The researcher trying to see the future of AI

Theimagine if the world’s response to climate change relied solely on speculative predictions from pundits and CEOs, rather than rigorous—albeit still imperfect—models of climate science. “Two degrees of warming are coming soon, but they will change the world less than we all think,” one might say. “Two degrees of warming is just not close. That will take a long time,” another might argue.

That’s pretty much the world we’re in with artificial intelligence, with OpenAI CEO Sam Altman saying that AI systems that can do any task a human can do could be developed in the “reasonably near future,” while Yann LeCun, Facebook’s Chief AI Scientist, argues that human-level artificial intelligence systems “will take a long time.”

Jaime Sevilla, a 28-year-old Spanish researcher, is trying to change that. It is not clear whether and how the capabilities of more advanced AI systems will continue to rapidly advance, and what the effects of these systems will be on society. But given how important AI already is, it’s worth trying to bring some of the rigor that characterizes climate science to predicting the future of AI, Sevilla says. “Even if AI innovation stops, this is already a technology that will affect the lives of many people,” he says. “That should be reason enough for us to take it seriously.”

Read more: 4 charts showing why AI progress is unlikely to slow down

To do this, in 2022 Sevilla founded Epoch AI, a non-profit research organization that investigates historical trends in AI and uses those trends to help predict how the technology may develop in the future. “We want to do something similar for artificial intelligence to what William Nordhaus, the Nobel laureate, did for climate change,” he says. “He laid the groundwork for rigorous study and thoughtful action guided by evidence. And we want to do the same. We want to follow in his footsteps.”

Sevilla grew in Torrejón de Ardoz, an industrial suburb of Madrid. His early interest in technology led him to pursue degrees in mathematics and computer engineering from the Complutense University of Madrid. There, he inadvertently planted the first seeds of Epoch AI—in his first year, he returned to his high school to give a presentation about rationality and artificial intelligence, impressing Pablo Villalobos, a student at audience that would go on to be Epoch AI’s first volunteer employee.

In 2020, Sevilla began her Ph.D. in artificial intelligence at the University of Aberdeen. Trapped at home by the restrictions of COVID-19 and feeling abroad like a sun-loving Spaniard in dreary Scotland, he had time to think more seriously about where AI might go. “Surprisingly, no one was doing systematic analysis of what the trends in machine learning have been over the last few years,” he says. “I thought: well, if no one is doing it, then I better go for it.”

He and Villalobos began spending their spare hours pouring over hundreds of academic papers, documenting the amount of computing power and data used to train important AI models. Feeling confident in the importance of this work, but daunted by the magnitude of the task, Sevilla put out a call for volunteers, whose respondents became the initial Epoch AI team. Together, the small group documented the critical inputs of every significant AI model ever created. When they published their findings in early 2022, the response was overwhelmingly positive, with the paper going viral in some areas of the internet. Encouraged, Sevilla put his PhD on hold, sought funding from philanthropic donors, and in April 2022, Epoch AI was born.

Since then, the organization, where Sevilla is director, has grown to 13 employees, spread around the world. Team morale is maintained through a vibrant Slack culture and occasional retreats in which the small team strategizes and sings karaoke. It’s a modest operation that was professionalized only two years ago, but already Epoch AI’s work is widely used by governments trying to understand the rapid progress of AI. The Netherlands Government’s vision for generative artificial intelligence cites the work of Epoch AI, as well as a report commissioned by the UK government that aims to synthesize evidence on the safety of advanced AI systems.

Two of the most significant efforts to put guardrails around advanced AI models—the EU’s AI Act and President Joe Biden’s AI Executive Order—set a threshold in terms of the computing power used to train a model. AI, on which stricter rules apply. Epoch AI’s database of AI models has been an invaluable resource for policymakers undertaking such efforts, says Lennart Heim, a researcher at the nonprofit RAND Corporation, who was a member of Epoch AI’s founding group and still is. associated with the organization. “I think then it’s fair to say that there is no other database in the world that is as exhaustive and as rigorous.”

Researchers at Epoch AI now aim to go a step further by using their research on historical trends to inform predictions of AI’s future impacts. For example, in a paper published in November 2022, Epoch AI analyzed how the amount of data being fed into AI models was increasing over time and assessed how much useful data is readily available to AI developers. The researchers then noted that AI developers could soon run out of data if they didn’t find new ways to feed their creations. Another study attempts to predict when AI systems that, if widely available, will result in societal change comparable in scale to the industrial revolution—their model estimated that such an outcome is 50% likely to occur by the year 2033. This is just one model—Sevilla points out that Epoch AI team members’ personal predictions for such an event range from a year to a century away.

Read more: When can AI overtake us? It depends on who you ask

Such uncertainty underscores that despite Epoch AI’s efforts to bring rigor to the issue, a great deal of unpredictability remains about how AI will impact society. Sevilla hopes his organization will catalyze a broader effort to address the issue. “We want to motivate everyone to think more rigorously about AI—to take seriously the possibility that this technology could bring about drastic changes in the coming decades,” he says, “and to try to rely on current evidence or thinking good science when making decisions about technology.”