated content, there’s the possibility of creating artificial worlds and replicas of this one for future training.

Geopolitics

Perhaps the leading geopolitical dimension of AI has been increasing US-China rivalry on the technology. Each nation’s leadership has committed funding, political capital, and prestige to a kind of LLM race aimed in the direction of artificial general intelligence (AGI), albeit in different ways. We’re seeing this play out in various forms: China’s emphasis on smaller implementations (Deepseek) and also open source; academic politics; the possibility of divergent AI ecosystems. The Trump administration followed the Biden in trying to block Chinese access to key hardware, which seems to have backfired as Chinese technologists innovated new ways to make LLMs work on less than stellar infrastructure.

Other nations have mounted efforts to catch up, from British and Middle Eastern data centers to Switzerland’s public benefit AI Apertus. An interesting theme which emerges from this global competition is “AI sovereignty,” the attempt to run LLMs on a nation’s own infrastructure. (Perhaps “AI autarky” was too extreme a term?) We can find examples around the world, from Europe to South Korea and Taiwan.

Within each nation there are many political responses and uses of AI. Some nations and subnational units have sought to encourage AI development, or to take advantage of it. Israel used AI to aid in targeting during its Gaza war. In the United States the Trump administration and Republican party have generally doubled down on supporting the tech, while some progressive opponents critique it. America’s federal system also saw a struggle over the rights of states (as opposed to the federal government) to issue AI regulations.

Economics

The largest question looming over AI was probably the bubble one. Was the AI sector overvalued, overloaded with too much investment, and about to pop or collapse? Or were we witnessing the warts and all buildout of necessary infrastructure, along the lines of 19th century railroads, early 20th century phone lines, and 21st century broadband? As of this writing we see investment continue to flood into LLMs, especially for building out a vast data center ecosystem. Electricity and water demands have soared.

2025 saw the first impact of AI on the labor market as some firms laid off workers, citing their replacement by LLMs, and others stopped hiring young staff. Unemployment as a whole hasn’t taken off - 4.6% in the US, a mostly steady rate - but we’re seeing what might be employers taking the first step up a tall staircase.

On the business side we’ve seen OpenAI and Google continue to struggle for market dominance, matching each other’s offerings and pushing hard to take the lead on many metrics. Gemini 3 won impressed reviews while ChatGPT 5 underwhelmed.

Struggles over copyright continue, with lawsuits against AI companies proceeding through legal systems. One major finding against Anthropic hit that firm’s use of literally pirated materials, not the large general data scraping practice.

Culture and society

How have we acculturated and responded to the AI revolution? I’ve tracked this in some detail. One fascinating aspect has been the rising use of AI as companionbots, or our anthropomorphizing the tech more broadly. 2025 has seen story after story of people treating LLMs as companions, conversation partners, romantic objects, soul mates, therapists, religious advisors, government officials, and friends.

Another aspect is the rise of what I very awkwardly called the intermediary layer: when AI mediates between people, simply put. I use AI to prepare an email or web page, then you use AI to find, understand, and respond to it. This can happen at a relationship level, as well as in domains of economics, politics, culture, and more. I haven’t seen a lot of commentary on this point, but the practice certainly seems to exist in the world and to be growing.

AI artifacts continue to appear across the digital world, as people and bots use LLMs to produce all kinds of media objects. The derisive term “slop” seems to have become the word of art to describe such artifacts, and you can see a great deal of resistance and dismay in it. Given human ingenuity in the arts and storytelling, it should come as no surprise that people use AI to craft a range of items, like producing videos of deceased people to pop songs to science fiction shorts. Less obviously, AI use appears in mixed forms, as people employ the tech while making their own material, like this Japanese writer did.

The cultural divide over AI persisted through the year, with hype and proponents meeting critique and opponents.

2 Then AI helped me make this post

Getting meta: I drafted an outline of this post in Substack and wrote some paragraphs, first from memory of the year, then by poking into the archive of 20+ posts.

Next, I turned to AI to see what it might add. Specifically, I created a new notebook in NotebookLM, feeding it URLs from all of the year’s Substacks. After it ingested them, I used the chat interface to ask it questions about what it saw of the year through my writing. The responses were pretty good, hitting themes I had in mind, which wasn’t redundant, as NotebookLM confirmed my intuitions. It also offered some interesting summaries, which led me back to some Substacks in new ways. It gave education a lot of attention, which I’m taking to the next post.

NotebookLM is a busy creature, providing a bunch of options in its Studio. It whipped up an infographic for the year:

It created a mind map:

It also created a data table (of policy developments), an audio overview, some slides, and a video overview. I think of these as reflections or refractions on my writing, a version of readers responding to the posts. I found it useful.

I didn’t use NotebookLM to write these paragraphs, although I changed up some sentences in response to what the AI generated. Nor did I use it to make the images in the first part; those were Midjourney’s. (As I’ve said before, I’d prefer to use Creative Commons-licensed photos, but it’s hard to find the right ones for my purposes, and it is of course apt to use AI given the subject of this newsletter.)

3 Thoughts for 2026

So what does all of this suggest as we consider the next year?

At a first order of approximation we can extend these trends forward under the principle of “if those goes on…” In this way we should expect a continued US-China AI rivalry, pushes for AI sovereignty, Open AI vs Google, more AI-explained job freezes and cuts, more companionbots, more use of AI to create stuff, more opposition to LLMs, and so on.

The trick is to anticipate how those trends might change, how they could grow, shrink, or mutate beyond simply continuing as they did in 2025. There are a lot of factors to consider, from major political decisions to wild cards. At the tech level, will LLMs start to run into quality issues as they run out of data to scrape? Will open source advance further in quality and adoption? Might AI hit a market correction or bubble in the US? Would a disaster popularly attributed to AI cause public revulsion and pop the bubble? Would Trump drop the AI cause as a result, hating to be contaminated by losers? If not, will American politics break firmly on AI along partisan lines, with Democrats denouncing the stuff and the GOP doubling down on it?

I’m agnostic on many of these for the simple reason that generative AI is now so vast a thing, so embedded in so many human and technological systems, that it’s hard to get good data on much of it and difficult to solve the resulting foggy, many-body problem. My gut tells me anti-AI sentiment and critique will develop still further, that we could see a market correction, that the intermediary layer will expand as will agentic capacity, and that the Trump administration will be chaotic on AI. But that’s intuition, albeit informed by continuous horizon scanning, analysis, and reading the literature. I feel on firmer ground in forecasting more creative uses of AI will occur, from art to storytelling to technological development, based on what I know of the history of technology and human innovation. I also feel a bit more confident, and also sad, in projecting AI will continue to gnaw at the web. So many institutions, including academia, still struggle with using the web in its full or even basic form that we might hit peak web in 2026.

I’m also skeptical that we’ll see anything which most people would recognize as human-level intelligence from an LLM next year. The past year’s incremental improvements suggest more of the same, unless a staggering breakthrough appears.

Thinking through such questions, I’ve been reading and revisiting commentaries new and old. A recent one from Tim O’Reilly and Mike Loukides asks us to view forecasting AI in terms of a different question: is generative AI an ordinary or extraordinary tech? That is, either

AI faces the same barriers that every enterprise technology faces: integration costs, organizational resistance, regulatory friction, security concerns, training requirements, and the stubborn complexity of real-world workflows. Impressive demos don’t translate smoothly into deployed systems. The ROI is real but incremental. The hype cycle does what hype cycles do: Expectations crash before realistic adoption begins.

or:

we aren’t experiencing an ordinary technology cycle. We are experiencing the start of a civilization-level discontinuity. The nature of work changes fundamentally. The question is not which jobs AI will take but which jobs it won’t. Capital’s share of economic output rises dramatically; labor’s share falls. The companies and countries that master this technology first will gain advantages that compound rapidly.

