Pro-worker artificial intelligence

I just finished reading Building Pro-Worker Artificial Intelligence by Daron Acemoglu, David Autor, and Simon Johnson, and it is one of the most thoughtful reports on AI and work that I have read in a long time.

What I appreciated most is how clearly the authors move beyond the usual “AI will replace jobs” narrative. Instead, they ask a deeper question: what kind of AI do we actually want to build? Their concept of pro-worker AI—technologies that expand human capabilities and increase the value of human expertise—is both simple and powerful.

The report also offers a very helpful framework for thinking about technological change. It distinguishes between automation that replaces workers and technologies that create new tasks, new expertise, and new opportunities for people to contribute. That distinction feels crucial at a time when the direction of AI development still remains open.

Perhaps most importantly, the report reminds us that the trajectory of AI is not inevitable. It is shaped by market incentives and policy choices. If we care about inclusive prosperity, we should be actively steering AI toward ensuring that it works with humans rather than against them.

One passage (p. 4) of the report particularly captured my attention:

[…] pro-worker AI is not the norm, and we explain the reasons why it is not. The first is incentives: leading firms perceive greater economic return to building and deploying technologies that automate expertise than those that create new tasks or workers and increase the value of skills and expertise. The second reason is ideology. Even though it is a technical and highly quantitative field, computer science—and the AI community in general—is nonetheless gripped by an ideological vision that places AGI, meaning machines that exceed all human capabilities, as its highest possible pursuit.

In other words, the authors argue that the current trajectory of AI development is not accidental. Firms tend to invest in technologies that automate expertise rather than those that expand human capabilities because automation typically offers clearer and faster economic returns. In contrast, technologies that create new tasks or enhance worker expertise often produce benefits that are more diffuse, slower to materialize, and harder for individual firms to capture.

The problem is compounded by the intellectual orientation of the AI field itself. Much of the field is organized around the pursuit of AGI—systems that outperform humans across most economically valuable activities. When the ultimate goal is machines that replace human capabilities, it becomes less surprising that many innovations are directed toward substitution rather than collaboration.

What makes this argument compelling is that it reframes the debate about AI and jobs. The issue is not simply whether AI will replace workers. The deeper question is which direction technological change will take. AI is being developed primarily to automate tasks and reduce the role of human expertise. But it could also be developed to augment workers, expand the range of what they can do, and create entirely new forms of work.

The report makes an important point here: it is unrealistic to expect individual firms to solve this problem on their own. Companies respond to market incentives, and the technologies that replace workers often generate stronger private returns than those that empower them. Yet the broader social consequences—employment opportunities, wage growth, and the distribution of economic gains—affect society as a whole.

This means that the direction of AI development is not just a technological question. It is also a policy question. If we want an AI-intensive future that works for workers, we will need policies that encourage the development of pro-worker AI—technologies that complement human expertise rather than render it obsolete.

In other words, the future of work will not be determined by AI alone. It will depend on the choices we make about how AI is developed and deployed.

Highly recommended reading for anyone interested in the future of work and AI.