How is AI changing the world of work?
In my last post I shared some results from my scanning of AI and economics. For reasons of time I focused there on business strategies and the investment world. Today I’ll shift the scanner to another economic dimension, that of labor.
We’ll focus on AI replacing versus redefining jobs, then shift to two other workforce implications, with glances at several regions with different experiences.
(If you’re new to my Substack, welcome! This issue is an example of my environmental scanning research. Each scanner report focuses on one particular slice of the world of AI and the future: politics, education, economics, culture, etc. Beyond the scanning, I also offer analyses and scenarios.)
When does AI replace jobs? The history of the industrial revolutions gives us plenty of examples wherein businesses replaced workers with machines - swapping labor for capital, in economics-speak. The car replaced the horse, buggy, and driver; refrigeration the ice shipping business, steam shovels the human hammerer, and so on. One of the great challenges of AI is that it might replace human workers entirely, costing jobs, following that industrial age template.
Or it might not. Alternatively, AI can take over some professional functions without replacing them completely, redefining those positions. For one example of this thinking, Daron Acemoglu and Simon Johnson, who recently won the Nobel economics prize, laid out several ways for AI to improve worker’s lives, by:
Improving the productivity of workers in their current jobs.
Creating new tasks with the help of machine intelligence augmenting human capabilities.
Providing better, more usable information for human decision-making.
Building new platforms that bring together people with different skills and needs.
The United States Department of Labor went further, recently issuing guidelines for using AI to improve workers’ experience, “to center the well-being of workers in the development and deployment of AI in the workplace and to value workers as the essential resources they are”:
[North Star] Centering Worker Empowerment: Workers and their representatives, especially those from underserved communities, should be informed of and have genuine input in the design, development, testing, training, use, and oversight of AI systems for use in the workplace.
Ethically Developing AI: AI systems should be designed, developed, and trained in a way that protects workers.
Establishing AI Governance and Human Oversight: Organizations should have clear governance systems, procedures, human oversight, and evaluation processes for AI systems for use in the workplace.
Ensuring Transparency in AI Use: Employers should be transparent with workers and job seekers about the AI systems that are being used in the workplace.
Protecting Labor and Employment Rights: AI systems should not violate or undermine workers' right to organize, health and safety rights, wage and hour rights, and anti-discrimination and anti-retaliation protections.
Using AI to Enable Workers: AI systems should assist, complement, and enable workers, and improve job quality.
Supporting Workers Impacted by AI: Employers should support or upskill workers during job transitions related to AI.
Ensuring Responsible Use of Worker Data: Workers' data collected, used, or created by AI systems should be limited in scope and location, used only to support legitimate business aims, and protected and handled responsibly.
[italics and bold in original]
These guidelines won support from an important labor leader:
AFL-CIO President Liz Shuler, meanwhile, said that DOL’s “new AI best practices are an important new tool to help employers understand the value of engaging their workers to ensure that these technologies actually improve productivity, work and workers’ lives.”
Note that these are nonbinding guidelines, suggestions for businesses and nonprofits. They may also disappear if Trump win’s next month’s election.
Meanwhile, state governments could pass their own AI and work laws. Illinois, for example, has a law which kicks in in 2026.
Effective Jan. 1, 2026, the Illinois Human Rights Act will be amended to prevent employers from using AI in a discriminatory manner, including using an individual’s ZIP code as a proxy identifier for characteristics. The law also requires notice to be sent when AI is being used in processes related to recruitment, hiring, promotion, renewal of employment, selection for training or apprenticeship, discharge or conditions of employment.
So will AI replace or redefine jobs? We have recently had some further signals of AI doing the former. Some of those signals are actual changes to workforces. Tiktok laid off hundreds of human moderators, especially in Malaysia, planning to replace them with AI. In one confirmation of this prophecy Swedish finance firm Klarna laid off several hundred workers, arguing that "our investment in AI… has driven down operating expenses and improved gross profits.”