AI is strange, in this sense, a weird mutation which blazes a new path.

I think it could go in either direction now. There’s a lot of pressure to domesticate LLMs, to see them through Gartner’s famous hype cycle and into legible enterprise and consumer slots. We are very busy people, after all, with a lot on our plates. It would simplify things very much if we could use AI like Excel or an XBox. Business owners would like to slot LLMs into the classic modes of either replacing labor with capital or adding tech to leaven productivity. Investors would like this to happen in a way which produces profit. Yet… the technology is strange, alien enough to break the mold. Human ingenuity can run with it in bizarre directions, as we’ve already seen.

I’m not sure this is entirely or even mostly a technical problem. I suspect it’s more a question of culture, economics, and politics - of perceptions - how we think about and use the tech en masse. That evolving collective shift will either lead us towards a singularity (in O’Reilly and Mike Loukides’ sense) or into another turn of the old tech wheel. If that’s right then 2026 is shaping up to be a struggle over where we take AI. There will be immense and broad-ranging competition to frame and reframe LLMs. We should watch pop culture, political arenas, business plans and marketing, boycotts, and workshops to see what we make of it. We should watch - and participate.

Now it’s over to you, dear reader. What do you make of these scans of 2025 and glances to 2026? What do you think of using NotebookLM? And was this issue too terse or compressed to a good level?

(Next up: education!)

Today we’ll examine what’s been happening in the world of AI research and development. New releases and iterations keep appearing, churning the environment and giving us more choices. Let’s scan the AI tech horizon.

(If you’re new to this newsletter, welcome! This is one of my scan reports, which are examples of what futurists call horizon scanning, research into the present which looks for signals of potential futures. We can use those signals to develop trend analysis, which we can then use to create glimpses of what might come next. On this Substack I scan various domains where I see AI having an impact. I focus on technology, of course, but also scan culture, government and politics, economics, and education, this newsletter’s ultimate focus.

It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for examples.)

Now, on with the scan. There’s a lot to explore here. I’ve divided the following by businesses or business-like entities, followed by some quick reflections as time and space allow.

OpenAI

…has been busy. They released ChatGPT 5.1. This has two versions, Instant for general use and Thinker to show its work, with an Auto function which picks a version based on a user’s prompt. The company describes this release as “warmer by default and more conversational,” improving or adding more personas or personalities: “Default, Friendly (formerly Listener), and Efficient (formerly Robot) remain (with updates), and we’re adding Professional, Candid, and Quirky.” OpenAI claims improved accuracy and speed.

OpenAI also provided a branching function. With this you can fork a discussion into two or more paths, returning along them as you like. Ars Technica offers a nice analogy: “Think of it almost like creating a new copy of a ‘document’ to edit while keeping the original version safe—except that ‘document’ is an ongoing AI conversation with all its accumulated context.”

The House of Altman also released ChatGPT Pulse, a kind of agentic summary which summarizes the past day of a user’s activity as seen through the app. Tom’s Guide offers this glimpse:

At the end of the day, Pulse takes a look at your chat history, memory, feedback and then does a round of asynchronous research, synthesizing all the information you’ve shared or asked during the day.

The next morning, it delivers a curated feed of updates in the form of visual cards you can scan quickly or expand for details. Think of it as a morning briefing, but one that reflects your personal goals, habits and even your calendar.

I think it’s only available for Plus users.

What received more attention that these offerings was the new version of Sora. Sora 2 (only available for iPhones now, irritatingly) is a video generator, much improved from 1.0. For example:

Users can also insert their own video content into a clip. Additionally, there is now a Tiktok-style Sora app where you can browse a stream of Sora 2 clips. Improved quality has given rise to a new wave of concern about the trustworthiness of all visual media, if Sora can convince many viewers of its reality.

Google

The search+ giant has also been very busy, fighting to lead the AI revolution. It released a new version of its main AI service, Gemini 3, to a lot of respect and applause. Many reviewers gave it improved ratings from its predecessors, citing improved speed, output quality, better coding, various kinds of reasoning. It scores well on multiple ratings on LMArena’s leadeboard. It has more agentic abilities. It competes well with OpenAI’s ChatGPT 5.1. Salesforce’s CEO rhapsodized:

Marc Benioff @Benioff Holy shit. I’ve used ChatGPT every day for 3 years. Just spent 2 hours on Gemini 3. I’m not going back. The leap is insane — reasoning, speed, images, video… everything is sharper and faster. It feels like the world just changed, again. ❤️ 🤖

There’s much more. Google launched a new version of its image creator, Nano Banana Pro, which I think is embedded within Gemini. Reviews have claimed sharper images, faster results, more fine grained user control. They updated Veo to version 3.1, which is paid only, as far as I can determine. This promises better audio and video output.

Google Scholar now has a conversational AI feature. So far I’ve seen that it lets me ask questions in natural language, but the output is formatted as before, a list of citations. Google Earth now has AI features, using LLMs to connect multiple data streams for what it calls geospatial reasoning. The company updated Sima, an agent which can follow instructions in virtual worlds, following and describing a reasoning chain. Google posted about VaultGemma, an effort to reduce the amount of private data LLMs retain.1

Google also launched Mixboard, a whiteboard tool somewhere on a continuum between the company’s Jamboard (light) and Miro (advanced). You can prepopulate a Mixboard by prompting it, then add or edit the results. Here’s one example I quickly whipped up:

Google Cloud and Stanford University researchers published a framework for small language models to use multi-step reasoning for the first time. Supervised Reinforcement Learning (SRL) “reformulates problem solving as generating a sequence of logical ‘actions’.”

Also on the academic front, Google announced Learn Your Way, which looks like an AI tool for teaching a learner about a document in a personalized fashion. Users apparently upload a pdf of something they’d like to study, then the site asks some particular questions: grade level (K-12), other interests, media formats. It then generates questions, quizzes, videos, summaries, and assessments. I wrote “looks like” because I’m still on the waitlist, and am working from a video clip on the announcement page. At this distance I can imagines Learn Your Way becoming an out of the box tutor for challenging topics, then users (or teachers) daisy chaining a series for a curriculum.

On the hardware side, Google announced its latest customized silicon chip, Ironwood, the 7th generation of its Tensor Processing Unit (TPU), will be available for purchase. More ambitiously still, the company announced plans for putting AI in orbit. Project Suncatcher “envisions compact constellations of solar-powered satellites, carrying Google TPUs and connected by free-space optical links.”

As a result Google seems to have taken the lead in AI at present. To illustate, here’s what Gemini gave when I asked for an image of “an AI company taking the lead over other AI companies”:

Everyone else

Moonshot.ai launched open source Kimi K2 Thinking, which is getting fine reviews (for example) (for example). You can find more info here, code here, and API docs here.

Anthropic published Claude Opus 4.5, claiming improvements over previous versions, including a broader context window and ameliorated coding output. An Ars Technica report notes Opus 4.5 exceeds all comers in accuracy. Anthropic also continues to publish research, such as an analysis of user productivity on its tools. Claude can now end conversations the software deems to be harmful.

