What is this?
Health AI applies models to medicine: reading scans, drafting clinical notes, designing drugs and answering medical questions at expert level. Earlier detection saves lives here. A wrong model also does real harm, so the stakes run both ways.
Key tools & players
- AlphaFold / Isomorphic Labs (DeepMind) — protein structure and AI-designed drugs
- Med-Gemini (Google), medical GPT models — expert-level medical Q&A
- OpenEvidence — the fastest-growing clinical reference for doctors
- AI scribes (Abridge, Nuance DAX) — auto-writing doctor's notes
- Radiology AI: hundreds of regulator-cleared diagnostic tools
Milestones
- 2020 — AlphaFold2 solves protein folding, biology's 50-year grand challenge
- 2022-2026 — Med-PaLM passed licensing exam questions, then Google's AMIE interviewed 98 real patients in The Lancet (in our news)
- 2024 — AlphaFold3 models drug interactions; its creators win the Nobel Prize in Chemistry
- 2025 — AI scribes reach thousands of clinics; first AI-designed drugs progress toward human trials
- Jul-Sep 2026 — Harvard and Stanford launch MAST, then OpenAI opens MentalHealthBench, where the best model scored 57% (in our news)
- August 2026 — An audit of all 1,357 FDA-cleared AI medical devices finds only three tested on patient outcomes
- Sep 2026 — OpenAI links ChatGPT to Epic, letting clinicians read records covering 325 million patients
- Sep 2026 — DeepMind opened its AlphaGenome DNA map and Alibaba opened Damo Radar, which beat most radiologists (in our news)
- Oct 2026 — Utah first let an AI write an initial prescription without a doctor's signature (in our news)
- Oct 2026 — Biohub, the US government, Google and Meta commit $1.8 billion to open AI-ready cell data (in our news)
Mini glossary
- AlphaFold: AI that predicts a protein's 3D shape from its sequence
- AI scribe: assistant that listens to a visit and writes the clinical note
- Clinical decision support: AI suggesting diagnoses or treatments for doctor review