AI Agent - Aug 13, 2026

AMIE Video vs Text: OSCE Results and Limits

Bottom Line

The AMIE study is a three-arm simulated OSCE, not a trial showing that a video AI is better than doctors in real care. It compares AMIE Video, AMIE Text, and primary care physicians using a common set of standardized scenarios. The result supports further audio-visual clinical-AI research while leaving real-patient safety, utility, workflow, and outcome questions open.

The Three Arms

ArmInterfaceWhat it helps isolate
AMIE VideoReal-time audio and videoThe combined research system’s conversation, reasoning, and perception
AMIE TextText chatThe incremental effect of moving AMIE from text to audio-visual interaction
PCP VideoReal-time videoPerformance against 10 board-certified primary care physician comparators in the same simulated interface

Fifteen professional patient actors completed 300 consultations spanning 100 scenarios. A separate panel of 20 experienced primary care physicians evaluated consultations using general and case-specific criteria.

Which Findings Are Reasonable to State?

Within the reported setting:

  • clinical evaluators rated AMIE Video on par with or better than PCP Video across several core competency rubrics;
  • AMIE Video improved over or matched AMIE Text on multiple evaluated dimensions;
  • physical-observation and guided-examination measures were a relative strength for the video configuration;
  • patient actors preferred the video interface to text on communicative effectiveness, convenience, and feeling understood;
  • the paper reports that PCPs were preferred for rapport and partnership building in the clinician comparison.

Those statements refer to different raters, comparisons, and rubrics. Compressing them into “AMIE beat doctors” discards the study design and produces a claim the research does not justify.

Why Video vs Text Matters

Text requires a patient to translate a visible or audible sign into words. Video gives the system an opportunity to observe movement, expression, posture, breathing, voice, and guided physical maneuvers. The comparison therefore tests whether adding audio-visual perception changes performance and experience—not whether a video call can replace an in-person exam, diagnostic testing, or a clinician’s full context.

The multi-agent architecture guide explains how AMIE Video separates the fast conversation path from planning and perception.

Six Limits That Must Stay Attached to the Result

  1. Simulated participants: professional actors followed scenario packs; they were not presenting their own uncontrolled symptoms.
  2. Scenario coverage: the set emphasized signs that could be portrayed through acting and omitted presentations that could not be authentically enacted.
  3. Selected comparators: 10 PCP comparators in one protocol are not the entire range of clinicians, teams, specialties, or care settings.
  4. Rubric outcomes: consultation scores and preferences are not morbidity, recovery, missed-escalation, or long-term patient outcomes.
  5. Known failures: the sources report occasional perception and reasoning errors plus intermittent technical issues.
  6. Prototype boundary: the evaluated research configuration does not establish public access, operational controls, regulatory status, or an intended clinical use.

A Better Evidence Ladder

Future claims about real-world readiness should distinguish:

Evidence levelQuestion it can answer
Automated component testsCan a component detect or reason about a defined cue?
Simulated multi-turn testsCan the system maintain a bounded clinical conversation?
Actor-based OSCEHow does it perform on standardized, controlled encounters?
Prospective real-patient studyWhat happens with real presentations and clinical workflows?
Outcome and safety evidenceDoes intended use improve care without unacceptable harm?
Deployment controlsCan approved users operate it with privacy, oversight, monitoring, and escalation?

The August 2026 AMIE Video paper provides evidence in the first three rows. It does not complete the later ones.

Frequently Asked Questions

What did the AMIE Video study compare?

It compared AMIE Video, AMIE Text, and board-certified primary care physicians using video across standardized simulated consultations with professional patient actors.

Was the AMIE Video study conducted with real patients?

No. Fifteen trained professional patient actors enacted 100 scenarios. The sources say real-patient studies are required before conclusions about real-world clinical utility.

Did AMIE Video outperform doctors?

That is too broad. Clinical evaluators rated AMIE Video on par or better on several study rubrics, but the paper reports PCP preference for rapport and partnership building, and the controlled simulated design does not establish overall superiority in real care.

Primary Sources

See the AMIE Video research guide for the broader interpretation and access boundary.