Elsewhere, Microsoft continues its AI work. The company is apparently developing its own AIs, independent of OpenAI’s. Two have been announced, MAI-Voice-1 and MAI-1-preview. Redmond launched Agent 365, a “control plane” for its agents. On the hardware side, Microsoft announced a “superfactory” in Georgia and Wisconsin with “hundreds of thousands of advanced Nvidia GPUs.”

Meta continues its AI work. The company is pushing to release a new version of its open weight tool Llama before January. Meta’s AI for Good is working to build bad weather detection tools for Florida. And some retraining is under way to keep bots from writing inappropriate content to minors. On the hardware side, they apparently reached out to Google to use its chips, which depressed Nvidia’s stock price.

Deepseek upgraded its China-facing app to work more effectively with chips produced in that nation. V3.1 also improved in quality.

A group of three Swiss public universities - ETH Zurich, the Swiss Federal Technology Institute of Lausanne (EPFL), and the Swiss National Supercomputing Centre (CSCS) - launched Apertus, an open source LLM intended for the public good, and which we noted this summer. According to one Swiss acount,

“We aim to provide a blueprint for how a trustworthy, sovereign and inclusive AI model can be developed,” said Martin Jaggi, professor of machine learning at the Swiss Federal Institute of Technology Lausanne EPFL.

The Verge notes that it followed European regulations. One interesting point about Apertus concerns language:

Trained on 15 trillion tokens across more than 1,000 languages – 40% of the data is non-English – Apertus includes many languages that have so far been underrepresented in LLMs, such as Swiss German, Romansh, and many others.

And here’s a kind of FAQ.

Amazon continued to work on AI. Leaked documents point to developing agentic software. Amazon also won a $50 billion contract with the American federal government to provide AI to multiple agencies.

Apple added AI to its Airpods, designed to translate languages for the user. Apple also started offering code to work with the MCP agentic protocol.

X.ai launched an AI-powered encyclopedia called, inevitably, Grokipedia. Business owner Elon Musk claimed it would be free of political bias. Reviewers have noted times when it espouses right wing views or draws on such sources (for example). One reviewer suggested that it shows how to use AI to improve the least-developed Wikipedia entries. X.ai also launched two sex-themed chatbots. Additionally, the company is apparently working on world models.

Friend launched its hardware pendant of the same name. It contains a microphone which picks up nearly sounds. It transmits text to its user’s smartphone. Friend elicited backlash, with people tearing down and mocking posters in New York City and criticism appearing online. Meanwhile, AI toys are a success in China.

What might we derive from these developments?

Overall, research and development continue along a series of domains we’ve been tracking. Many LLMs received incremental, versioning improvements, especially in chat, as well as image and video creation. There are pushes to expand the scale and capacity of LLMs, as well as some efforts to produce smaller ones.

Agentic work keeps coming, but often folded in with other tools. In our AI seminar we debated what precisely defined an AI agent. My sarcastic idea was “agents are what companies can charge more for,” but a more sober answer might note that agents are less separate from non-agentic AI and more blended in.

There’s rising interest in software fixes for conversations which turn abusive or dangerous.

Open source AI keeps improving, usually without much attention. Chinese developers are especially learning into open source. Once again I remind readers that generative AI is not a US-only thing.

Over to you, dear readers. Have you tried any of the new tools listed here?

(thanks to Steven Kaye, Garthster Lucerne, Ruben R. Puentedura)

Double, double, toil and trouble;

Fire, burn; and cauldron, bubble. (Macbeth Act IV scene 1.)

Greetings from the Halloween season, dear readers. Here in the mid-Atlantic United States we are deep in autumn, with leaves and temperatures descending and the dark advancing. Summer has truly passed and winter looms, such as it is here. The mood is dramatic, which makes a fine setting for today’s topic: the latest economic developments in artificial intelligence.

There is a *lot* going on, and this issue feels like it could be endless, especially given the torrents of commentary and analysis out there. On revision, Substack tells me I’ve already broken the “too long for email” barrier. So I’m going to hit some highlights, hopefully spark some comments, and start working on the next posts.

(If you’re new to this newsletter, welcome! This is one of my scan reports, which are examples of what futurists call horizon scanning, research into the present which looks for signals of potential futures. We can use those signals to develop trend analysis, which we can then use to create glimpses of what might come next. On this Substack I scan various domains where I see AI having an impact. I focus on technology, of course, but also scan culture, government and politics, economics, and education, this newsletter’s ultimate focus.

It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for examples.)

Now, on with the scan. We’ll look at actual uses of AI in the workplace, touch on emerging business models and practices, then dive into the bubble question, followed by a look at workforce issues. We’ll conclude with some futures reflections.

Actual uses of AI in the workplace

How people actually use AI is a topic I’m fascinated by. It’s not easy to determine, as many studies rely on self-reporting, which can be dubious. It’s also a field of great scope (hundreds of millions, soon more than a billion users) and complexity (so many programs, applications, and possibilities!). So let’s see what we can best learn now.

A new paper by Microsoft researchers identified leading AI uses, based on Copilot data from 2024. There’s quite a lot in the document, so I’ll pull out the parts which struck me. One top-level finding was that no job was 100% replaceable by Copilot, at least in 2024. Another viewed the technology as a conversational partner, rather than as an information retrieval service: “In terms of AI performing actions, we show that it often does so in a supporting role to the human acting as a coach, trainer, or advisor.” (cf this post)

A third result concerned specific kinds of jobs and their use of AI:

we find the major occupation categories with the highest AI applicability scores are Sales; Computer and Mathematical; Office and Administrative Support; Community and Social Service; Arts, Design, Entertainment, Sports, and Media; Business and Financial Operations; and Educational Instruction and Library.

The authors broke professional functions down to general tasks:

I notice information-gathering is the leading user purpose, but not what the team thought Copilot was doing. Instead, leading AI actions were communicative.

Next, the researchers decomposed functions still further:

Again that divergence between information-gathering and communication.

In another use case, American financial giant JP Morgan Chase continued its drive to include AI in daily operations. The current focus seems to be equipping staff with AI assistants, especially agents, which can speed up task performance. At a smaller scale, Nextdoor will deploy AI as a recommendation engine for local services and Dow Jones added a third AI-powered translated content service (French, after Korean and Japanese). Delta Airlines is rolling out an AI tool to personalize fares for individual customers (archived copy).

There were also cases of companies using AI badly. Consulting firm Deloitte generated a report for the Australian government, which turned out to have used AI poorly, throwing up errors which a human expert would have caught. Let Brent Anders explain:

AI business models evolve

The AI industry continues to try new business models, as do companies using AI. Perhaps the biggest story in this area is OpenAI which once again shifted its unusual corporate structure. Now there is a nonprofit organization, the OpenAI Foundation, which owns part of the for-profit OpenAI Group PBC (public benefit corporation). The official line is for the business engine to power the non-profit public good. On a sarcastic note, I do enjoy Matt Levine’s line about how OpenAI’s CEO wants to ask AI for business ideas: “It really is the greatest business plan in the history of capitalism: ‘We will create God and then ask it for money.’” This is especially sharp in the wake of news that the firm lost $12 billion last quarter.

OpenAI also set up a marketplace function, allowing users to buy stuff through the platform, especially agentically. Retail behemoth Walmart partnered with them, so that users can use AI to buy their products.

