Google’s multi-agent AI teams solve decade-long medical puzzles in just 2 days

Google DeepMind has developed multi-agent AI systems that enable multiple artificial intelligence agents to collaborate, debate, and refine ideas together for complex problem-solving in medicine and scientific research. The breakthrough represents a significant leap beyond single AI agents, with teams of AI entities working together to accelerate medical discoveries that traditionally take human researchers years to complete.

The big picture: This collaborative AI approach mirrors human teamwork dynamics, where AI agents assume different roles—including “boss” and “team members”—to tackle challenging scientific problems through natural language discussions that can continue for days or weeks.

How it works: The multi-agent system builds on self-play principles used in game-playing AI, but instead of competing, the agents collaborate through structured interaction.

• Multiple AI agents with different specializations, training emphases, and reasoning skills work together on complex tasks.
• The agents engage in natural language critiquing, refining, and improving their collective ideas over extended periods.
• Google’s Gemini 2.5 “Deep Think” capability orchestrates these internal agents to develop comprehensive results for human users.

Remarkable speed advantage: The AI team approach dramatically outpaces human research timelines, as demonstrated in a collaboration with Imperial College London, a prominent British research university.

• A complex antimicrobial resistance puzzle that took human researchers “about a decade of painstaking, brilliant research” was solved by the AI co-scientist team in just two days.
• The AI system successfully “recapitulated their entire decade of discovery” in this compressed timeframe.

Real-world medical applications: The technology is already generating concrete results across multiple areas of medical research and clinical care.

• AI has helped identify new drug-repurposing candidates for acute myeloid leukemia.
• Stanford researchers discovered treatments for reversing liver damage using the system.
• The technology has assisted in designing more efficient proteins for cellular rejuvenation.
• An AI clinician called “Amy” generated better diagnoses and care plans compared to human doctors while demonstrating superior empathy.

What they’re saying: Vivek Natarajan of DeepMind, whose father died from Parkinson’s disease, emphasized the collaborative nature of this technology rather than replacement of human expertise.

• “This isn’t about replacement. This is about partnership,” Natarajan explained during a recent TED talk.
• “We really need to be able to teach the AI system to think, specifically to do the kind of slow, deliberate, system-to-style thinking that is a hallmark of real science and scientists.”
• “The real promise of such AI systems is to radically democratize healthcare access and health plan to billions.”

Why this matters: The multi-agent approach could fundamentally transform healthcare delivery and medical research accessibility worldwide.

• Natarajan envisions “having a doctor in the pocket of everyone, everywhere,” particularly benefiting remote villages in India and rural African communities.
• The technology promises to deliver “healthcare at zero marginal cost to everyone, everywhere” within a decade.
• The dual capability offers both democratized healthcare access through AI co-physicians and accelerated biomedical discovery through AI co-scientists.