# Google Trains Clinical AI Through Simulated Residency Instead of Exam Questions

Reinforcement learning against multi-turn simulated patient encounters targets diagnosis, management, communication, documentation and safety together, a step past the single-answer benchmarks that made medical AI look solved.

- Published: 2026-08-13T06:26:09.114Z
- Canonical: https://polylog.news/ai/2026-08-13/google-trains-clinical-ai-through-simulated-residency-instea
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [Google DeepMind](https://deepmind.google/blog/putting-sign-language-ai-into-users-hands/)

Medical language models have long scored well on multiple-choice licensing questions, which is a poor stand-in for actual patient care. The work now being circulated, described by the Telegram channel AI Post as training an AI through a sim…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-08-13/google-trains-clinical-ai-through-simulated-residency-instea (subscription information: https://polylog.news/pricing).