{shortcode-d80ef3359fb4f1aae9c8d5a176e3b7972b70ff60}

Workers in occupations most exposed to artificial intelligence can earn substantially more after retraining, according to a new working paper co-authored by Harvard Kennedy School Ph.D. candidate Karen Ni.

The paper, submitted in August with co-authors Benjamin G. Hyman, Benjamin Lahey of New York University, and Laura Pilossoph of Duke University, examines how AI reshapes occupations and whether workers in exposed fields can retrain without long-term career losses.

The study defines “AI exposures” as jobs where technology can perform a large share of human tasks. It identifies legal, computer and mathematical, and arts and entertainment occupations as the most AI-exposed. But Ni said the findings show that exposure does not necessarily translate into permanent displacement.

“In terms of retrainability, what we really aim to capture here is the capacity of a worker who is coming from an AI exposed field to be able to — through job training — gain skills to allow them to adapt and go back into AI intensive work without necessarily taking huge pay cuts or being displaced again,” she said in an interview with The Crimson.

The project began when Ni was a pre-doctoral researcher at the New York Federal Reserve Bank studying labor markets and job training. The release of large language models like ChatGPT shifted the team’s focus to AI in particular.

“We basically saw this as an opportunity to look at a very specific application of future work by looking at the disruption-slash-integration that AI has had in the labor market,” she said.

The researchers drew on national data from federal training programs, which showed a wide variety of retraining paths. Some workers pursued AI-adjacent skills such as prompt engineering, while others sought credentials in fields ranging from trucking to nursing. The breadth, Ni said, shows that retraining cannot be understood only through narrow technical skill sets.

“Instead of thinking about this as a very doom and gloom kind of thing — AI is going to come in and take all our jobs and bad things will happen — we want to look at the potential upside of this,” she said. “Could AI skills be potentially helping workers become more productive or gaining new opportunities that might not have been available to them previously?”

Ni, who is in her final year at the Kennedy School, credited HKS with shaping her development as an independent researcher and pointed to Harvard College Dean David J. Deming’s research on the future of work as influential in framing her own projects.

“I know that Dean Deming does talk a lot about just the upshot of AI and art, the reality of education today,” she said. “I know I've seen it as a teaching fellow during the last three years of my time here, and I think the reality is that AI is here to stay, whether we like it or not.”

Deming, who studies the economics of education and the labor market, has written extensively on how soft skills and social skills grow in value as technology automates routine tasks. In his inaugural convocation address earlier this year, he urged Harvard undergraduates to embrace liberal arts training as a complement to technical knowledge in an era marked by AI.

Looking ahead, Ni said she and her co-authors plan to move beyond national data toward field experiments at the state level, where they hope to isolate what kinds of retraining programs succeed. Ni added that she is also interested in studying the equity effects of AI adoption, especially for women and underrepresented groups in the workforce.

Still, for Ni, AI is only one case study in a much larger story.

“With technology changing all the time, it might not be AI that we’re talking about in 10 years,” she cautioned. “What we ultimately hope to do in this work is to not just speak to AI specifically, but towards technological change more broadly.”