The “AI for Science” opportunity
Including a new fellowship program for AI talent interested in public mission projects
Last November, the Trump administration launched the Genesis Mission, the Department of Energy’s moonshot program to accelerate science with AI. It has already announced hundreds of millions on projects like automated labs and AI models built for science. DOE has long been a major player in AI, particularly through its 17 national labs, which house large-scale world-class scientific infrastructure (like particle accelerators and supercomputers) that often serve as the deployers of first resort for early-stage technology.
Charles Yang knows this landscape from the inside. He helped stand up the DOE’s AI work prior to the Genesis Mission, and now works on AI for science at Renaissance Philanthropy. He also writes The Republic of Science on Substack. We asked him what’s so exciting about AI for science, what the Genesis Mission actually is, what the national labs do, and how he made the jump from machine learning engineering to government and on to philanthropy.
As we discuss with Charles, one of the major bottlenecks to AI for science work is talent. To close that gap, Horizon, in partnership with Renaissance Philanthropy, SeedAI, and Fulcrum Science, is launching the AI for Science Fellowship to bring technical AI talent to the national labs to work on Genesis Mission projects.
Applications for Horizon’s AI for Science Fellowship are open through July 31.
Remco Zwetsloot: Why are people excited about AI for science? What are some existing examples of AI transforming science, and what are the kinds of breakthroughs AI could unlock in the future?
Charles Yang: For me, AI for Science has always been one of the most exciting, unqualified goods that can emerge from transformative AI.
AlphaFold is perhaps the most well-known example of a scientific AI model that has advanced the frontier of our understanding and become a massive accelerant for biomedical research. There has similarly been tons of exciting new work applying large language models to understanding biological sequence data.
Outside of biology, there is also enormous opportunity in how AI can accelerate high-resolution weather forecasting, discover new catalysts for energy production, or even unlock control of fusion power plants.
AI will undoubtedly have an immense impact on our society in a variety of ways, and already has. There are many important problems and opportunities to tackle, but for me, AI for Science remains one of the brightest shining opportunities for AI to improve human wellbeing—from curing disease to managing climate change—and to make real progress on the hardest challenges of our era.
What role has the government played in making this field happen? And what is the Trump administration trying to do with the Genesis Mission, its big AI for science project today?
Charles Yang: While general-purpose frontier model development is mostly funded by the private sector, the government still plays an important role in ensuring the development of AI is applied towards important scientific challenges.
First, coordinating resources to generate and host scientific training data is a necessary public good for AI for Science. The protein data bank, which enabled AlphaFold, was started by Brookhaven National Lab and later funded by National Science Foundation and National Institutes of Health. Similarly, historical weather data collected by public weather agencies provided the training data for AI powered weather prediction.
More importantly, AI models will ultimately need feedback loops in the real world to evaluate predictions. This includes autonomous labs but also includes large-scale scientific infrastructure, ranging from particle colliders to national fusion test reactors, that can be used to validate AI models.
Genesis Mission is an important initiative in driving AI for Science capabilities and deployment across the national lab complex. So far, it has included both foundation model development and public compute capability building, but also integrating scientific user facility infrastructure and autonomous labs to be in-the-loop with AI agents. The endeavor of making our national lab complex fully integrated with AI systems is not a trivial one, but is certainly critical for ensuring the American public benefits from AI development through accelerated scientific research. The administration has also made clear it intends for Genesis Mission to ultimately expand beyond just DOE to include other scientific agencies as well.
What are the National Labs and what role do they play, both in the Genesis Mission specifically and in the R&D ecosystem more broadly?
Charles Yang: The Department of Energy national labs are perhaps one of the most underdiscussed institutions in our public science ecosystem. Born out of the Manhattan Project, the 17 national labs today provide large-scale scientific infrastructure and user facilities. These facilities, which range from supercomputers and particle colliders to beamlines and high-end microscopes, are enormous national investments—far too large for any one university or company to build, but made publicly available for American researchers and companies to use.
The Department of Energy national labs are perhaps one of the most underdiscussed institutions in our public science ecosystem.
The national labs also employ thousands of scientists across the country. While the National Science Foundation (NSF) funds scientists for basic, undirected research and for workforce training, the DOE national labs focus on capability building—specific research programs that require large teams and are considered strategic priorities. This can range from the nuclear weapons research the national labs conduct to their role in supporting the Human Genome Project, all of which were specific public research efforts.
Besides DOE, are there other places in government doing exciting AI for Science work right now?
Charles Yang: Certainly! The National Institute for Standards and Technology (NIST) has historically played an important role in industry coordination and standard setting. In AI for Science, they’ve been quite active in setting standards for autonomous science instruments and prototyping autonomous labs for chemistry research.
