Collaborating on a nationwide randomized study of AI in real-world virtual care

Google Research just announced something that moves its medical AI ambitions out of the lab and into actual patient visits. Working with Included Health, a virtual care provider, Google plans to run a nationwide randomized controlled trial testing its conversational AI system against standard clinical practice, pending IRB approval. Real patients, real conditions, real geography spread across the country. Not actors reading scripts, not retrospective chart reviews.

This is the next rung on a ladder Google has been climbing for a few years now. It started with AMIE, a diagnostic dialogue system tested against primary care physicians in simulated consultations, where it reportedly matched or beat human doctors on diagnostic accuracy and conversation quality. Then came a single-center feasibility study with Beth Israel Deaconess Medical Center, watching how often a human safety supervisor had to step in and interrupt the AI during real clinical use. Google says early signs from that study look good, though full results aren't out yet.

What's different this time is scale and rigor. A single hospital feasibility test tells you whether something is safe enough to keep studying. A nationwide randomized trial with consented participants is designed to tell you whether the thing actually works better, or at least as well, as normal care — the kind of evidence bar drugs and devices have to clear, applied here to a chatbot doing clinical reasoning. Google frames this as deliberate: each phase generates more data on safety and usefulness before the next step gets greenlit.

The AI going into this trial isn't just AMIE. Google has also been building a Personal Health Agent that chews through wearable data like sleep and activity metrics to produce coaching insights, and a