Contact The Editor
Articles By This Author
Minutes after winning the most prestigious award in mathematics, 38-year-old University of Toronto professor Jacob Tsimerman announced that he was joining the company behind ChatGPT.
"I'll be starting a position at OpenAI in their safety division very soon. So that's what's next for me," Tsimerman said Thursday at a press conference following the awards.
The mathematician's pivot was the latest example of a star professor at their intellectual peak taking a job at a company racing to make money from artificial intelligence.
For decades, Silicon Valley has poached leading lights from fields such as computer science and physics to work on problems like building smartphone apps or cloud computing services.
More recently, AI firms facing public anxiety and political backlash over the technology's impact on society have siphoned experts from a wider range of disciplines, raiding departments of philosophy, economics and public policy to answer questions about the potential consequences of the industry's breakneck growth.
Mounting uncertainty around AI's effect on workers and labor markets has directed corporations to focus on economists.
Since May, Google and Anthropic, maker of the Claude chatbot, have hired three top economics professors focused on AI's effect on jobs from Stanford University and the universities of Chicago and Virginia.
The three men will now help the AI companies produce research designed to be read by decision-makers in government and business. Although each has an impressive track record as an independent scholar, some industry observers worry that corporate backing can distort how researchers approach their work and the solutions they propose.
Lawmakers are reluctant to act without more data on AI's effect on jobs, said Molly Kinder, a former researcher at the Brookings Institution, and CEO of a new organization focused on AI's impact on workers and the economy. "There's a fog of war right now. The public's anxious, the economists are all over the map in what they're predicting," she said.
Kinder said she is confident that Anthropic's and Google's new hires will push to make more data available. Anthropic has openly shared data on how people use Claude, she said.
But it will be a loss to the public square if the new recruits can't speak freely, Kinder said. "There isn't an endless bench of experts who are deep into this technology, who really have the public's trust and experts' trust," she said.
Alexander Imas, who recently joined Google, and Chad I. Jones and Anton Korinek, who both joined Anthropic, all previously published research exploring the potential economic benefits from highly advanced AI.
Tech companies can make offers that are difficult to refuse for top academics, whose high-prestige jobs come with moderate salaries compared to elites in other sectors.
Joining a leading AI company promises access to coveted data and immense computing resources, and a chance to work on what's billed as the world's most urgent, high-stakes problem - all for better pay and potentially also stock options in a trillion dollar company.
In many cases, working for a tech giant doesn't require cutting ties with the ivory tower. Jones, Imas, Korinek and Tsimerman are all on leave from their universities. Jones was named to a Federal Reserve task force on AI this month but serves in his capacity as an independent expert, Anthropic said.
On the other side of the bargain, companies that hire academics can get fresh intellectual firepower that retains some of the authority of an outside expert.
But Silicon Valley companies have not always allowed all internal findings to see the light of day. Over the past decade, tech employees who worked on social platforms like Facebook and YouTube have alleged or leaked evidence that research showing potential harm to users was kept private.
Stephen Casper, a professor at Harvard's Kennedy School, said even knowing that tech companies are hiring can artificially restrict research agendas and policy recommendations. "Most of the people on this floor have some potential idea in their mind about a future in which they work for Anthropic," he said from his former office at MIT.
In May, Casper wrote in a social media post that he would never work for a Big Tech company. He said the declaration was inspired by a friend who works on Capitol Hill, who claimed lawmakers prefer to talk to experts who have some industry affiliation.
Brad Carson, formerly president of the University of Tulsa and now chair of a super PAC backed by Anthropic and employees from top AI firms, said the limits on corporate research are plain. "If you go to work for Exxon, they're not going to let you write whatever you want to about climate change. They just won't," he said.
Carson said he has been sent policy papers by former academics who have been told by their new tech industry employers that they cannot be published. He declined to identify the firms.
"I don't really blame the companies for that because they're not in the business of truth-telling, but it is a loss to the broader society," said Carson.
As questions grow about the effects AI will have on workers and the economy, corporate research can find a ready audience. On Thursday, the nonpartisan nonprofit Pew Research Center reported that 49 percent of Americans say they're "very concerned" about AI's role in the economy. The same day, Google released a report on AI usage that found AI is helping workers, not replacing them, a conclusion repeated in headlines from major news outlets.
Companies ramping up their research on the effects of AI on jobs and the economy say they are doing so to help the world prepare for the future.
"Society needs more research and data so decision makers can respond to what's actually happening," Google spokesperson Michael Appel said in a statement.
Anthropic's research helps "produce better data for the world," even when the findings tell a negative story about AI, Jack Clark, head of the Anthropic Institute, the company's internal think tank, told the podcast Odd Lots in May.
OpenAI's chief economist, Aaron "Ronnie" Chatterji, who is on leave from his position at Duke University, wrote on Substack last week that by publishing quickly, his team helps policymakers and business leaders act "while the window for making decisions is still open."
Tech companies are pledging to help society research and adapt to AI's economic effects at a time when the American public and many members of Congress have become concerned about the technology's potential to wipe out jobs.
This month, Anthropic launched a new ad campaign saying it "invites hard questions" around AI's impact on jobs and governance. One 90-second commercial opens with the image of a house engulfed in flames and voice recordings of apparently real users asking questions like, "Can I trust AI?" and "If it ends up taking, like, almost all the jobs then what does it mean to work?"
Outsiders may assume that AI companies are uniquely able to forecast economic disruption, but the type of internal data that companies share is less useful as a way to predict job loss, said Pamela Mishkin, who lead OpenAI's economic research team for a time before leaving the company in May.
During her time at OpenAI, Mishkin pushed to share internal data about how people use ChatGPT. But when policymakers approached Mishkin at an international summit on AI in India in February, she had to explain that a worker's conversation with ChatGPT would only tell them part of the story.
"Usage data won't tell you whether [an outsourced] worker is about to be replaced by an AI agent. Watching a company redesign its entire workflow to eliminate that worker will," she said.
Mishkin said AI companies sometimes focus on a version of the future "where every worker gets automated one day," at the expense of looking more closely at the other ways AI will affect labor, such as reducing the quality of work or giving employers permission to restructure their workforce.
Casper, the Harvard professor who vowed not to work for Big Tech, said he would nonetheless be open to accepting funding that came from Anthropic or other AI labs indirectly, from a third party. He cited recent uncertainty around federal funding for science.
"It's just kind of hard to navigate doing AI research without ever touching something that's questionable," Casper said.
"You can't realistically try to be a fish who doesn't want to swim in water."
(COMMENT, BELOW)