AI co-scientist
In plain wordsAn AI system built to work alongside human scientists, proposing and debating new research ideas and hypotheses that people then test in the lab.
A Google multi-agent systemmulti-agent system, n.: A setup in which several AI agents, each with its own role or point of view, work together — or compete — on a problem, passing messages and results among themselves., built on Gemini, that acts as a virtual research partner: it turns a goal stated in plain language into candidate hypotheses and research proposals and then refines them. Six specialized agents (generation, reflection, ranking, evolution, proximity and meta-review) draft ideas, check them against the literature and make them compete in tournament-style debates scored with Elo ratingsElo rating, n.: A number that ranks players — or AI models — from the results of head-to-head contests, rising when you beat a strong opponent and falling when you lose to a weak one.; giving the system more test-time computetest-time compute, n.: The computing effort an AI model spends while answering a question, as opposed to while being trained. Letting a model think longer or try many answers can make it markedly smarter. produced better-rated hypotheses. Laboratory follow-ups confirmed some of its drug-repurposing candidates and combination therapies for acute myeloid leukemia; it also pointed to new treatment targets and, on its own, reached a not-yet-published explanation of how one mechanism of antimicrobial resistance works. More loosely, the phrase names any AI system designed to collaborate with scientists rather than replace them.
The co-scientist's top-ranked hypothesis matched what our lab had found but not yet published.
Origin
Announced by Google Research and Google DeepMind on 19 February 2025 with the paper 'Towards an AI co-scientist' by Juraj Gottweis, Wei-Hung Weng, Alexander Daryin, Vivek Natarajan and colleagues (posted to arXiv on 26 February), and offered to research groups through a trusted-tester program.
Related
- AI scientistAI scientist, n.: An AI system meant to do research largely on its own: reading the literature, coming up with ideas, running experiments and writing up the results.
- multi-agent systemmulti-agent system, n.: A setup in which several AI agents, each with its own role or point of view, work together — or compete — on a problem, passing messages and results among themselves.
- AlphaEvolveAlphaEvolve, prop. n.: A Google DeepMind system that combines AI language models with an evolution-like search to invent and improve computer programs, finding new results in mathematics and computing.
- automated AI researchautomated AI research, n.: Using AI systems to do the work of AI researchers, such as running experiments and writing code for new models, so that AI helps build its own successors.
- test-time computetest-time compute, n.: The computing effort an AI model spends while answering a question, as opposed to while being trained. Letting a model think longer or try many answers can make it markedly smarter.
- deep researchdeep research, n.: An AI feature that spends minutes rather than seconds on a question — planning, running dozens of web searches, reading many sources — and then writes a long report with citations.
- AI scientistAI scientist, n.: An AI system meant to do research largely on its own: reading the literature, coming up with ideas, running experiments and writing up the results.
Sources
- Gottweis, J., Weng, W.-H., Daryin, A., … Natarajan, V. (2025). Towards an AI co-scientist. arXiv:2502.18864. arxiv.org/abs/2502.18864
Mentioned in
AI scientistAI scientist, n.: An AI system meant to do research largely on its own: reading the literature, coming up with ideas, running experiments and writing up the results., Genesis MissionGenesis Mission, prop. n.: A US government program launched in 2025 to speed up science with AI, using the national laboratories' supercomputers and data to build AI tools for research.