What is pro-worker artificial intelligence (AI), and how can we build it? On February 25, The Hamilton Project at the Brookings Institution hosted a virtual event to discuss pro-worker AI, why it matters, and if and how public policy could channel advances in AI to be more pro-worker. The event coincided with the release of a new essay focused on pro-worker AI.
Aviva Aron-Dine, director of The Hamilton Project, introduced the panel discussion with Daron Acemoglu (MIT Institute Professor, NBER, and Stone Center on Inequality and Shaping the Future of Work), David Autor (MIT Economics, NBER, and Stone Center on Inequality and Shaping the Future of Work), and Simon Johnson (MIT Sloan School, NBER, and Stone Center on Inequality and Shaping the Future of Work), moderated by Natasha Sarin (Yale University).
Acemoglu, Autor, and Johnson opened with an overview of their new essay on pro-worker AI.
The authors define pro-worker technologies as those that expand human capabilities and make workers’ skills and expertise more valuable. To better understand technology’s effects on workers, Acemoglu described five types of technologies: labor-augmenting, capital-augmenting, automating, expertise-leveling, and new task-creating technologies. Only the last category, new task-creating technologies, is unambiguously pro-worker. Though AI has tremendous potential to create new tasks, Acemoglu warned that “this is not the direction that AI is going.”
“AI has tremendous potential to create new tasks. … But importantly, this is not the direction that AI is going.” –Daron Acemoglu
Next, Autor offered examples of different AI tools, both real and hypothetical, evaluating the extent to which each tool would be pro-worker. He emphasized that a pro-worker technology is collaborative with, as opposed to automating, workers: “Does a technology make human skills and expertise more useful, rather than less necessary?” These examples illustrate that pro-worker AI is technically feasible and has potential across a wide range of jobs and industries. The question then is, “Why isn’t the market producing more [pro-worker AI tools], and should it be?”
Johnson presented policy recommendations to advance pro-worker AI, including:
- Focusing on health care and education as sectors where the government can shape technological change toward pro-worker AI
- Building AI expertise in government
- Using grant-making to support pro-worker AI development
- Reexamining the tax code to reduce asymmetry
- Creating new markets and new activities around worker voice and ownership of expertise and data
- Addressing barriers created by licensure requirements
Johnson challenged views that the direction of technology can or should not be changed. “We are not currently on the pro-worker AI path,” he said. “And that’s unfortunate, it’s regrettable, it’s avoidable, and we’re working pretty hard to try and get us onto another path—a much more pro-worker path.” Getting on this path, Acemoglu added, cannot be done by government alone. “If you put all of the onus on the government, that only the government has to steer us in that direction, I don’t think that’s going to work. It needs to be a collective effort.” In particular, he highlighted the need to change the priorities of researchers within the industry.
In discussion with Sarin, the panel emphasized the importance of shifting away from the current focus on automation. Acemoglu, Autor, and Johnson argued that automation is not inevitable—in fact, automation is difficult and takes time. History reflects, too, that automation is not always beneficial for workers or even society as a whole.
The panelists also explored why the market is underinvesting in pro-worker AI. Acemoglu and Johnson pointed to path dependence, industry and research priorities, and industry concentration. Many leading AI firms are centered around developing and selling automation tools, rather than pro-worker AI. Beyond automation, the AI community is gripped by the ideological vision of artificial general intelligence. Established players often buy promising start-ups, increasing the concentration of the industry.
Johnson and Autor urged policymakers to use “all levers available” to seize the opportunities that pro-worker AI offers. “This is a great opportunity. We’ve never had an opportunity to make so much progress on some of our greatest challenges—in education, in health care, in energy generation, in dealing with disease, poverty, and climate change,” Autor concluded. “So there’s huge upside, there’s huge risk. We should be looking for those upsides, we should be attempting to mitigate the risk, and we should be steering it modestly in a direction that we think is going to be socially beneficial to the degree we can.”
This is a recap of “AI + work: Building pro-worker AI.”
