People actually using AI: the latest from Gallup

How do people actually use AI? Answering this question can be challenging, as it’s much easier to find anecdotes and generalizations in the crash of hyper and counter-hype.

This is why I’m fascinated by a new Gallup poll. This one focuses on people at work, as opposed to in their private or personal lives. And the results are sobering. It also doesn’t seem to have gotten a lot of attention, so here I’ll summarize and reflect on its findings.

Now, this is a shorter post, written during travel. Meanwhile, several larger posts are progressing in the hopper. Economics is up next!

To begin with, most workers aren’t seeing their workplaces roll out AI. Only about 33% experience it:

Notice the big differences between professions. White collar jobs are nearly at one half, which health care and production are below one quarter. I was surprised that admin/clerical was so low. And turning around the headline: two thirds of workers don’t see their workplaces as using AI.

That’s what workers see businesses doing. What do workers say they themselves are actually doing with AI? Not much - or rather, very few. As the survey puts it, “Weekly use remains limited, and many employees never use AI.”

Two-thirds (again that number) say “never.” The ones that use AI daily or weekly are under 20%.

Note the curve, too: “These figures remained essentially unchanged from 2023 to 2024, suggesting that AI adoption requires leadership and training to increase usage.” That’s not a rising number.

So what is that minority using AI for? The results are interesting:

Quickly, those are individual, not group uses, generally. Collaboration and other social interactions are way down on that list. Note the top three: generating ideas (does that including creating first drafts or images?). working with data, and the classic automation of basic tasks. I’m fascinated by how learning new things is medium level and identifying problems is low, low.

Now we have a sense of how many workers are using AI and what they use it for, what do workers think of those uses? Efficiency is the leading value:

The rest of that Gallup report contains advice for AI adoption, which I’ll skip today.

What can we take away from this?

I interpret this as a caution. AI adoption is much lower than some thought, at least in the workplace, with a solid majority not using the technology. Usage levels also seem to have plateaued, rather than rising, as some of us forecasted.

This matters for education, given the argument that we should teach students AI to prepare them for an AI-using workplace. Now, if colleges and universities are explicitly preparing graduates for white collar jobs, this isn’t a bad call, since Gallup’s data finds almost one half of those jobs using AI. But not all grads head to those jobs. Note, for example, the low adoption numbers in health care.

The usage data fascinates me. AI for brainstorming is a leading use, but I don’t think we often teach people to use it that way. AI as ideation partner can be powerful, and perhaps we should do more with it. (Note that this runs the risk of anthropomorphization) Yet this use is apparently for preexisting workplace tasks, as identifying new problems is far down on the list.

One more usage point catches my eye. These look like primarily individual use cases. Collaboration and human interaction are, in contrast, rare on the ground (or office or shop floor). This strikes me as an area for AI growth. As a teacher, I support students working together to learn. As someone who supports democracy, I want people to learn how to connect productively with each other.

Why are we not seeing this with AI? Perhaps it’s an interface design question, one which emphasizes individual profiles and makes sharing awkward. Or maybe it’s a cultural issue, as some of us see generative AI as an advantage, and don’t want to share it with competitors. Alternatively, we see AI as somewhat illicit or strange, and want to use it privately. This might be a lot to overcome, but perhaps changes in interface design and use behavior can lead us to social AI.

Last point: these low usage numbers are evidence for an AI contraction or bubble, if the massive AI industrial growth assumes widespread adoption.

That’s all for this one. Let’s see if dodgy train internet lets this post reach you.