The National Science Foundation has also continued to play its role in supporting scientific workforce development through its “AI-Ready America” regional grant program to drive AI adoption across businesses and institutions. NSF has also recently launched several initiatives around scientific instrument innovation, which will play an increasingly important role in driving novel science forward as AI-driven hypotheses become more commonplace.
And of course, Congress is also a critical stakeholder in funding the development and deployment of AI for Science across the scientific ecosystem. The bipartisan American Science Acceleration Project in the Senate, supported by the Accelerate Science Now network of companies and think tanks, is an important initiative in driving congressional discussion around how science policy should evolve with AI. [Horizon is a member of the Accelerate Science Now coalition.]
On your backstory—you yourself worked at DOE before, after having been an ML engineer. What was your portfolio there, and what was it like transitioning from tech to government?
Charles Yang: Yes! I was a ML engineer at an AI hardware startup, building apps on top of custom silicon accelerators. While I loved the startup environment and even worked on a national lab funded project while I was there, I wanted to more directly work in areas where my contributions would have an outsized impact.
It was clear to me then, and still is now, that we face a deficit of technologists working on public good projects, be they in government or elsewhere in the public sector.
It was clear to me then, and still is now, that we face a deficit of technologists working on public good projects.
So I made the leap in 2023 from a San Francisco Bay Area AI startup to the Department of Energy in Washington DC! It was a whirlwind tour, where I helped deploy $10B in investment tax credits for manufacturing and critical mineral projects and helped stand up DOE’s AI policy office and coordinated across the 17 national labs on AI for Science efforts which served as the predecessor to Genesis Mission.
There are lots of stereotypes and tropes about working in government. They are all true to some degree, but what is much less commonly discussed is not only how meaningful the work is, but how much room there is to drive change in systems and institutions. There is much I could say about my time at DOE, but perhaps the most important is this: the thesis I made the leap on—that a high-agency, technical person can make a real difference inside government—turned out to be true, and then some.
After government, you ended up doing AI for Science work at Renaissance Philanthropy. How did you jump into philanthropy work, and what were the biggest similarities and differences with government?
Charles Yang: As I said earlier, I was motivated to work in places where I felt I could directly impact the public good. Philanthropy felt like a natural next step after government, and Renaissance Philanthropy, a new science and tech philanthropy whose team I already knew well, seemed like the right place to do it.
Philanthropy is quite different from government. Within a federal agency, you have a clear mandate and authorization, and a relatively legible ecosystem of actors to engage with. In philanthropy, you work with all kinds of funders, and then source and support grantees, while also trying to articulate how to create a non-market mediated public value. So the work is much more open-ended, which both allows you to think big about the problems we should be tackling but is also much more free-form in terms of structure and roadmap.
In my case, I was focused on how we could enable and accelerate autonomous lab development across materials and chemistry research. My work ranged from hosting a workshop on autonomous science instruments to organizing a whitepaper with autonomous lab researchers on the bottleneck science instrumentation posed to autonomous labs, to helping incubate new autonomous lab platforms.
This also meant I started writing much more about AI for Science. I’m also grateful for my particular experience, where I was able to explore a particular field and arrived at several theses around the role of science infrastructure and science instrumentation in advancing AI for Science that were quite different from where I began.
We’re excited to now be running this fellowship program together with Horizon, Renaissance Philanthropy, and our other partners! Why should people apply?
Charles Yang: Yes, very excited to be launching this as well!
One clear message from the national labs during my time at DOE was how talent constrained they were. It was not compute, or data, or even funding that they were most lacking in—though those are also at times in short supply—but the number one constraint lab directors conveyed was the need for high-skilled AI talent with industry experience. While the national labs bring significant scientific infrastructure and expertise, they are constrained in their ability to bring in talent with strong understanding of where the state of the art in AI is today. The AI for Science Fellowship is meant to be a direct response to this public challenge.
While the national labs bring significant scientific infrastructure and expertise, they are constrained in their ability to bring in talent with strong understanding of where the state of the art in AI is today.
I imagine our final fellow cohort will have a pretty diverse set of backgrounds and motivations, but you may be a good fit if you have experience training or developing AI models in the private sector and you:
Are tired of building ad optimization or general model infra and want to work on something more mission-oriented
Want to pivot into potentially building something in AI and deeptech, and are interested in working alongside some of our nation’s best scientists for a year to quickly upskill in a specific domain
Are similarly excited by the positive societal opportunities AI can bring to scientific domains and want to contribute to those outcomes
And of course, if you’re post-economic and looking for a change of scenery.
Big thanks to Charles for taking the time and for his work getting the AI for Science Fellowship off the ground. Apply here by July 31 if you have a technical skillset and want to use AI to help tackle some of the country’s biggest scientific priorities.
If you enjoyed this conversation and want to learn more:
Charles’ Substack, The Republic of Science, including his post reflecting on his two years at DOE
Horizon’s guides to the Department of Energy and the National Labs on emergingtechpolicy.org
The Accelerate Science Now coalition’s resources