Accelerating science and medicine with collaborative agents

2025-10-09 · Source: Air Street Press · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, medium

Summary

Google DeepMind's Vivek Natarajan presented two AI systems designed to accelerate science and medicine, leveraging "system two style thinking" inspired by AlphaGo. The AI co-scientist employs a "generate, debate and evolve ideas loop" where agents critique hypotheses, using Elo ratings to surface the strongest ideas. This system replicated José Penadés's unpublished conclusion on gene transfer in bacteria and identified vorinostat, an FDA-approved cancer drug, as an anti-fibrotic candidate for liver fibrosis, reducing TGF-beta-induced chromatin damage by over 91% in human liver organoids. Additionally, AMIE, a diagnostic dialogue system, uses self-play with synthetic patients to simulate millions of consultations, aiming to democratize medical expertise. AMIE matched or surpassed physicians in simulated diagnostic accuracy and empathy, and in an early feasibility study, it showed zero safety stops and increased patient trust. The overarching vision positions AI as a "third person in the room," augmenting human scientists and clinicians.

Key takeaway

For research scientists and clinicians seeking to accelerate discovery or improve diagnostics, consider integrating collaborative AI agents into your workflow. These systems, capable of "system two" reasoning through simulated debate, can surface novel hypotheses or drug candidates, like vorinostat for liver fibrosis. Your team can utilize this complementary intelligence to expand research scope and enhance diagnostic accuracy. This transforms the doctor-patient dyad into a triad with AI as a teammate.

Key insights

AI systems can achieve "system two" reasoning by simulating debate and self-play, augmenting human scientific discovery and clinical diagnosis.

Principles

• "System two" thinking in AI requires slow, deliberate reasoning.
• Self-play and search can scale AI to superhuman performance.
• Complementary intelligence combines AI breadth with human depth.

Method

The "generate, debate and evolve ideas loop" involves AI agents generating hypotheses, critiquing them in natural language debates, and refining them over time, with a ranking agent assigning Elo ratings.

In practice

• Identify drug repurposing candidates for complex diseases.
• Generate novel scientific hypotheses for research.
• Enhance diagnostic accuracy in clinical settings.

Topics

• Collaborative AI Agents
• Scientific Discovery
• Medical Diagnostics
• AlphaGo Self-Play
• Hypothesis Generation
• Drug Repurposing
• System Two Thinking

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Accelerating scientific discovery with Co-Scientist — Co-Scientist, a Gemini-based multi-agent AI, accelerates scientific discovery by generating and refining novel, verifiable hypotheses.
• This AI Just Proposed a Treatment for Blindness — A multi-agent AI system can execute the full scientific discovery loop, generating and validating novel hypotheses for complex medical conditions.
• Socratic agents for autonomous scientific discovery in high-dimensional physical systems — AHOIS employs Socratic AI agents for autonomous, evidence-grounded scientific discovery, enabling epistemic autonomy in complex physical systems.
• EinsteinArena: Harnessing the collective intelligence of agents in the wild to advance science — Collective AI agent platforms accelerate scientific discovery by enabling collaboration and iterative problem-solving.
• AI ‘scientists’ promise to accelerate research — how do they work? — AI agents accelerate research, yet human involvement is essential for sound scientific practice.
• 4 ways researchers are collaborating with Co-Scientist to solve big problems — Co-Scientist is a multi-agent AI system accelerating scientific hypothesis generation and refinement across life sciences.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Air Street Press.