Today, we are announcing Future House, a new philanthropically-funded moonshot focused on building an AI Scientist. Our 10-year mission is to build semi-autonomous AIs for scientific research, to accelerate the pace of discovery and to provide world-wide access to cutting-edge scientific, medical, and engineering expertise. We have chosen to focus on biology because we believe biology is the science most likely to advance humanity in the coming decades, through its impact on medicine, food security, and climate.
Biology research today is set to scale. New techniques allow us to test tens or hundreds of thousands of hypotheses in a single experiment. New tools allow us to design thousands of proteins computationally in parallel. However, the fundamental bottleneck in biology today is not just data or computational power, but human effort, too: no individual scientist has time to design tens of thousands of individual hypotheses, or to read the thousands of biology papers that are published each day.
At Future House, we aim to remove this effort bottleneck by building AI systems -- AI Scientists -- that can reason scientifically on their own. Our AI Scientists will augment human intelligence. In 10 years, we believe that the AI Scientists will allow science to scale both vertically, allowing every human scientist to perform 10x or 100x more experiments and analyses than they can today; and laterally, by democratizing access to science and to all disciplines that are derived from it.
Our work at Future House is informed by several core hypotheses.
Firstly, we believe that all of the key components and capabilities for an AI Scientist can be built today. Scientific knowledge is represented in language, so the advent of language models is a major enabling advance.
Secondly, we believe that it is impossible to build an AI Scientist without participation from human scientists. Games like Go, Starcraft, or Minecraft have well-defined rules and win conditions, but in science there are no rules, no rewards, and no manual. Human scientists, working on concrete science projects, are the closest thing we have to ground truth. Thus, we will operate a wet lab in house, where our researchers will pursue new inventions and discoveries assisted by our AI Scientist, to discover concretely how AI will enable biology to scale.
Finally, we believe that building an AI Scientist is a key step towards building more capable and better-aligned general intelligences. Scientific reasoning, the ability to form a model of the world and to update that model in the face of uncertainty, is an essential aspect of human cognition. Biology is the most unknown science, and is thus the perfect playground in which to determine, under conditions that are free from overfitting, whether an AI Scientist can make predictions, plan experiments, or conduct analyses at a superhuman level. At Future House, integrated teams of machine learning researchers and biology researchers will iterate rapidly on constructing AI systems that can formulate hypotheses, plan experiments, reason mechanistically about the world, and apply those skills to concrete problems in biology.
To measure our progress, we keep a “Challenge Book” with specific tasks that an AI scientist should be able to accomplish, ranging from the mundane (“design primers to clone mScarlet into a pAAV CAG GFP backbone”), to those that require complex scientific reasoning (“This cell isn’t clustering with the other cells in my dataset. Why not?”). AI systems today cannot complete even the simplest tasks in our challenge book. However, work over the past six months on systems such as ChemCrow, a language-based agent that can design and execute chemical reactions, has made it obvious that we will soon be able to build systems that can.
In pursuit of our mission, we are pioneering a new approach to scientific research. Future House is an independent, non-profit research organization, headquartered in San Francisco, with the freedom to remain laser-focused on our long-term mission even if it takes us away from shorter-term, value-generating targets like drugs or industrial catalysts. We are fiercely committed to a flat structure, team science and individual contributions. We believe that the way to enable discoveries is to enable small, integrated teams of outstanding biologists and AI researchers to iterate rapidly towards big-if-true ideas. Seniority, tenure and even specific disciplines may blur as our teams forge a path forward together. As a non-profit, we also have a unique ability to prioritize responsible use of AI, which will be critical to ensure that our AI Scientists accelerate science without sacrificing safety or empowering bad actors.
Finally, we are lucky to have an exceptional network of supporters. We are backed by Eric Schmidt, who is providing us with extraordinary freedom to pursue big ideas aligned with our mission. Andrew White, one of the primary architects of ChemCrow, is joining us as Head of Science. Tony Kulesa has worked with me on building Future House since the very beginning, and is joining our Board of Directors, along with Adam Marblestone and Tom Kalil. We would also like to call out Matt Rubashkin, who is joining us as Head of Engineering; Will Barnett, who is joining us as Head of Operations; and Sam Cox, Jakub Lála, Mike Hammerling, and Jon Laurent, who are joining as members of the technical staff. If this mission and these ideas are exciting to you, get in touch. Join our mailing list at the bottom of the page, follow us on X at @FutureHouseSF, and reach out to us at firstname.lastname@example.org.