Vivek Natarajan of Google DeepMind at RAAIS 2026

Key Takeaways

• Med-PaLM 2 achieved 86.5% on MedQA, surpassing predecessor by 19 points
• AMIE’s diagnostic AI included correct diagnosis in 90% of 100 patient chats
• AI co‑scientist supported US Genesis Mission across 17 DOE labs in 2025
• DeepMind to open UK automated research lab for materials science in 2026
• Vivek Natarajan, Google DeepMind research lead, speaking at RAAIS 2026 London

Pulse Analysis

The latest wave of AI breakthroughs is no longer confined to laboratory benchmarks; it is entering the corridors of hospitals and national laboratories. Med‑PaLM 2, led by DeepMind’s Vivek Natarajan, demonstrated an 86.5% score on the MedQA dataset, eclipsing earlier models by more than 19 points and earning physician approval comparable to human clinicians. This performance leap underscores a broader trend: large language models are being fine‑tuned for domain‑specific reasoning, rigorous evaluation, and the trust standards demanded by medical licensing bodies.

Project AMIE (Articulate Medical Intelligence Explorer) pushes the envelope further by embedding conversational AI directly into primary‑care workflows. In a March 2026 feasibility study at Beth Israel Deaconess Medical Center, 100 patients interacted with AMIE via text chat, and the system’s differential diagnosis contained the eventual diagnosis in 90% of cases without triggering safety stops. The ongoing nationwide randomized trial with Included Health aims to validate these results at scale, illustrating how AI can move from single‑turn question answering to continuous, multimodal clinical reasoning that mirrors real‑world diagnostic processes.

Beyond medicine, Natarajan’s AI co‑scientist platform is reshaping scientific discovery. Built on DeepMind’s Gemini models, the multi‑agent system has already identified drug repurposing candidates for acute myeloid leukemia and new targets for liver fibrosis. Integrated into the US Genesis Mission, it now serves researchers across all 17 Department of Energy national laboratories and will be housed in DeepMind’s first automated research lab in the UK, slated for 2026. Together, these initiatives highlight a pivotal shift: AI is transitioning from knowledge organization to knowledge creation, a narrative that will dominate discussions at RAAIS 2026.