Content licensing stories continue to appear, as when Perplexity licensed Getty Museum materials. Gannet licensed USA Today content to Perplexity as well.

Another story appeared which we might consider an anti-model. Reddit announced it would block the Internet Archive from archiving its contents, as the former feared AI companies were scraping material from the latter.

Investment boom or bubble?

Are we in an AI bubble? I’ve tracked this question for years, collecting data, issuing forecasts and scenarios (most recently), and advising audiences to pay careful attention to possible ways the boom might fall apart. Ever since my last post bubble talk has continued to loom large. A series of major players raised the idea, including Jeff Bezos, who offered the idea that we’re in an industrial bubble, where some companies fail but the overall sector keeps growing and providing value. Starting in August Sam Altman started speaking of bubbles, overvaluation, and losses to come. (Here’s the most passionate and extensive argument for a bubble about to pop.)

Money keeps thundering into AI as speculation grows at a blistering pace. Startups keep receiving funding, even in the billions (for example) (another example) (perhaps notoriously). Anthropic raised more investments on a value of $180 billion. The market now values OpenAI at $500 billion. Nvidia’s value just broke $5 trillion. Microsoft, often derided in the technology press, saw its value crack $4 trillion based in part on its AI work. Other AI businesses are seeing prices balloon:

Players are taking on more and more debt now. Hundreds of billions of dollars are pouring through the sector now, especially to keep building out data centers and their requisite infrastructure. One sign of this is that we spend nearly as much on those centers as on building offices, and seem likely to exceed that number soon:

For another small (!) example, Anthropic made a deal to spend billions on chips from Google. In other words, if the bubble is inflating to astonishing heights, it shows few signs of popping as of this writing. Obviously things could burst.

Yet the scale of AI’s financial footprint and the sheer amount of money it’s drawing in from the rest of the economy is causing worries beyond the sector itself. Analysts are furiously trying to figure out just how much of the American economy is AI, and how much of the nation’s growth depends on the nascent industry. One popular estimate notes that seven AI-heavy tech firms now account for one third of the entire American stock market’s value. More: “Harvard economist Jason Furman recently said that AI investments accounted for nearly 92% of U.S. GDP growth in the first half of 2025.” In those same months “AI-related capital expenditures contributed 1.1% to GDP growth, outpacing the U.S. consumer as an engine of expansion,” according to JP Morgan. More starkly,

Ruchir Sharma, chair of Rockefeller International, told Fortune that in a way, “America has become one big bet on AI,” and warned that “AI better deliver for the U.S., or its economy and markets will lose the one leg they are now standing on.” [emphasis added]

There are several cases where the money is running in the other direction. LLM prices have been falling on a per unit basis, especially due to the excellent work in open source AI:

Ernie Tedeschi (Budget Lab at Yale University) offers some caution. Yes, AI spending is contributing to GDP growth, but a chunk of that is imported, so not part of the American economy in a key sense. There are also strong lags in data, sometimes lasting years.

Yet problems are emerging. Some of the major AI spending takes place within the sector itself, as companies buy from or trade with each other. Chipmakers, cloud companies, and AI firms are very close with each other:

And now it looks like Amazon and Google each own a chunk of Anthropic. You could call it a financial ouroboros, each creature in the AI bestiary ultimately gnawing on the others’ tails.

Meanwhile, some consumer utility bills are now rising, driven to some degree by datacenter build outs, that driven to some degree by the AI boom.

Let me end this section with two cautions. First, economist Noah Smith isn’t as disturbed by the snail-biting as others. He sees the practice as likely a mix of business diversification and prudent financing. Second, this Neutron post suggests a bear market is likely in the immediate term (so, a bad market correction, a bubble bursting) followed by continued growth, a bit like my Jagged Infrastructure scenario. Overall, I am worried that demand for AI, both paid and unpaid use, will fall short of very, very, very expensive supply.

AI and the world of work

How AI will impact labor is still an open question with signals and developments pointing in different directions.

At the most ambitious level, Walmart’s CEO stated that AI would impact every job in his vast company, and not just there: “Maybe there’s a job in the world that AI won’t change, but I haven’t thought of it.” He mention position cuts as well as creating new ones.

What kinds of jobs will be changed, ended, or created anew? Recent weeks have seen hiring freezes and cuts to numerous white collar professions. That Wall Street Journal article cites thousands of layoffs to corporate staff at Amazon, Booz Allen Hamilton, General Motors, Molson Coors, Rivian Automotive, Target, and United Parcel Service. Amazon is leaning into both AI and robotics to reduce and redefine its workforce. A Washington Post report notes cuts and hiring slowdowns at Goldman Sachs, JPMorgan, and Nestlé. A Goldman analyst described a hiring drop in the technology sector, especially for young workers. CNBC reports that YouTube offered buyouts to some workers. One job facing automation might be the very venture capitalists who fund AI. A new research project trained LLMs to pick startup winner and performed, apparently, at a comparable standard to human peers.

There are countervailing signs. An Australian bank laid off four dozen staff, then rehired them when AI underperformed internally. I keep hearing reports of companies struggling to integrate AI into operations, having to expend more staff time to address issues with output quality and workflow redesign.

Reflections

Overall it feels like 2025 is something of an inflection point for AI and the economy, a moment of furious work in the midst of deep uncertainty, or risk. We’re building out a huge ecosystem in great uncertainty with vast sums of money swarming through multiple nations and it’s not clear if we’re headed for a railroad future of a rich, globally used infrastructure or a technological misfire, a 8-track for the 21st century.

Companies are changing work and laying off staff, but tentatively, to a degree indicating it’s still early days. I can imagine firms using the fear of AI to keep down compensation and quash unionization drives.

Some business models are continuing, like content licensing and expanding on-platform marketplaces.

The use cases are very interesting. There’s a split between enhancing human work and replacing it, which has been a long-running theme. There’s also a split between users turning to AI to get information versus those having conversations with the bots, which for me reflects our tension between information retrieval and AI content generation. Licensing and its opposite continue to chug along.

That’s all for now. The scanner turns its attention to other domains.

(thanks to Todd Bryant and Hernan Carvallo for fine links)

Just a short post today with an announcement an invitation. This Thursday I’m holding an experimental video session. On the Future Trends Forum* we’ll have a group brainstorm/design exercise, where together we imagine how a new campus might appear and function in the age of AI. I’ll facilitate, using various tools to help people develop and share ideas: polling, whiteboard, chat, design thinking prompts.

This is a new thing for us. We’ve held scenarios and role-playing events on the Forum, but this is the first time we’re doing a design from scratch exercise. It’s an experiment! (Any thoughts or suggestions ahead of time appreciated, along with reflections afterwards.)

If you’re interested, you can register here or just join us from 2-3 pm ET on the 23rd.

*If you’re new to the Future Trends Forum, it’s my weekly live video program about the future of higher education. The usual session hosts one or several guests who have a particular angle on the subject. I facilitate the discussion, which is largely driven by audience questions and ideas. There isn’t any presentation or slideshow; instead, the program centers on conversation, as we believe combining multiple voices together is a great way of grappling with the future.

We’ve been doing this for nearly ten years now. There’s more information on our little website, including our YouTube archive with more than 450 recordings so far.

If there is an AI bubble, and it pops, how might that play out? And what may come next, if so? In this post I’ll introduce the bubble argument, then offer several short scenarios about how it might unfold.

