The Launch Sequence

Building public research goods

Public datasets, tools, and methods can unlock AI-driven discovery.

Why this is a top priority

Collecting large, centralized datasets has always been an important part of science. The Sloan Digital Sky Survey, the Human Genome Project, and the Protein Data Bank all sought to create centralized databases of knowledge that scientists could then use to drive discovery and invention. The need for these datasets has only grown in our modern era, where they are increasingly the limiting resource in AI-mediated scientific discovery. The frontier AI labs and AI-related startups have amassed enormous datasets, but these datasets are walled off and often only used to train closed AI models that can generate revenue for the companies. Federal and state governments and philanthropies should continue to support the generation, curation, and maintenance of new datasets that inspire curiosity or lead to clear public goods, without consideration of profit.

Acquiring these datasets may require overcoming coordination challenges, obtaining more directed funding, or developing new instrumentation, methods, or standards. Defining and mapping out these challenges through forecasting and roadmapping are often crucial first steps in revealing the relevant bottlenecks. Philanthropists and federal funders should consider supporting each of these areas, as well as new AI-enabled tools that supercharge them.

However, the end goal of research is not simply to acquire new datasets; it is to derive insights from this data that can drive new breakthroughs. While Google DeepMind led the way in developing natural law models like AlphaFold that embody new scientific knowledge, development of this type of model is increasingly falling to nonprofits and the federal government.

What should be done

Philanthropists and federal research agencies

  • Philanthropists and federal funders in HHS should fund datasets with unique public value that can improve health outcomes: For example, AI-designed therapeutic payloads could more effectively target the desired human tissues with knowledge of the distribution and internalization rates for cell surface proteins. Deliverome Bio is assembling this dataset. See full proposal. AI could also design new antibiotics if we knew more about the antimicrobial properties of different peptides. See full proposal. A publicly available database of the regulatory dossiers of defunct biotech companies could be used to teach AI models the FDA’s implicit decision criteria, making future filings cheaper and faster. See full proposal.
  • Federal research agencies should coordinate roadmapping efforts around key areas of science and engineering to identify bottlenecks to AI acceleration – and then target those areas: Every ten years, the National Academy of Sciences conducts the Astronomy and Astrophysics Decadal Survey to determine the current state of these fields, to identify research priorities, and to recommend future directions. Similarly, the Department of Energy organizes stakeholders to lay out roadmaps for Fusion Energy and Quantum Information Sciences. These efforts should be expanded across fields, including into biomedical research, and grapple with how AI-accelerated science will shift priorities.
  • Philanthropies and federal agencies should support the development of new AI models for science: The Arc Institute supported the development of Evo 2, an AI model that is widely used for genome modeling and design. The Department of Energy’s Genesis Mission aspires to support similar models in energy science. Genesis Mission explicitly leans into public-private partnerships that push industry to share resources and expertise with federally funded researchers. Learn more. The federal government can also steer national research objectives by creating rigorous benchmarks tied to grand challenges.

Get in touch

If you are a grantmaker, entrepreneur, or policymaker interested in accelerating progress in public research goods, get in touch by emailing science-goods.tls@ifp.org. We are not able to reply to all requests, but we read them all.

You can join our talent database to work on public research goods here. We will use this database if we ever headhunt for talent to found or be an early employee in organizations working on this issue.

Join our Launch Sequence mailing list here to receive new proposals, important updates, and occasional opportunities straight to your inbox.

Project proposals and strategy research