This AI Expert Advises the White House, Google, JPMorgan and Corporate America

Within the first few hours after ChatGPT was released in November 2022, Univer[1]sity of Pennsylvania busi[1]ness school professor Ethan Mollick began furiously tex[1]ting with a colleague about it. He had been obsessed with the technology before, but that night, it became clear it would change everything. A lifelong eager and early adopter, Mol[1]lick immersed himself in the technology. On social media, he shared discoveries made while experimenting with ChatGPT, including findings from his M.B.A. stu[1]dents at Wharton, who use it in his classes. His embrace of large-language models has been so early, and so deep, that he has become a go-to AI expert for policymakers and corporate leaders alike. Mollick doesn’t charge for what amounts to behind-the-scenes consulting work for companies, but he is on Zoom or a plane two to four times a week, meeting with their leaders. He has talked to groups of workers and executives at Google, Meta Platforms, OpenAI and Anthropic. He’s also spoken to leaders at big banks like JPMorgan, a slew of private-equity firms, and the White House. “There is not a single large-scale gen[1]eral purpose technology that does not have upsides and downsides, and part of my message has been that we have agency right now, to call certain things out as be[1]ing bad,” says Mollick, perched on the edge of his seat, his gaze unwavering. “And we need to be doing that.” Mollick is a motormouthed, appropri[1]ately disheveled academic. His preparation for the photographer meeting us was an $18 haircut at Supercuts. He grew up in Milwaukee and is decidedly not cut from the cloth of Silicon Valley, insisting that he has little in common with “people who take cold plunges and want to live for[1]ever.” To listen to him speak is like zooming in on a never-ending fractal—his digres[1]sions have digressions, everything is three layers deeper than expected, the parts echo the whole and yet, one must agree, are critical to understanding the bigger picture. At some point, 10 or 15 minutes after heading off on a tangent, he will loop back to the original question, recall it with perfect clarity, and continue galloping down the main course of discussion. AI experiments Mollick begins spilling ideas im[1]mediately, coruscating waves of them, one after another. Microsoft’s new co-pilot AI is “dangerous,” he says at one point, because it “auto[1]mates middle management in the worst possible way.” Popping open his laptop, he shows off the video[1]analysis powers of Google’s Gemini AI, and explains how similar tech[1]nology will allow employers a level of surveillance and control over white collar employees that has previously been the lot of blue col[1]lar workers at places like Amazon. Then he shows me what he’s built with his early access to Devin, an AI unlike any other commer[1]cially available today. Devin is an “AI software engineer” with access to the internet, and the ability to use it just like a person would. Whereas today’s AIs merely give advice, Devin takes action. At one point in his experimenta[1]tion, Mollick instructed Devin to figure out how to create an account on Reddit, then offer to answer coding questions. Not only did the AI manage this feat, but it also spontaneously began demanding $50 to $100 an hour for its work, and writing back to real humans. He took it down before it could fool anyone, but what it says about the future is clear. AIs will be navigat[1]ing the internet, and eventually our world, with as much autonomy as we’re willing to give them. Mollick is generally upbeat about the future of AI—a self-described “rational optimist.” He’s insistent that now is the time for people in every field to engage deeply with it, while they still have the power to shape a nascent general-purpose technology that will someday affect every aspect of our society. He’s an authority on how AI can be applied to education, but insis[1]tent that every field needs people like him, figuring out how AI will be used in ways peculiar to that area. Little sleep, fun house “He doesn’t sleep much,” says Lilach Mollick, Ethan’s wife and fre[1]quent collaborator. The two have worked together for seven years, partners in research as well as in raising their two children, now teenagers. Their collaboration began when Lilach quit her string of human-re[1]sources jobs at tech startups to be[1]come the director of pedagogy at Wharton Interactive, a group that builds games and simulations for use in courses. Their work includes a variety of digital teaching aides, including a six-week computer sim[1]ulation that walks students through the process of launching a startup. These days, Ethan will work un[1]til 3 a.m.–“late at night he’s super productive,” says Lilach. She picks up their collaboration when she rises, at around 5 a.m. On her lap[1]top, she’s greeted by a string of Slack messages from Ethan, elabo[1]rating on what he’s been up to while she was asleep. “It’s great, because I just know that something interesting in the morning will pop up,” she adds. The two also share a Microsoft Teams account so that they can both work with all the custom AIs they’ve created. “I feel like this moment was tai[1]lor-made for them,” says Claudine Gartenberg, who teaches manage[1]ment at Wharton. The first time Gartenberg went to their house, Ethan showed her a large, two-dimensional array of LED lights on the wall, and asked her what she thought it was. She had no idea. It turned out to be a live map of the locations of trains on the New York City subway sys[1]tem, updated with data pulled from the MTA’s public feed. “Their house is filled with all these really bizarre, cool, fun toys,” adds Gartenberg. That manic curiosity and inven[1]tiveness, channeled into building things, has been a theme through[1]out Mollick’s long career in and around tech. His first startup, eMeta, launched in 1998 and helped pioneer the paywall. He went on to pursue a Ph.D. at MIT, and in the early 2000s advised AOL on AI at a time when the subject of AI was unpopular even in academia. He also helped Darpa, the advanced re[1]search division of the U.S. Depart[1]ment of Defense, adapt first-person shooter games into simulations to help troops prepare for ambushes. In 2003, he saw the movie “Ter[1]minator 3” with Marvin Minsky, who founded MIT’s AI laboratory. After, he quipped to the living leg[1]end, “I guess you didn’t succeed, because no one’s come back in time to kill you.” In his time at Wharton, Mollick has focused on how to democratize the kind of education that you’d normally have to go to Wharton to receive, by creating teaching mate[1]rials and simulations that can be distributed on the internet—and also at least one tabletop board game. Long before the debut of Chat[1]GPT, he was tweeting papers and other interesting findings from the broad array of fields his work as an economic sociologist touches on. His newsletter on AI now has more than 140,000 subscribers. He talks to the media a lot. Lilach, whom Ethan credits as the “greatest prompt writer ever,” regularly creates elaborate scripts, or prompts, for AIs that the pair share with the companies building them—especially when those prompts don’t yield the results they would like. Big banks like JP Morgan, and a slew of private-equity firms, were among the first to come calling for Mollick’s expertise. A handful of major law firms have also re[1]quested his help, as generative AI is quickly and profoundly disrupting things like legal discovery. Before the White House issued its far-reaching executive order on AI, the President’s Intelligence Ad[1]visory Board reached out to ask, basically: what is AI? As a thank[1]you, Mollick was given Biden signa[1]ture Hershey’s Kisses