Googleの医療AI「AMIE」、リアルタイム臨床ビデオ診察機能を初の研究で実証AMIE, our research medical AI system, demonstrates real-time clinical video consultation capabilities in a first-of-its-kind study.

Google has demonstrated for the first time that AMIE, its experimental medical artificial intelligence system, can conduct real-time clinical consultations over video in simulated settings. The result matters because it pushes the system beyond text-based chat and closer to the format in which many patients actually meet clinicians today, offering an early signal of how conversational AI might one day support remote or telehealth-style care.

AMIE, short for Articulate Medical Intelligence Explorer, is a research effort from Google focused on diagnostic dialogue rather than one-shot answers. Earlier work centered on typed conversations, in which the system gathered a patient's history, asked follow-up questions, reasoned about possible causes, and proposed a differential diagnosis and next steps. Those studies were run in controlled, simulated environments and evaluated using formats modeled on the Objective Structured Clinical Examination (OSCE), a standardized method medical schools use to assess trainees through interactions with actors playing patients. The new study extends that approach into live video, where the exchange unfolds in real time rather than through asynchronous text.

Moving to real-time video introduces technical demands that text does not. A system must manage the natural rhythm of spoken or live conversation, including turn-taking, interruptions, and pacing, while keeping latency low enough to feel responsive. It also opens the door to visual information that a chat interface cannot capture, such as a patient's appearance or a visible physical sign. Google's description frames this as a demonstration that AMIE can operate in this consultation format, which is a meaningful engineering and design milestone even though the work appears to remain confined to simulated scenarios rather than encounters with real patients.

It is worth being precise about what has and has not been shown. According to the source material, the study establishes that AMIE can carry out real-time clinical video consultations in simulated settings, marking a step toward AI-assisted medical consultations. That is different from a clinically validated tool, a regulatory clearance, or a deployed product. Research demonstrations like this typically measure how the system performs against defined benchmarks and, in some cases, against human clinicians on specific conversational or diagnostic metrics, but results from simulations do not automatically transfer to the complexity, liability, and safety requirements of real-world medicine. The appropriate reading is that this is likely an incremental but notable advance within an ongoing research program.

The broader context helps explain why Google is investing here. AMIE builds on the company's earlier medical AI work, including research systems such as Med-PaLM that were evaluated on medical question answering, and it reflects a wider industry push to apply large language models and multimodal models to healthcare tasks. Prior AMIE research has also explored multimodal capabilities, such as interpreting medical images alongside conversation, so adding a real-time video channel fits a trajectory toward systems that can both talk and perceive. Google's medical models are generally associated with its Gemini family of foundation models, and the company has repeatedly emphasized that these are research investigations conducted with clinical input rather than consumer products.

Other technology and health organizations are pursuing adjacent goals, from clinical documentation assistants that transcribe and summarize visits to diagnostic support tools and AI scribes now being piloted in hospital systems. Telehealth adoption, which expanded significantly in recent years, provides a plausible setting in which conversational and video-based AI could eventually play a supporting role, for example in triage, intake, or follow-up. At the same time, medical AI faces persistent challenges that this study alone does not resolve, including the risk of confidently stated errors, gaps in handling rare or atypical presentations, questions of bias across patient populations, privacy and data governance, and the regulatory frameworks that govern software used in diagnosis or treatment.

For now, AMIE's real-time video capability is best understood as a research demonstration that widens the range of interaction formats the system can handle, not as evidence that AI is ready to replace clinicians. Whether the approach proves safe and useful in practice will depend on rigorous testing with real patients, independent evaluation, and oversight from medical and regulatory bodies, steps that typically take considerable time.