IImagine what would happen if the world's response to climate change relied solely on speculative predictions by experts and CEOs, rather than on rigorous (albeit still imperfect) climate science models. Some might say, “2 degrees is coming, but the world won't change as much as we think it will.” Others might counter, “2 degrees is not coming. It's going to take a long time.”
This is more or less the world we live in with artificial intelligence, with OpenAI CEO Sam Altman saying that AI systems that can perform any task a human can are “fairly near future,” while Facebook's chief AI scientist Yann LeCun argues that human-level AI systems are “a long way off.”
Jaime Sevilla, a 28-year-old Spanish researcher, is trying to change that. It's still not clear whether and how the capabilities of cutting-edge AI systems will continue to advance so rapidly, and what impact such systems will have on society. But given how important AI is already, Sevilla says it's worth trying to predict its future with some of the rigor that characterizes climate science. “Even if AI innovation stops, this is a technology that already impacts the lives of many people,” he says. “That should be enough reason for us to take it seriously.”
read more: 4 charts showing why AI progress is unlikely to slow down
To achieve this, Sevilla founded Epoch AI, a nonprofit research organization, in 2022. The organization will study historical trends in AI and use those trends to predict how the technology will develop in the future. “We want to do for artificial intelligence what Nobel Prize winner William Nordhaus did for climate change,” he says. “He laid the foundation for rigorous, evidence-based research and thoughtful action. We want to do the same thing. We want to follow in his footsteps.”
Seville has grown Born in Torrejón de Ardoz, an industrial area on the outskirts of Madrid, his early interest in technology led him to earn a degree in Mathematics and Computer Engineering from the Complutense University of Madrid, where he unintentionally planted the first seeds of Epoch AI: He returned to high school during his freshman year to give a presentation on rationality and artificial intelligence that impressed a student in the audience named Pablo Villalobos, who went on to become Epoch AI's first volunteer employee.
In 2020, Sevilla began a PhD in artificial intelligence at the University of Aberdeen. Stuck at home by COVID-19 restrictions and feeling out of place as a sun-loving Spaniard in dreary Scotland, Sevilla had time to think hard about where AI was going. “What surprised me was that no one had systematically analysed the trends in machine learning over the last few years,” he says. “I thought, if no one else is doing it, I'll do it.”
He and Villalobos began spending their spare time poring over hundreds of academic papers and recording the amount of computing power and data used to train important AI models. Confident in the importance of the work but overwhelmed by the scale of the task, Sevilla put out a call for volunteers, who responded to the call and became the original Epoch AI team. This small group worked together to record the key inputs of every important AI model ever created. When they published their findings in early 2022, the response was overwhelmingly positive, and the paper went viral in a certain internet niche. Encouraged, Sevilla paused his doctoral studies, raised funds from philanthropic donors, and in April 2022, Epoch AI was born.
Since then, Sevilla's organization, which he directs, has grown to 13 employees spread across the globe. Team morale is maintained through a vibrant Slack culture and occasional retreats where small teams strategize and sing karaoke. Though a modest operation that only professionalized two years ago, Epoch AI's work is already being widely used by governments trying to understand the rapid advances in AI. The Dutch government's vision for generative AI cites Epoch AI's work, as does a report commissioned by the UK government that seeks to synthesize evidence on the safety of advanced AI systems.
Two of the most significant efforts to put guardrails on advanced AI models — the EU AI law and President Joe Biden's executive order on AI — set thresholds for the computing power used to train AI models above which stricter rules apply. Lennart Heim, a researcher at the nonprofit think tank RAND Corporation and a founding member of Epoch AI, who still works there, says Epoch AI's AI model database has been an invaluable resource for policymakers working on such efforts. “It's safe to say that there is no other database in the world that is as comprehensive and rigorous as this one.”
Epoch AI researchers are now aiming to go a step further by leveraging research into historical trends to predict the future impact of AI. For example, in a paper published in November 2022, Epoch AI analyzed how the amount of data fed into AI models has increased over time, estimating how much useful data is readily available to AI developers. The researchers then noted that unless AI developers come up with new ways to incorporate data into their creations, they may soon run out of data. Another study sought to predict when societal changes on a scale comparable to the Industrial Revolution would occur if AI systems were widely available, and their model estimated that there was a 50% chance that such an outcome would occur by 2033. This is just one model, and Sevilla emphasizes that personal predictions of such events by members of the Epoch AI team range from a year from now to a century into the future.
read more: When will AI outperform humans? It depends on who you ask
Such uncertainty highlights that despite Epoch AI's efforts to bring rigor to the issue, there is still a great deal of unpredictability about how AI will impact society. Sevilla hopes the organization will spark a broader effort to address the issue. “We want to motivate everyone to think more rigorously about AI, to take seriously the potential for this technology to drive dramatic change in the coming decades, and to strive to rely on real evidence and good scientific thinking when making decisions about this technology,” Sevilla says.