Other signals of the John Henry AI future are more speculative or are attempts to influence leaders to replace works with AI. This summer Citi released internal research suggesting massive replacement of jobs in the financial sector. In August someone leaked comments by the CEO of Amazon’s cloud computing division where he imagined AI taking over a lot of his company’s coders’ jobs:
"If you go forward 24 months from now, or some amount of time — I can't exactly predict where it is — it's possible that most developers are not coding," said Garman, who became AWS CEO in June.
The CEO went further, forecasting a definition of programmers’ jobs, fairly abstractly:
"Being a developer in 2025 may be different than what it was as a developer in 2020…”
"Coding is just kind of like the language that we talk to computers. It's not necessarily the skill in and of itself," the executive explained. "The skill in and of itself is like, how do I innovate? How do I go build something that's interesting for my end users to use?"
This means the job of a software developer will change, Garman added.
"It just means that each of us has to get more in tune with what our customers need and what the actual end thing is that we're going to try to go build because that's going to be more and more of what the work is as opposed to sitting down and actually writing code," he said.
That’s how legal software company Spellbook described the launch of its new AI assistant: “By fundamentally changing the role of the junior associate and enabling AI to take on hundreds of new legal workflows, Associate eliminates drudgery and unlocks a new level of efficiency within legal teams.” “changing the role of the junior associate” is different from “letting the firm hire fewer such associates.”
That’s also how Vialytics, a data analytics business, described its use of AI. It
contracts with city or county public works departments to equip employee vehicles with phones that record road conditions as people drive around. On the back end, an AI-powered software analyzes the video, identifying issues, cataloging them by severity, and then compiling timelines for repairs. The company has around 50 clients across the US.
“Our product helps public works departments — which often get forgotten and are largely understaffed — allocate money most effectively … and allows road workers to show up to job sites more prepared,” Andy Kozma, chief revenue officer at Vialytics, told Technical.ly.
This is using AI to improve work, in other words, rather than taking it over. Back to the Department of Labor/Acemoglu and Johnson model. Yet it’s no stretch to imagine employers using such improvements to reduce, rather than just redesign, the amount of work to be done, and hence reducing their labor force.
Perhaps neither of these options will occur at scale. A survey by Uplevel, a business intelligence company, examined computer programmers’ use of AI and found they received no benefits from working with Microsoft’s CoPilot. It looks like the software’s code was buggy enough to require humans to spend as much time debugging the results as they saved by having CoPilot generate lines.
That’s a lot resting on one study, I know, but let me place it on the table for now to enable three options, big picture ones:
Businesses, governments, and nonprofits replace human workers with AI.
Employers use AI to redefine work, keeping about the same number of workers.
AI is not effective enough to redefine work or replace workers.
We currently have at least some evidence for each of these possible futures. Which intuitively makes me think what lies ahead is a mixture of all three.
Hiring website Indeed published a report which supports my intuition. Their top-level finding was that at present AI can’t replace humans or even augment us without a lot of restructuring. But with some redesign, Indeed saw AI helping some workers: for “jobs with a high share of skills that require hands-on execution, including nursing, GenAI could help with some repetitive tasks (like documentation) and allow workers to refocus on the core skills necessary in these roles.” Similarly,
[i]n more stereotypical “office jobs,” including software development, GenAI may potentially be able to offer significant knowledge and solve modest problems, emphasizing the importance of continued upskilling and ongoing learning for human workers.
For that to occur, “meaningful changes in digitalization and working norms will need to happen first.” Otherwise, Indeed found AI best at theoretical knowledge, which points to its use in knowledge work.
Besides those three alternative pathways, we can consider how AI might change work on a different level. What roles might AI play in hiring? We know employers have been using more artificial intelligence in that key process for some time. Here’s a 2023 Forbes article describing how businesses do this: “AI screens résumés and identifies the most qualified candidates for a job, matches candidates with job openings based on their skills, experience and preferences, schedules interviews and reduces the time it takes to fill open positions.” Employers can go further:
With predictive analytics, AI determines which employees are most likely to leave a company. Employers can use this insight to course-correct and have transparent conversations about how they can alleviate the risk of voluntary turnover. It also evaluates employee performance and identifies areas for improvement.