These scenarios rest on some readings of technological history, along with observations of the present. They occupy the short and medium term future horizons, from the near future out to around five years. Drivers of change I’m especially interested in here concern government actions, cultural attitudes, geopolitics, and, of course, macroeconomics.

I rule out several things here. I’m assuming no nightmare scenario occurs whereby AI does terrible things to humanity. There is no artificial general intelligence breakthrough in this time frame. No clear successor to the LLM architecture appears at scale. The focus is on AI now as it hits a (putative) bubble.

Economic bubbles burst when the value of a thing drops quickly, after it has been shooting up for a while. After a peak an item’s price rises for a moment but nobody wants to pay it, and then the price starts to decline, then to fall. People tend to view the last stage of the price inflation as driven by speculation, poor analysis, hype, or frenzy. One classic example is the 17th century Tulip mania, where people invested ever-increasing funds into what appeared to be a very exciting plant, only to wake up and realize they’d massively overvalued it. In living memory we might think of the dot com collapse around 2000 (a/k/a “dot bomb”) or how we recently passed peak streaming video services.

(There’s also an argument that the US economy is basically flat, if you remove the AI boom. Possibly the finance capital now firehosing into AI would redirect elsewhere, but this is another topic.)

If you’d like to look into more analysis on the subject, I recommend Noah Smith’s post. He takes the discussion further with his characteristic combination of solid research and clear writing.

Noahpinion

America's future could hinge on whether AI slightly disappoints

A burning question that’s on a lot of people’s minds right now is: Why is the U.S. economy still holding up? The manufacturing industry is hurting badly from Trump’s tariffs, the payroll numbers are looking weak, and consumer sentiment is at Great Recession levels…

Read more

9 months ago · 184 likes · 39 comments · Noah Smith

Now how might this play out over the next few years? Here are some possibilities.

Extinction Event

AI’s value crashes quickly and deeply. Investors lose historically vast amounts of capital. OpenAI and Anthropic go out of business or just shrink to something very marginal. Google, Microsoft, Amazon, Apple cut back their services and eat the losses. A new AI winter appears as the field collapses for years.

As a result, the American economy goes right into recession, based on the huge capital losses and the collapse of so many loans. Other nations also suffer to the extent they invested in failed enterprises. Meanwhile, finance slides capital into other causes, like biotech, health care, and private space exploration (launches, mining, satellites), gradually spurring economic growth in a year.

Governments around the world turn away from AI in different ways. Politicians who once boosted the technology either fall carefully silent or attack it. Those who were skeptical proclaim their foresight. State funding levels drop. Militaries end contracts. President Trump mocks AI leaders as fools and con artists.

Culturally, we sour on the whole AI enterprise. People quietly or performatively drop creating content with AI. The collapse accelerates anti-AI backlash, with critics and opponents feeling vindicated. LLM leaders and boosters appear as villains or clowns in movies, tv, songs, computer games, and stories. Then, given short attention spans, we move on, culturally consigning AI to the dustbin of failed inventions.

There are interesting precedents for such deep collapses. Pneumatic tubes became an exciting technology in 19th century Europe, competing with the telegraph for communications and mobile humans for delivering stuff, but “les pneus” collapsed badly, broadly disappearing from life. I also think of how America’s human spaceflight enterprise went from the Apollo triumph and space shuttle successes, then stopped altogether for decades, reduced to renting seats on old Russian spacecraft just to get into low Earth orbit.

Jagged Infrastructure

In this scenario the LLM crash happens but AI work keeps going. How? The knowledge of how to create and maintain LLMs is now largely public, embodied in a large number of open source projects. There is also a lot of infrastructure now established which new projects can use. Perhaps cloud service costs start to decline, even for GPUs, after a peak. In short, the AI industry rebounds quietly, without so much furor this time.

Investors and government officials are chastened, but resume support in quieter terms than they once used. No longer is AI an existential threat to be wrestled with or a global revolutionary forces. Instead, finance and politics resume their embrace of LLMs as necessary infrastructure for the 21st century.

The model here is the bust and boom experienced by fiber optic networks around 2000, where people spent fortunes to pay networks underground then flopped. Another antecedent are the repeated railroad crises in the late 1800s, when startups used immense amounts of capital to lay tracks across the nation and repeatedly went bust, often in dubious circumstances. In both cases after the bubbles popped positive material achievements remained. Dark fiber was in the ground and railroad tracks upon it, ready for use. Bubbles inflate and pop but infrastructure elements are available for subsequent development.

We could think of the dot.com bubble bursting this way, too, as the 1990s saw the build out of web infrastructure: HTML, the browser, servers, growing awareness of web design know how, etc. The web grew after the bubble burst, and for the 2000s it saw some unevenness. We might think of Flash’s rise and fall, or the rise and demise of so many web editors before web 2.0 really settled in.

(See also “Tarnished Persistence” below)

Governments Take Command

The bubble pops and we power through the crisis via the force of state industrial policy. Governments are frustrated by the economic overshoot but think that AI’s strategic value remains and so they keep funding it. China, America, the European Union, Britain, the rich Middle Eastern nations step up to support the LLMs movement. They impose some new regulations and agendas, partly in an attempt to ward off another flop.

Public funding increasingly replaces private in the United States and some European nations. China and the Eurozone also see more governmental efforts to encourage businesses developing AI and its uses. States take a role in identifying, supporting, and promulgating productive uses of AI.

Culturally there are some interesting possibilities. We might see Americans pay less attention to AI since the government has it now and therefore it is dull. People in other nations might rethink AI as part of their national identity, either complaining about it accordingly or taking pride in it, depending on circumstances. Critics may charge AI with enabling governmental policies and attributes they disagree with, such as militarism, surveillance, corruption, or waste.

My model here is the American space program once more, because it was and is almost never popular. NASA has usually polled badly in the face of public disinterest and opposition. A steady stream of popular criticism expressed a range of arguments against space exploration, from “Whitey on the Moon” to anti-SpaceX sentiment. But Washington maintained support for space over decades, seeing it as vital to American interests: military, reputation, scientific, economic (in terms of spinoffs). The federal government managed multiple private entities in producing hardware and systems, much as the Chinese state is doing now with its energetic space effort.

Tarnished Persistence

As an alternative to Jagged Infrastructure (see above), after the bubble collapses the AI industry shambles along, mixing bursts of innovative activities with misfires and defeats. Money sloshes in and out of different projects. Overall, the sector grows over the next few years, albeit in stops and starts.

Governmental attitudes are all over the place, varying by nation, region, and political ideology. Some figures will try to rein in the dangerous technology through policies and legal action. Others will embrace AI for its exciting, outlaw nature. Regulation will become more intense, reaching the levels of policies concerning atomic power or biotech.

Culturally, we tend to view post-bubble AI as a misfit. In this view it’s a sector riddled with scams and dangers, with shadowy actors and strange projects. Legal actions occur against some AI actors. Yet there’s also some affection for the rascal, its ability to attract money, its eager welcome for new participants, its manic creativity. New ideologies appear, building on posthumanism, effective altruism, accelerationism, and various forms of artificial general intelligence aspirations and dread. Celebrities with alternatively promote or war against the tech with personal reputations and fortunes at stake.

Readers might guess that my model here is bitcoin, with its ragged rise punctuated with disasters, but with an overall upwards direction. There is lots of public disdain for it along with some love. It has a bad reputation yet keeps attracting money and demonstrating creativity.

