Three Carnegie Mellon University professors are researching the influence of artificial intelligence on the economy and workforce as part of Google’s inaugural AI & Economy cohort. Their projects range from how job applicants feel about interacting with AI agents in the hiring process to why certain types of work tasks are more susceptible to AI than others.
The inaugural cohort(opens in new window) is made up of eight researchers from institutions across the globe, three of them from CMU. It gives each member $100,000 in unrestricted funding, dedicated compute resources and early access to frontier AI to research the impact of AI on labor markets and organizational productivity. These projects highlight CMU researchers studying the ways in which AI changes hiring, how work is divided and ways in which its economic benefits can be measured.
What if job seekers would rather be interviewed by an AI agent than a human?
Brian Jabarian, an assistant professor of economics and technology in the Heinz College of Information Systems and Public Policy with a courtesy appointment in the Kenneth C. Griffin School of Computer Science studies the transformative social and economic impacts of AI and how institutions should respond to them. He partnered with PSG Global Solutions, a company that handles outsourced hiring needs, to conduct a field experiment with an unusually large sample examining AI agents in hiring. Jabarian’s team randomly assigned roughly 70,000 job applicants to interviews with human recruiters, an AI voice agent or, for a third group, a choice between the two. When given that choice, 78% of applicants chose the AI interviewer.
As part of his research in the AI+Economy cohort, Jabarian will ask the question: What happens if employers all rely on AI systems that make similar mistakes in the hiring process?
What do economic measures miss about AI’s value?
Avinash Collis, an associate professor of digital economy at Heinz College, studies the impacts of AI that don’t appear in typical economic statistics, such as gross domestic product (GDP). Collis believes that worker productivity — a common way researchers measure the positive impacts of AI — is only part of the picture when it comes to the net benefits of AI, especially since most people use free versions of AI tools and use them mainly for nonwork activities. He is developing new metrics to quantify the economic benefits users are reaping from AI and give the tool to policymakers so they have a more complete understanding of the costs and benefits of the technology as they weigh policy decisions.
Collis is particularly interested in understanding how costs and benefits to users change by geographic location and income levels, and how AI benefits change depending on whether users live in a wealthy or poor country.
How should labor be divided between humans and AI?
There’s plenty of discussion about AI taking jobs away from human employees. Other studies have examined what jobs and employment sectors are most vulnerable to AI. But Christophe Combemale, an assistant research professor of engineering and public policy in the College of Engineering, believes understanding the data on a more granular level can provide a clearer picture of why certain jobs — and the tasks they require — are more or less vulnerable to AI.
Combemale says his preliminary research shows that some tasks are economical to separate between a person and AI, which can enable “partial” automation. But he says other tasks are harder to separate. The cost of that separation might be due to information structure, such as a lawyer following the evolution of a business dispute over many communications, or due to physical constraints, like a manual operation by a technician that requires vision, judgment, dexterity and strength based tasks to operate simultaneously. These constraints mean that it is not enough for AI to perform one task adequately, but rather that it must perform all tasks simultaneously to achieve a productivity gain.
Combemale said he hopes the research can help guide employers reconsider the productivity of AI in specialized sectors.