A Virtual Doctor with a Camera and Voice: How Emmy and the Chinese Pre-A System Work

Emmy and Pre-A belong to different classes of medical artificial intelligence. According to their creators’ descriptions, Emmy combines voice dialogue, video analysis, and the generation of diagnostic hypotheses. Pre-A addresses a different task: it interviews the patient in advance, collects complaints and medical history, and then passes structured information and its suggestions to the doctor.

The main question here is not whether the model can deliver a convincing medical answer. More important is understanding which part of the work it takes on, who checks the result, and how closely the trial resembles real clinical practice.

From ELIZA to a Multimodal Medical System

A program that asked medical questions and changed them depending on the user’s answers already existed in the year 1966. It was ELIZA, created by Joseph Weizenbaum to study communication between humans and machines. The program imitated conversation but did not possess medical understanding in the modern sense.

Modern language models differ from early expert systems in the flexibility of their dialogue. They continue a conversation after an answer the developer did not anticipate, clarify an unclear complaint, and connect information from different parts of the conversation. This advantage is related to language and context processing, but by itself it does not prove clinical reliability.

According to the author of the material under review, the ability to respond to unanticipated answers may become the main distinction between new models and old questionnaires and decision trees. Flexibility is useful when collecting medical history, but it also makes verification more difficult: the doctor needs to understand why the system asked a specific question and on what basis it suggested the next step.

How Emmy Works

In the material under review, Emmy is described as a virtual diagnostician with voice and video channels. According to this description, the system talks with a person and analyzes video and audio. The material also describes a participant’s interaction with the system, but the available evidence does not specify exactly what actions the participant performed in front of the camera. Information about the specific underlying model is not provided in the confirmed description, so Emmy cannot be linked to Gemini or Project Astra without an additional source.

According to the description presented, Emmy contains three coordinated agents:

• the first agent conducts the conversation;
• the second agent generates medical hypotheses based on the information obtained;
• the third agent continuously analyzes video and audio, matching its observations to the course of the consultation.

This division of roles does not mean that there are three independent doctors here. It is an architectural way of distributing tasks among components of a single system. One module supports the dialogue, another evaluates possible explanations for the symptoms, and the third monitors the video and audio signs described in the demonstration. The final conclusion still depends on the quality of the model, the data, access restrictions, and the rules that connect the agents.

A multi-agent architecture also does not guarantee independent verification. If all the agents use similar data, the same underlying model, or shared erroneous assumptions, their agreement does not prove that the diagnosis is correct. For clinical use, separate validation protocols would be needed for each component and for the system as a whole.

What the Camera Adds

In the description of Emmy, the camera expands the consultation beyond text and voice because the third agent analyzes video and matches its observations to the course of the conversation. The material also mentions the participant’s interaction with the system, but the available evidence does not disclose the specific actions performed in front of the camera. More specific capabilities, such as recognizing a cough or runny nose or assessing particular aspects of appearance, are not disclosed in the confirmed evidence, so they cannot be considered established system functions.

Multimodality can be useful when a patient cannot describe a symptom precisely. Video and audio can supplement the medical history if the system actually receives a signal suitable for analysis. But camera observation is not the same as medical measurement. Lighting, microphone quality, body position, accent, room noise, and appearance all affect what the system sees and hears.

And what happens if the patient speaks quietly, is in a poorly lit room, or shows the camera the wrong part of their body? The system may receive an incomplete signal and then express its conclusion with excessive confidence. In a medical product, such a scenario should end not with an attractive diagnosis but with a record that the data quality is low, a clarifying question, or a referral to a specialist.

Why Emmy’s Demonstration Is Not a Clinical Trial

Emmy was tested with real people, but all the participants played patients. This is a significant limitation. Such a format can test interaction, response speed, camera operation, the ability to ask questions, and the overall coherence of the dialogue. It does not show how the system behaves in the presence of actual pain, fear, an incomplete medical history, comorbidities, and ambiguous symptoms.

In a role-playing scenario, the participant knows they are portraying a patient, and the organizers control the experimental conditions. An actual patient may forget the name of a medication, conceal a symptom, misjudge how long an illness has lasted, or report several problems at once. They may also be unable to calmly carry out the system’s request in front of the camera.

It is more accurate to view Emmy as a research demonstration of a multimodal consultation rather than a proven substitute for a doctor. In the author’s view, the system already looks like a virtual diagnostician, but this is an assessment of its technological potential, not an established clinical status.

How the Pre-A Trial Was Conducted

Pre-A was tested in China on more than 2 000 patients scheduled to see approximately 100 doctors from different specialties. The participants were divided into three scenarios:

1. the patient independently completed an interview with the AI;
2. staff helped the patient interact with the system;
3. the patient visited the doctor as usual, without a preliminary AI interview.