The dragon’s turn

The AI bubble pops in many countries, but not in China. Earlier I mentioned the possibility of Trump backing AI at the governmental level even as the private sector sours on it. Here I imagine that he doesn’t, that he follows his usual practice of dumping anything and anyone smacking of loserdom. In contrast, European states lack the capacity to take the reins, especially as they devote more resources to remilitarizing against Russia.

In contrast China’s Xi Jinping stays the course, deeming AI to be a strategic necessity. Beijing continues to direct lavish support to that nation’s very energetic AI industry. Gradually Chinese research, development, products, and services become the leading ones in the world. AI suffuses China, but Chinese AI also appears in other countries’ services, perhaps as they participate in One Belt One Road, or because they see a good deal. Propaganda embeds Chinese AI within the national character, a source of pride, one more sign of the nation’s peaceful rise to global leadership. Every so often Beijing takes down an AI leader who appears too arrogant, reminding everyone who’s leading this tech revolution.

As this becomes evident some in Europe and America simply turn away, thinking generative AI a lost cause and China to be wasting resources on it. Others argue that LLMs are now aligned with the Chinese Communist Party, and think the world should resist it on ideological and geostrategic grounds.

In the cultural realm, AI figures in prominent and positive ways in Chinese stories and art. LLM inventors appear as heroes of the intellect and the economy, perhaps roguish at times, otherwise supporting Confucian virtues. We should see various AI tools and personas teaching Xi Jinping thought. Other countries which participate in Chinese AI may not make much of it culturally, or could portray the tech as exciting with a cosmopolitan flair. Countries which reject the Chinese technology may just not express it at all, culturally, or they might feature Beijing-aligned technology people and organizations as villains either formidable or deluded.

We can think of these scenarios as different pathways the world might take. We could also pick one or several which we’d like to achieve and plan towards it. Alternatively, we might select a scenario which we dread and figure out what’s needed to avoid its occurrence.

It’s also worth thinking about another classic feature of scenario thinking. When considering a set of possible futures, rarely does one come true and all others fail. Instead, reality offers a messy mashup. Sometimes multiple scenarios or elements thereof do occur, just spread across societies and/or over time. One region or population heads into one future while others enter another, or we cycle through several in a row.

Over on Mastodon Dave Wilburn offers one such take in response to an earlier, very truncated version of the above:

I feel like it’s going to be “all of the above”. The investor bubble will collapse under the weight of the industry’s financial schemes and lackluster revenue. And it will survive in niches here and there where it actually has some usefulness alongside conventional ML [machine learning]. And governments and defense departments will continue chasing the wild goose, both out of an interest in greater autonomy and to corruptly funnel cash to favored oligarchs and regime supporters. And the same credulous idiots and grifters that gamble on cryptocurrency will continue to do so with GenAI. And at the end of it all, maybe some of us will be lucky enough to be able to afford a decent graphics card for gaming again.

What do you think of this passel of scenarios? Do you envision one in particular as more likely than the others to occur? How might several play out simultaneously?

(bubble photo by Juha Riissanen)

We’ll start with international uses and dimensions of AI, then shift to national and subnational cases. We’ll add notes on copyright developments and conclude with some reflections.

1 AI and geopolitics

Data center expansion for AI support occurs mostly in some parts of the world. An Oxford study found the United States, China, and the European Union were very much in the lead. Here’s a New York Times visualization:

Other countries are not, as you can see from that map. As the Times sums it up, “Africa and South America have almost no A.I. computing hubs, while India has at least five and Japan at least four, according to the Oxford data. More than 150 countries have nothing.” This is another digital divide.

There are many AI bots creating political content around the world and across national boundaries. The Ottawa Citizen has an interesting report on one, crafted by a local man and aimed at American politics. One of his X/Twitter accounts focuses on New York City (note that its X/Twitter profile includes the word “automated”). It’s hard to determine if he’s a partisan or a digital sword for hire:

Singh, who moved to Ottawa in 2019 from Minnesota, claims he is not a propagandist, but a “technologist” who wants to show how AI can be leveraged for public relations. He said he would be open to working for campaigns across the political spectrum.

He added that he “does not discriminate” and that “the technology can help anyone.” But his known accounts have targeted progressive politicians and sought to aid conservative ones so far.

(I tried to find his other X/Twitter account, but could not, unless this is it. DOGEai_gov claims to be an “Autonomous AI uncovering waste & inefficiencies in government spending & policy,” but based in Washington DC. Maybe?)

The Chinese government is increasingly using AI to analyze intelligence content and to generate information operations worldwide, according to some American officials speaking to the New York Times. One sketch of what might be going on:

A new technology can track public debates of interest to the Chinese government, offering the ability to monitor individuals and their arguments as well as broader public sentiment. The technology also has the promise of mass-producing propaganda that can counter shifts in public opinion at home and overseas.

Along those new Cold War lines, Anthropic announced it would not sell its products to any firm with a majority Chinese ownership. Similarly, consulting giant McKinsey has cut back on its AI work in China, even though such contracts must likely be financially rewarding. The Financial Times speculates this may be a response to Trump administration pressure, trying to either decouple American businesses from China or pressure that nation to come to more favorable terms in an upcoming agreement.

Turning to Europe, the European Union appears to be considering regulations which would block American AI companies from accessing the continent’s financial data.

2: AI at the national, provincial, or state level

India is building up its AI capacity, inspired by China’s AI triumph with the relatively low cost Deepseek. That Tech Review article notes a flurry of national government activity, including spending on local efforts. It also notes a language problem based on that country’s diversity of tongues. For example,

Whereas a massive amount of high-quality web data is available in English, Indian languages collectively make up less than 1% of online content. The lack of digitized, labeled, and cleaned data in languages like Bhojpuri and Kannada makes it difficult to train LLMs that understand how Indians actually speak or search.

Global tokenizers, which break text into units a model can process, also perform poorly on many Indian scripts, misinterpreting characters or skipping some altogether. As a result, even when Indian languages are included in multilingual models, they’re often poorly understood and inaccurately generated.

Great Britain is in talks with OpenAI, trying to get a deal on ChatGPT access for some or all of that nation.

Elsewhere in Europe, the Albanian government announced an AI would serve as corruption minister. It’s not a new program for that state:

Diella, which is Albanian for “Sun,” has been up and running the Albanian government since January of this year, when it launched as a virtual assistant designed to help citizens navigate e-Albania, a digital platform for accessing government services. According to reporting from CNN affiliate A2 at the time of its launch, the chatbot is built on OpenAI’s large language model and Microsoft’s Azure cloud platform. And while it was initially tasked with helping people interact with government systems, it is now being charged with monitoring for potential corruption within the government.

Diella addressed the parliament:

In the United States there’s an interesting division opening up on the American right in terms of party politics and political ideologies. While president Trump and some of his allies lean hard into supporting and deregulating AI, others are coming out against it. We can see signs of this in the divide over the power of technology firms in this administration, or the related split over immigration policy, with the pro-tech splinter wanting more (tech-) skilled immigrants, and opponents wanting to block them.

I found another version of this divide in recent statements by conservative senator Josh Hawley (Missouri). For years he’s urged policies and laws to control the power of Silicon Valley, corralling social media and even authoring (putatively) a 2021 book, The Tyranny of Big Tech. Now at a major conservative event Hawley issued a blistering attack on AI, which his official site described as “Senator Hawley Warns ‘AI Threatens the Working Man’ at NatCon5.“ Here’s the full video:

From the other side of the political spectrum Mariana Mazzucato and Fausto Gernone call for national governments to more expansively fund creators in response to the AI revolution.

