Accelerating the Possible

Lukas Althoff is a Stanford economist studying how AI is changing the way scientists work. He joined the School of Humanities & Sciences as an assistant professor of economics in 2025. His current research includes the creation of SciNet, a public database that maps scientific disciplines into their component research tasks—making it possible, for the first time, to measure where AI is accelerating discovery, where bottlenecks remain, and which researchers have the most unrealized potential to benefit. 

This year as a Stanford Impact Labs Design Fellow, he’s been working to translate that evidence into practical guidance for the funders, universities, and policymakers who shape how science gets done. Stanford Impact Labs’ Alex Carr sat down with Althoff to reflect on these efforts.

Alex Carr: You describe yourself as an economic historian with computer science training, a combination that hasn’t always been the norm. How did you end up at this intersection, and what drew you to the question of how researchers adopt new tools?

Lukas Althoff: I've always been fascinated by what new technology lets people do that they couldn't before. My mum helped me set up an email account at age 7, and one of my early obsessions was "EarthCams"—public webcams pointed at beaches, skylines, and city streets around the world. I would watch one near Times Square for hours, wondering about the strangers passing through. When my mum and I finally traveled from Germany to New York, I gave my grandma the link and called her from a telephone booth at that exact spot. She watched us wave at her from across the Atlantic.

That same curiosity pushed me toward computer science in high school and then taking computer science classes in college. Neither of my parents went to university; computer science and economics seemed like a natural path with good economic prospects. In grad school, I became increasingly excited about the long-run view on economic progress, how we can make more of it, and how everyone gets an opportunity to participate.

AI is the most exciting version of that question to me right now. First, in one working paper, my coauthor Hugo Reichardt and I show that AI tools can simplify complex work, helping lower-skilled workers compete and reducing wage inequality. Second, in SciNet, I'm studying how AI is changing scientific discovery: where it accelerates progress, where bottlenecks remain, and which researchers have the most unrealized potential to benefit.

Alex: You argue that uneven AI adoption across fields and demographics isn't just an equity issue; it has consequences for which scientific problems get solved. Can you walk us through that logic? What's at stake if the productivity gains from AI stay concentrated in a narrow set of disciplines?

Lukas: What kind of science gets done depends on who becomes a scientist. If the gains of AI acceleration stay concentrated in a narrow set of labs, we're going to leave a big chunk of discoveries on the table, narrowing the set of problems humanity will solve.

The U.S. is a leader in global R&D, which means a disproportionate share of scientific effort goes toward problems shaped in the U.S. Important problems facing the rest of the world get less attention not because they're less tractable, but because fewer people with the tools are working on them. AI has the potential to accelerate research everywhere, enabling more people to drive discovery, making all of us better off.

Alex: You built SciNet as a kind of "O*NET for Science", a database that maps scientific disciplines into their component tasks. What does that make possible that wasn't possible before? And what did it take to build it?

Lukas: SciNet lets us ask where AI tools actually accelerate science—at the level of specific tasks rather than whole fields. Hinton predicted in 2016 that AI would make radiologists obsolete. It didn't happen, because AI only automates a small slice of what a radiologist actually does. There are now more radiologists than there were then, and they're more productive.

I view science as a bundle of tasks: running experiments, designing instruments, doing literature reviews, analyzing data, communicating findings. AI affects each one differently. Economists have used task-level analysis in labor markets for years, but nobody had done it systematically for science. SciNet maps every discipline into its component research tasks, so we can measure where AI is accelerating discovery, where bottlenecks remain, and which researchers have the most unrealized potential to benefit.

Building it took scale and access. Scale in the form of open data on more than 400 million publications, plus protocols and full paper text, evaluated with AI at a level of detail no human team could match. Access in the form of being at Stanford, talking to researchers across fields about how they actually use these tools, and to AI companies thinking about what role they want to play in science.

Alex: The SIL Design Fellowship is designed to push faculty to engage stakeholders outside the university and think carefully about who their research is actually for. Was there a moment during the program that reframed how you think about the relationship between your research and the people it's meant to serve?

Lukas: The moment that reframed my thinking about impact was meeting my peers at SIL. It's been really inspiring to see the problems that are being tackled by some of the most amazing people I've met on campus. On top of that, the SIL team is really unique. Everyone is incredibly talented and driven. That combination has enabled the group to make a big step towards translating research into long-lasting impact outside of academia.

Alex: A lot of faculty are passionate about a social problem but uncertain how to connect their research expertise to real-world change. What would you tell a colleague who's considering the leap you've taken?

Lukas: I suspect that many of us overestimate the gap between our research and the impact it can have in the real world. My experience has been that people in industry and government are incredibly curious to share their experiences, learn from academics, and more open to dialogue than I expected. The hardest part is building the discipline of reaching out proactively. When you do, it tends to make the research better and the impact larger at the same time. I have learned a ton and met many new friends.