Forbes now thinks two-thirds of employers will use AI in HR.
One more recent example concerns the restaurant chain Chipotle, which announced a hiring bot:
The “virtual team member,” built by recruitment software company Paradox, will chat with candidates in various languages, answer their questions, collect basic information, schedule interviews, and even send offer letters, according to the company.
(I freely admit to confusing that Paradox with the game company of the same name and getting some odd ideas as a result.) Perhaps these efficiencies will lead to layoffs in HR: “Chipotle anticipates the tech will reduce hiring time by 75%.”
Meanwhile, some businesses are eager to hire AI talent, continuing a years-long trend, despite recent layoffs in non-AI jobs. Perhaps the most extreme version of this is the story of several Google engineers who quit that company to start up Character.ai, only to be hired back after the successful launch of that service. A recent study found that the number of statistics jobs requiring AI skills increased by 31 times over the past decade.
One more way to consider the topic of AI and labor is to look for regional differences.
Let’s start with one American city. Seattle is apparently full of AI businesses who aren’t hiring many people. As one analyst put it, “Seattle has a higher supply of AI résumés than AI job postings.” The area is apparently a magnet for AI-skilled workers, but the local giants - Microsoft and especially Amazon - are being more careful with hires now, preferring applicants with more years of experience than the average tech. Now, Seattle is an atypical city for technology. But perhaps this is an early indicator of labor market change as more people appear with AI credentials.
Britain is having a similar yet distinct hiring issue. The Financial Times reports that while university graduates can use AI to send out more applications, but they are not necessarily well suited for jobs, leading to higher rejection rates. Much of this might be explained by Britain’s sagging economy overall.
Australia might be having yet another AI and labor experience. One report from early 2024 summarized experts and other reporters, finding some room for job replacement and augmentation. They cite McKinsey claiming 9% of workers needing to find new jobs. They reference Pearson which “estimated that almost 46 per cent of banking roles, 28 per cent of accounting positions and 36 per cent of finance bookkeeping jobs could be automated by 2027.” Professor Niusha Shafiabady, at Charles Darwin University, anticipates “clerical or secretary roles, and jobs in administration… plus positions overseeing data entry, basic accounting, bookkeeping and payroll, ticketing and cashiering, and executive secretarial services” are most susceptible to automation now.
Circling back to America, one Federal Reserve bank surveyed New York state and New Jersey businesses, and found a complex picture. Unlike the case of Australia, “[t]he most widely cited use [of AI] was for marketing or advertising, as well as for business analytics and customer service.” One quarter of service industry firms and 16% of manufacturers reported using AI. Some of the former said they’d consider using AI in the next half year, while no additional firms of the latter did. Also on service businesses, “Five percent of service firm AI-users had hired new workers to accommodate the technology, while 10 percent reported laying off workers—most of whom had at most a high school diploma or GED.” Overall, the bank found its results “suggest firms plan net hiring due to the use of AI, not net worker reductions.” Overall, “it appears that firms are finding ways to utilize existing workers through training/retraining and are planning to hire new workers to work with AI.”
Summing up, the AI and labor market picture is deeply in flux, with the potential for many local and regional variations. I think we have three possible scenarios (replacement, redesign, no effect) and that we’ll likely see versions of each play out in the short and medium term future. Local economic conditions will strongly shape how employers and workers make use of AI.
Note government actions on this score. We should expect more of this, from guidance to actual laws.
There’s more to be said on labor, including workers resisting AI, but I’d like to pause for today as this is already long. Look for the next issues.
Alphafold did 200 million 3D proteins in weeks - that are on a FREE database. This is 'centuries' of research saved. QED
Love the John Henry reference!
I am pretty close to giving up on trying to make sense of surveys about AI use. Digging in on any of them is usually quite disappointing. and most of it is less than useless.
As you say, the picture is in flux, but there is little clarity about actual AI use outside of what I can see with my own eyeballs.