The Startup Building an AI Scientist

Scores of scientists are using reasoning AI models to help them come up with hypotheses and experiments in fields like nuclear fusion, biology and pathogen detection. However, some researchers believe that startups with backgrounds in specific domains like biology or chemistry would be better equipped to build the AI models for those domains, rather than more general AI labs like OpenAI, Anthropic and Google.

One of those startups is FutureHouse, a San Francisco-based nonprofit that’s aiming to build what it calls an “AI scientist.” That phrase refers to AI software that can autonomously analyze existing research and come up with new hypotheses to test in fields like biology and chemistry.

Developing such a “scientist” is simpler than you might think, according to FutureHouse cofounder and CEO Sam Rodriques. First, an AI scientist must be able to retrieve relevant information and search existing literature related to a research question, he told me. Then, it needs to be able to analyze that information and perform data analysis. Lastly, the AI scientist must be able to generate potential hypotheses to test based on that analysis and existing research, Rodriques said.