Pre-A collected complaints and medical history, suggested diagnostic hypotheses, and recommended additional examinations. The results were passed to the doctor before the consultation. In this arrangement, the AI does not replace the specialist or make the final decision. It serves as a preliminary interviewer and a tool for preparing the visit.

Pre-A’s role makes the trial substantively different from Emmy’s demonstration. The system is not required to conduct a complete examination independently or convince the patient of a diagnosis. Its value may lie in ensuring that an important complaint is not missed, collecting the patient’s history in advance, and freeing some of the doctor’s time for analysis and communication.

What a Preliminary Interview Can Give the Doctor

Before the appointment, the doctor receives an already collected sequence of complaints, information about previous treatment, and possible risk factors. If the patient has difficulty remembering dates, medication names, or the order in which symptoms appeared, a digital interview helps organize the account by topic. The doctor begins the consultation not with an empty screen but with prepared material that can be checked and corrected.

A structured record does not become true simply because a program compiled it. The doctor must distinguish the patient’s words from Pre-A’s interpretations, verify disputed information, and notice omissions. If the system has suggested an additional examination, the specialist evaluates whether it is necessary in light of the examination, medical documentation, and current clinical decisions.

Can this approach save time? Potentially, yes, if the questions reduce repeated history-taking and do not create a new body of unverified recommendations for the doctor. If, however, the specialist has to recheck every line and the patient answers a lengthy questionnaire twice, the benefit disappears.

Risks of Automatic Hypotheses

A diagnostic hypothesis helps the doctor consider possible causes of symptoms, but it can also direct their attention in one direction. The system’s suggestion may increase the likelihood that the specialist will look for confirmation of the proposed option and pay less attention to an alternative explanation.

Additional examinations create a similar problem. If Pre-A suggests an excessively broad set of tests, the clinic faces increased workload, costs, and the number of incidental findings. If the system misses a dangerous scenario, the preliminary interview creates a false sense of completeness. In both cases, responsibility for the decision remains within the medical process rather than being shifted onto the algorithm’s wording.

A separate risk is associated with errors in the source data. An incorrectly recognized medication, a missed allergy, or the mixing of information from different patients can affect subsequent recommendations. A medical system must verify patient identification, the origin of every fact, the time it was entered, and any corrections made to the record.

Emmy and Pre-A: Comparing the Approaches

Emmy attempts to expand the medical interview itself. Pre-A changes the clinic’s workflow and moves some questions to before the doctor’s consultation. The first system is closer to a virtual diagnostician; the second is closer to a digital medical secretary with analytical functions.

Trust, Empathy, and the Boundaries of the Term “Virtual Doctor”

A consistently benevolent system may make a pleasant impression, but an empathetic tone is not the same as clinical empathy. A doctor takes context into account, sees the patient’s reaction to bad news, understands their social circumstances, and chooses when a pause is needed. A model can reproduce a supportive manner of communication, but this does not prove that it understands a person’s emotions.

According to the author of the material, it is still unclear whom patients will trust more: a person or a system that always responds calmly and kindly. For some patients, the absence of judgment will be an advantage. Others will want to see a specialist who personally takes responsibility for the decision and can explain it in light of the entire situation.

The term “virtual doctor” should be used cautiously. It describes an interface and a set of functions, but does not confirm the system’s right to diagnose independently, prescribe treatment, or make decisions about the urgency of care. For now, the boundary lies where the algorithm turns data into a medical recommendation: this result must be reviewed by a qualified specialist and within established clinical procedures.

What Emmy and Pre-A Actually Demonstrated

Emmy demonstrates the possibility of combining conversation, video analysis, sound recognition, and several specialized agents in a single consultation system. Pre-A shows another approach: AI can be integrated into the process before the appointment to gather information and pass it on to the doctor. These results are important for designing interfaces and organizing workflows, but they do not justify considering any such system a ready-made autonomous doctor.

Clinical deployment will require studies involving real patients of different ages, with different diseases, language characteristics, and connection quality. What matters is not only agreement with the diagnosis, but also missed dangerous conditions, unnecessary examinations, the doctor’s time, the clarity of recommendations, and the consequences of errors. It is also necessary to separately assess how the system behaves under uncertainty and whether it can promptly hand a case over to a human.

The overall distinction is simple: Emmy demonstrates multimodal consultation, while Pre-A enhances the history-taking stage before the visit. The first technology seeks to see and hear the patient; the second helps the doctor obtain a more complete medical history. Both could become part of the medical process, but neither of the demonstrations described eliminates the need for clinical validation and professional responsibility.