OpenAI announced it would give federal workers access to ChatGPT Enterprise.

President Trump’s Truth Social platform now has search powered by Perplexity, although results are - unsurprisingly - skewed.

The state of Colorado is considering several AI laws. They include regulations requiring AI users to notify people of that usage. Another “would empower the attorney general to sue a developer or deployer that uses AI in a way that violates the Colorado Consumer Protection Act. It would also allow an individual to file a complaint against an AI developer if the system violates the state’s anti-discrimination law.” Legislators also failed to revise another AI law, delaying its implementation until 2026.

The Canadian province of Alberta is preparing to charge datacenters there a 2% hardware fee. Alberta, which is drawing a good number of centers (see map above), has been issuing a series of regulations on local AI work.

3: Copyright

A judge approved a deal between Anthropic and a group of authors, whereby the AI company would pay out $1.5 billion. Judge Alsup earlier found the business could claim fair use for some actions, but that they were liable for training software on clearly pirated versions of content. Here’s a statement by one of the authors. Here’s a site to look for infringement and to file a claim, which I did.

Researchers found they could recreate some source material through open AI tools. Llama 3.1 70B was an outlier, able to generate big swaths of the first Harry Potter book and Nineteen Eighty-Four. An Ars Technica article found the results were uneven,

giv[ing] everyone in the AI copyright debate something to latch onto. For AI industry critics, the big takeaway is that—at least for some models and some books—memorization is not a fringe phenomenon.

On the other hand, the study only found significant memorization of a few popular books. For example, the researchers found that Llama 3.1 70B only memorized 0.13 percent of Sandman Slim, a 2009 novel by author Richard Kadrey. That’s a tiny fraction of the 42 percent figure for Harry Potter.

Elsewhere in the copyright world, the Creative Commons launched a project to create a set of ways for content owners to signal what they’d like AI to do with their stuff. For example, here’s CC Credit Direct Contribution version 0.1:

Credit

You must give appropriate credit based on the method, means, and context of your use.

Direct Contribution

You must provide monetary or in-kind support to the Declaring Party for their development and maintenance of the assets, based on a good faith valuation taking into account your use of the assets and your financial means.

There are four such signals now on the official Github site: CC Credit 0.1, CC Credit Direct Contribution, CC Credit Ecosystem, and Credit Open. Here’s the CC paper explaining the effort.

Some reflections from a futurist angle

It looks like governments are increasingly using AI as weapons or tools of statecraft. These functions are spreading, from information operations to political economy. Yet AI capacity is diverging dramatically.

Other divides are continuing, starting with that stark data center split. The US-China conflict continues to deepen and ramify in general, with multiple AI implications and roles. We should expect more of this until and if Xi and Trump strike a major deal and if it addresses AI. Decoupling the two great nations is extremely difficult, as the tariff wars have shown. The EU doesn’t seem to be trying to decouple from the US per se, but that move on financial data looks like some effort to assert independence from Trump. In contrast, Britain continues its pro-poodle policy.

I also note the divide over the technology in the conservative world. Unless party discipline takes hold we may expect the schism to persist and develop, possibly intersecting with other topics such as policing, trade, or energy policy.

The picture of national and local AI governance continues to be diverse, with more efforts under way and a wide range of policies in the air. It’s important to track subnational developments in additional to the national ones.

Copyright: the Anthropic case is a hit to that company. Its seriousness depends on how many lawsuits follow. The question which results is will AI firms consider such payments as a cost of doing business, or as major obstacles to their continued development? We should expect more suits to continue working through the system and more people to file new ones.

Overall, the AI and politics/policy world is one of continued expansion, development, divides, and churn.

Over to you all. What are you seeing on these themes?

*If you’re new to this newsletter, welcome! This is one of my scanner reports, which are examples of what futurists call horizon scanning, research into the present which looks for signals of potential futures. We can use those signals to develop trend analyses, which we can use to create glimpses of possible futures. On this Substack I scan various domains where I see AI having an impact, documenting each instance. I focus on technology, of course (for example), but also scan government and politics, economics, and education, this newsletter’s ultimate focus.

It’s not all scanning here at AI and Academia! I also write other kinds of issues; check the archive for examples.

(thanks to Howard Rheingold, John Scalzi, and Jesse Walker for links)

It’s an Atlantic article whose title and subtitle are “The Question All Colleges Should Ask Themselves About AI: How far are they willing to go to limit its harms?” (ungated version). The answer is: pretty far. In it Tyler Austin Harper argues that AI is terrible for higher education, and that in response we must uproot not only AI, but also most of the digital ecosystem on campus.

Institutions of higher learning can abandon their centuries-long educational project. Or they can resist.

“The Question All Colleges Should Ask” begins by quickly summarizing some anti-LLM arguments which should be familiar to our readers: cheating, students learning less, ecological impact, copyright infringement, a blow to the creation of new knowledge, and negative impacts on labor. In sum, “[w]idespread AI use also subverts the institutional goals of colleges and universities.” In this light Harper views universities rolling out AI research and teaching programs as surrendering to the technology: “schools hav[ing] nonetheless responded to the AI crisis by waving the white flag… [I]nfusing the college experience with the technology is deeply misguided.”

In contrast, Harper approvingly notes some academic opposition to AI, such as a May 2025 article by Megan Fritts.

She called for these classrooms to ban large language models, which she described as “tools for offloading the task of genuine expression,” then went a step further, saying that their use should be shunned, “seen as a faux pas of the deeply different norms of a deeply different space.” [emphasis in original]

However, that is not enough for Harper. Here is the part which fascinates me. He would have us go further than resisting AI. Instead, the article calls for uprooting much of the academic technology infrastructure and its uses:

Get rid of Wi-Fi and return to Ethernet, which would allow schools greater control over where and when students use digital technologies. To that end, smartphones and laptops should also be banned on campus. If students want to type notes in class or papers in the library, they can use digital typewriters, which have word processing but nothing else. Work and research requiring students to use the internet or a computer can take place in designated labs.

To emphasize that this isn’t just about AI, the author sums up his broader agenda: “Nixing screens for everyone on campus, and not just those who self-select into antitech organizations, could change campus communities for the better… [A] full tech ban can benefit everyone.”

Harper explores how this de-tech move might connect with other features of campus life. For small liberal arts colleges he would like to have more students live on campus “so they can’t use AI tools at home undetected.” He would remove technological accommodations for students with disabilities and replace them with more “peer tutors, teaching assistants, and writing centers,” admitting that “[t]his support may be more expensive than a tech product.” That disability move would also reduce people misusing accommodations “for their own benefit.”

That’s Harper’s most extreme recommendation, and he doesn’t apply it to all of higher ed. As he adds, “Universities with tens of thousands of students might have trouble enforcing a campus smartphone-and-laptop ban, and might not have the capacity to require everyone to live on campus.” For them - that broad swath of higher ed - he recommends a curiously traditional approach, calling for honor codes, students writing proctored essays in class, and whatever this means: “creating a culture that prioritizes learning and creativity, and that cultivates the attention spans necessary for sustained intellectual engagement.” Plus blocking “websites such as ChatGPT from their campus networks.”

