# AI co-scientist by Google DeepMind

Canonical page: https://technooptimist.io/ai/co-scientist
Frontier: [AI](https://technooptimist.io/ai.md) · Field: AI scientists · Company: [Google DeepMind](https://technooptimist.io/companies/google-deepmind.md)
Stage: Pilot (3 of 5: Research → Proto → Pilot → Deploy → Scaling)
Updated: last update 2026-05-28, facts checked 2026-10-11
Goal: Give scientists a tireless research partner that reads the literature and proposes new, testable hypotheses, so labs spend their experiments on the most promising ideas.
Status: Published in Nature (May 2026); opening to researchers through Gemini for Science.
Next milestone: Hypothesis tool in Gemini for Science, gradual access, target May 2026

## Milestones

| Status | Milestone | Date | Slip |
| --- | --- | --- | --- |
| Done | Announced with lab-validated results | 19 Feb 2025 |   |
| Done | Trusted Tester access for research groups | Feb 2025 |   |
| Done | Peer-reviewed paper in Nature | May 2026 |   |
| Current | Hypothesis tool in Gemini for Science, gradual access | May 2026 |   |

## Upcoming events

- **May 2026**: Hypothesis tool in Gemini for Science, gradual access

## Most important updates

- **28 May 2026**: AI co-scientist published in Nature; becomes the base of Gemini for Science. Google's co-scientist paper appeared in Nature, and at I/O 2026 Google launched Gemini for Science, a set of experimental research tools whose hypothesis generator is built on co-scientist. ([our story](https://technooptimist.io/stories/ai-co-scientist-published-in-nature-becomes-the-base-of-gemini-for-01f7eb07.md) · [original source](https://research.google/blog/a-new-era-of-innovation-google-research-at-io-2026))
- **19 Feb 2025**: Google unveils an AI co-scientist whose drug ideas for leukemia held up in lab tests. Google introduced a multi-agent system on Gemini 2.0 that generates, debates and ranks research hypotheses. ([our story](https://technooptimist.io/stories/google-unveils-an-ai-co-scientist-whose-drug-ideas-for-leukemia-held-up-b6db945b.md) · [original source](https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/))

## Current obstacles

- **Unknown false-positive rate**: Published wins are selected cases; how many generated hypotheses fail in the lab has not been reported.
- **Self-scored quality**: Elo ratings come from the system judging itself, not from independent ground truth.

## Physics limits

- **Ideas need experiments**: A model can only propose; nature decides. Every hypothesis still costs lab time, which sets the real pace of discovery.
- **Bounded by what is written down**: The system reasons from published and provided knowledge; unpublished data and tacit lab know-how are invisible to it.

## How it works

### Goal: Scientist states the question

A researcher writes a goal in plain language; a supervisor agent splits the work among specialist agents.

### Generate: Many candidate hypotheses

One agent drafts ideas from the literature; another critiques them; a third pits them against each other in debates.

### Rank: Tournament

Hypotheses play pairwise matches and earn chess-style Elo ratings; the best are evolved and refined.

### Test: Humans run the experiment

Top ideas go to the scientist with reasoning attached; labs then test them, as in the leukemia and liver work.

## Spec sheet

| Spec | AI co-scientist | Provenance |
| --- | --- | --- |
| Specialised agents | 6 plus a supervisor | reported |
| Base model | Gemini (2.0 at launch) | reported |
| Idea ranking | Elo tournament between hypotheses | reported |

Provenance: reported = stated by the company; estimated = our estimate; sample = a sample figure.

## About Google DeepMind

Google DeepMind is Alphabet's AI lab, formed in 2023 from DeepMind and Google Brain. It builds the Gemini models and leads in AI for science. AlphaFold won a share of the 2024 Nobel Prize in Chemistry.

- Gemini 3 (Nov 2025) led many benchmarks. Gemini 3.5 Pro, announced in May 2026, is delayed.
- AlphaFold predicted structures for over 200 million proteins and won a share of the 2024 Chemistry Nobel.
- Google's own Ironwood TPU chip is used in-house and sold to outside labs, including Anthropic.

- Founded: 2010
- Headquarters: GB, US
- Status: subsidiary (Alphabet)
- Staff: ~6,000
- Website: https://deepmind.google
- People: Demis Hassabis (CEO and co-founder); Shane Legg (Co-founder and Chief AGI Scientist); John Jumper (AlphaFold lead; 2024 Nobel laureate)
- Partners, customers, investors: Alphabet (acquirer); Isomorphic Labs (partner); Anthropic (customer); Broadcom (partner)

Company page: https://technooptimist.io/companies/google-deepmind.md

## Related programs

More from Google DeepMind:

- [Gemini](https://technooptimist.io/ai/gemini.md) (Scaling)
- [AlphaFold](https://technooptimist.io/ai/alphafold.md) (Scaling)
- [Tensor Processing Unit (TPU)](https://technooptimist.io/computing/tpu.md) (Scaling)
- [Genie world models](https://technooptimist.io/ai/genie.md) (Pilot)
- [Veo](https://technooptimist.io/ai/veo.md) (Deploy)
- [Gemma 4](https://technooptimist.io/ai/gemma-4.md) (Scaling)
- [Antigravity](https://technooptimist.io/ai/antigravity.md) (Scaling)
- [AlphaEvolve](https://technooptimist.io/ai/alphaevolve.md) (Deploy)
- [Project Suncatcher](https://technooptimist.io/computing/project-suncatcher.md) (Pilot)
- [GNoME](https://technooptimist.io/materials/gnome.md) (Deploy)
- [Gemini Robotics](https://technooptimist.io/robotics/gemini-robotics.md) (Deploy)
- [Nano Banana](https://technooptimist.io/ai/nano-banana.md) (Scaling)
- [WeatherNext 3](https://technooptimist.io/ai/weathernext.md) (Scaling)
- [Google AI Hub Visakhapatnam](https://technooptimist.io/computing/vizag-ai-hub.md) (Proto)
- [Axion](https://technooptimist.io/computing/axion.md) (Scaling)
- [Gemini in Chrome auto browse](https://technooptimist.io/ai/chrome-auto-browse.md) (Deploy)
- [AlphaGenome](https://technooptimist.io/ai/alphagenome.md) (Deploy)
- [Lyria](https://technooptimist.io/ai/lyria.md) (Deploy)

Everything in AI: https://technooptimist.io/ai.md

---

The Techno Optimist tracks every frontier of technology, program by program: https://technooptimist.io/ (a Markdown version of any page: add .md to its URL; site map for agents: https://technooptimist.io/llms.txt).
Data API (JSON, realtime stream): https://technooptimist.io/developers. People read 5 pages free, then Reader $10/month; these Markdown pages are open to agents.