I have so many questions about how such acts of AI and digital resistance might work practically on real campuses. If digital typewriters are good (and thanks for the early 1980s flashback) because they are single purpose devices, would Kindle handhelds be appropriate companions, since they are primarily about reading? What would happen to academic libraries, who spend a great deal of resources in curating and providing access to digital materials? How would faculty and staff do their work without laptops and smartphones, via locked-down desktop computers? What happens to the large stack of campus operations software, like learning management systems and Banner? How would he fund the extra costs involved in extirpating this infrastructure and arranging for analog replacements?

Further, most of the article seems to focus on the humanities. How would things play out in the sciences, which have some significantly different digital uses? Does he think institutions will truly be able to accommodate disabled students without digital hardware and software, especially in this era of financial stress? (I am curious how he thinks staff will distinguish accurately between those with disabilities and those acting unethically.) What happens to the staff whose jobs involve working with such technologies, from educational technologists to digital librarians? What would Harper do with the enormous amount of online postsecondary classes? Will inter-library loan traffic (in print, presumably) escalate as students lose access to JSTOR? Is faxing acceptable? What happens to the academic content posted to individual and project websites, to class materials in the LMS? Will academics be able to use social media? And so on. This is without even getting started on AI.

I’m spending time on this fascinating article here for several reasons. First, I find it resonates with a scenario I developed years ago. “Retro Campus” posits colleges and universities which turn away from the digital world for various reasons. I’ve written about it and presented it various academic audiences and the responses are consistent: some tongue in cheek longing from various folks, curiosity about appealing to some chunk of the population who might want such an unplugged experience, and much dismissal as impractical. So as a futurist I’m interested in signals along this line. (Here’s one blog post. There’s also a chapter in Academia Next.)

Second, Harper’s call for uprooting technology resonates with some voices I’ve been hearing across higher ed. Some faculty, often in the humanities, have told me they find AI too much a threat to allow in their classrooms. They can’t see a way to blocking student access without removing access to the digital world more generally, so they institute device bans (laptops, phones, tablets), have students write on paper in class, and assign printed materials. Beyond the class they rarely go as far as Harper, but I’ve heard more than a few ask for colleagues to join them in these practices, for campus IT to somehow reduce access to AI, and to voice some of the many criticisms of our digital age (attention span degradation, misinformation, etc).

Yet perhaps a movement is building among academics. Spurred by their experience and considerations of AI, they would like to see their campuses become more retro. Maybe we’ll see calls for unplugging classrooms and other institutional spaces, for non-networked or otherwise locked down devices and labs, for carving out an academic space in defiance of a world increasingly reshaped by AI.

Again, I’m not referencing this article to support or criticize it. I think it describes a potential attitude within higher education, which might turn into actions.

Have any readers seen signs of this digital extirpation call?

PS: friend Jim Lai pointed out this Harvard student’s call for peers to use AI less often.

The class is in Georgetown University’s Learning, Design, and Technology graduate program. I’m team teaching it with LDT’s founding director, professor Eddie Maloney, who first designed and led the class last year. The subject is the impact of AI on higher education.

Every week students dive into one subtopic, from pedagogy to criticism, changes in politics and economics to using LLMs to craft simulations. During the semester students lead discussions and also teach the class about a particular AI technology. Each student will also tackle Georgetown’s Applied Artificial Intelligence Microcredential. Besides Eddie and myself, we are bringing in some fine speakers, including some Georgetown experts.

We also get meta in this class, as we often do in LDT. That is, we’re teaching *about* a technology while also teaching *with* it, so we ask students to think about their experience. We’ve got an AI policy and students were very attentive to it.

Almost all of the readings are online. Otherwise, we have two books scheduled: B. Mairéad Pratschke, Generative AI and Education and De Kai, Raising AI: An Essential Guide to Parenting Our Future.

Here’s the syllabus, minus dates and other logistical details:

Introduction to AI in Higher Education:
Overview of AI, its history, and current applications in academia

Delving deeper into LLMs
Guest speakers: Molly Chehak and Ella Csarno.
Readings:

1. AI Tools in Society & Cognitive Offloading

2. MIT study, Your Brain on ChatGPT. (overview, since the study is 120 pages)

3. Allie K. Miller, Practical Guide to Prompting ChatGPT 5 Differently

Macro Impacts: Economics, Culture, Politics

A broad look at AI’s societal effects—on labor, governance, and policy—with a focus on emerging regulatory frameworks and debates about automation and democracy.

Readings:

• Raluca Csernatoni, “Can Democracy Survive the Disruptive Power of AI?”

• Mauro Cazzaniga, Florence Jaumotte, et al, “Gen-AI: Artificial Intelligence and the Future of Work”

• EU AI Act

Optional: Daron Acemoglu, “The Simple Macroeconomics of AI”

Institutional Responses

This week, we will also examine how colleges and universities are responding structurally to the rise of AI, from policy and pedagogy to strategic planning and public communication.

Reading:

• The new Inside Higher Education student survey
  (click through to data!)

How Colleges and Universities Are Grappling with AI

We consider AI’s influence on teaching and learning through institutional case studies and applied practices, with guest insights on faculty development and student experience.

Guest speaker: Eddie Watson on the AAC&U Institute on AI, Pedagogy, and the Curriculum.

Reading:

• Lee Rainie and C. Edward Watson, Leading Through Disruption: Higher Education Executives Assess AI’s Impacts on Teaching and Learning

• Elon University, Student Guide to Artificial Intelligence.

Pedagogy, Conversations, and Anthropomorphism

Through simulations and classroom case studies, we examine the pedagogical potential and ethical complications of human-AI interaction, academic integrity, and AI as a “conversational partner.”

Readings:

• B. Mairéad Pratschke, Generative AI and Education, chapters 1-4

• Bryan Alexander, “Generative AI as simulation runner”

AI and Ethics/Humanities

This session explores ethical and philosophical questions around AI development and deployment, drawing on work in the humanities, global ethics, and human-centered design.

Guest speaker: De Kai

Readings: selections from Raising AI: An Essential Guide to Parenting Our Future (TBD)

Critiques of AI

We engage critical perspectives on AI, focusing on algorithmic bias, epistemology, and the political economy of data, while challenging dominant narratives of inevitability and neutrality.

Readings:

• Bender, Gebru et all, “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?

• Kirschenbaum and Raley, “AI May Ruin the University as We Know It”

• Helen Beetham, “Artificial intelligence is the opposite of education”

Human-AI Learning

This week considers how humans and AI collaborate for learning, and what this partnership means for workforce development, education, and a sense of lifelong fulfillment.

Guest Speaker: Dewey Murdick

Readings: TBD

Agentic Possibilities

A close look at emerging AI systems and agents, with attention to autonomy, instructional design, and how educational tools are integrating generative AI features.

Reading

• Pratschke, Generative AI and Education, chapters 5-6

Future Possibilities

We explore visions for the future of AI in higher education, including utopian and dystopian framings, and ask how ethical leadership and equity might shape what comes next.

Readings:

• Maya Georgieva and John Stuart, “Ethics Is the Edge: The Future of AI in Higher Education”

• Big claims for AI: Vinod Khosla, “AI: Dystopia or Utopia?”

One week with topic reserved for emerging issues

Topic, materials, and exercises to be determined by instructors and students.

Student final presentations

I’m very excited to be teaching it.

On a meta level, I think this is the first time I’ve posted about my teaching practice. I do a bunch of this on my blog, but as I’m teaching more about AI perhaps I should post here. Should I do more? And are you teaching AI this year?