Psychiatrist Maria Inês Quintana, MD, had a busy schedule in 2023: teaching, advising students, conducting academic research, and providing clinical care. A professor in the Department of Psychiatry at the Federal University of São Paulo (UNIFESP) in São Paulo, Brazil, and coordinator of the postgraduate program in mental health and psychosocial care at Unimed College in Belo Horizonte, Brazil, a Brazilian health cooperative’s education institution, she had recently established a treatment program for adolescents with personality disorders when she noticed she was losing muscle strength. A few months later, the diagnosis was confirmed: amyotrophic lateral sclerosis (ALS), a progressive neurodegenerative disease that destroys motor neurons while leaving cognitive function intact.

“When I received the diagnosis, I quickly met with my research group and my colleagues at UNIFESP to discuss what the future would hold. I handed over my postdoctoral research topic and spoke with colleagues so they could mentor my students. It wasn’t an easy task to do all of this, always with tears in my eyes,” she told Medscape’s Portuguese edition.

The disease progressed aggressively. In less than a year, Quintana completely lost her ability to move and speak, became dependent on mechanical ventilation, and had to decide whether or not to undergo a tracheostomy. “ALS took everything from me. Overnight, I was in a chair, unable to move a muscle, and had a tracheostomy. It was horrible. It’s like being alive, only you’re not,” said Quintana.

Psychiatrist Fábio Gastal, academic director at Unimed College, closely followed her journey. He was Quintana’s professor during her medical residency and remained in close contact with her over the years. “I’ve been following her illness practically since it began,” he said.

Even in the face of this new reality, Quintana continued to host students and colleagues at meetings organized in her own home and began seeking technological alternatives that would allow her to maintain part of her academic routine. “The mind continues to function in these diseases; patients are absolutely limited by their bodies, but not by their minds,” explained Gastal.

From this joint effort emerged the Brazilian initiative that today seeks to open a new front for professionals with severe motor limitations: using AI as an operational extension of intellectual activity.

Quintana began using an eye-tracking system called Tobii. Connected to a computer, the device uses infrared sensors to track eye movement, functioning as a mouse. “With Tobii, I was able to present a project I’m coordinating at UNIFESP,” she reported.

But the practical limitation remained enormous. To write, Quintana had to select letters one by one, which made writing even a simple text an extremely time-consuming task. The continuous effort caused eye strain and dryness, requiring constant breaks. The slowness then began to hinder activities that depended on continuous interaction, such as classes, academic supervision, and clinical discussions. “The major research challenge in this project is synchronicity,” said Gastal.

The interview for this article was conducted in this manner. Quintana answered the questions directly, using the eye-tracking system to type her responses in real time.

Artificial Extension

It was in an effort to overcome these limitations that ExtensIA, coordinated by Gastal, was developed. Created by the Brazilian startup WorkAI in partnership with the Unimed Foundation, a Brazilian cooperative health organization, the project — which is in its pilot phase with Quintana as its first user — combines eye tracking, a digital avatar, and multi-agent AI systems.

The central idea is not to replace the doctor but to create a digital extension capable of reducing the physical strain required for her professional communication and interaction. “As far as we’ve been able to determine, there is currently nothing similar being applied in this way,” said Eduardo Barros, CEO and co-founder of WorkAI.

To develop the system, the team gathered scientific articles, lectures, presentations, videos, photographs, and even old audio recordings sent by the doctor via messaging apps. “We had to ask friends and family to send back all the audio messages she had ever sent in the past,” said Barros.

The material was used not only to train the assistive agent’s clinical content but also to reconstruct characteristics of the psychiatrist’s communication style, including her accent, expressions, and manner of speaking. “We wanted to preserve her Rio Grande do Sul accent — the ‘bah,’ the ‘tchê,’ the playful banter, and her relaxed way of communicating,” he said. [Rio Grande do Sul is Brazil’s southernmost state, known for its distinctive regional dialect.] In addition to individual rehabilitation, the project aims to preserve highly specialized knowledge by helping the doctor recognize the avatar as a legitimate representation of her own professional identity.

Three Areas of Focus

Based on the eye-tracking system, the project was structured around three main areas of focus. The first is an academic and lecturing avatar — already in operation — capable of teaching classes, giving lectures, and making presentations using the doctor’s synthesized image and voice in Portuguese, English, and Spanish. “The content remains entirely hers. What we did was create a more efficient way to convey that knowledge,” explained Barros.

According to him, the academic version already operates asynchronously. To prepare a class, Quintana writes the content using the eye tracker, chooses one of the avatars available on the platform, sends the text to the system, and generates a video in which she appears delivering the presentation with a synthesized voice based on her own audio recordings. The feature is already being used in academic activities and scientific presentations.

The second front involves a multi-agent system focused on academic management. The goal is to assist with administrative and operational tasks that require constant manual interaction, such as student monitoring, meetings with faculty, curriculum organization, and analysis of graduate program activities. These systems would function as “digital copilots,” performing operational tasks under the psychiatrist’s supervision.

The third front, considered the most complex core of the project, involves the development of an assistive clinical agent trained using a digital repository of over 30 years of scientific output, clinical experience, lectures, and articles published by Quintana. This phase is still underway. The goal is not to create an autonomous AI for medical care but to allow the doctor to supervise clinical interactions with less need for manual typing and faster response times.

According to the developers, the system was designed to operate under the doctor’s constant supervision. In situations such as prescribing medication or making sensitive clinical decisions, the avatar does not act autonomously and relies on validation from Quintana. During interactions, the platform anticipates possible conversation paths, which the psychiatrist can accept, modify, or reject using commands triggered by eye tracking.

In practice, she must monitor the entire clinical interaction on a control panel while the avatar interacts directly with the patient. “The avatar will already be pretrained. If she doesn’t say anything, it will continue. But she can intervene, correct the direction of the conversation, request additional information, and do all of this using only her eyes,” said the CEO of WorkAI. “She’ll be backstage,” summarized Gastal.

This attempt to restore greater fluidity to communication — referred to by the team as “synchronicity” — is currently ExtensIA’s primary technical objective.

‘With AI, I Felt Human Again’

The demand for AI stemmed from a human need: to address the abrupt loss of identity. “Using AI made me feel alive again. Being able to communicate, send messages with my avatar and my voice to my children and friends was like magic,” Quintana said. “With AI, I felt human again.”

For attorney Ronaldo Fagundes, a specialist in medical and health law, the most innovative aspect of the project is not technological, but human. “This is an application of AI aimed not at replacing professionals, but at preserving the autonomy and professional identity of a physician whose cognitive function remains intact despite the motor limitations imposed by ALS,” he said.

Currently, Quintana has resumed academic supervision, teaching, and participating in scientific presentations using the avatar. She has also resumed some limited clinical interactions with previous patients, although the synchronous clinical integration phase of the AI is still under development and depends on ethical and technical validation. The system has already been used at academic events and is set to be presented at the Brazilian Congress of Psychiatry and at an international symposium on ALS in the Netherlands.

Between Accessibility and the Replacement of the Physician

The innovative nature of the proposal also raises ethical and legal questions that have yet to be fully explored in medicine. According to José Geraldo Romanello Bueno, MD, PhD, professor of civil law and health law, ExtensIA does not fit perfectly within traditional definitions of telemedicine. He explained that the initiative combines elements of telemedicine, AI, assistive communication, and what he calls a “supervised professional digital twin.”

Romanello Bueno argued that the legitimacy of the model depends on a fundamental distinction: The technology must function as a tool for accessibility and not as a substitute for the physician. “The digital twin must be an assistive extension of the person, not a full replacement,” he said.

For Fagundes, the project also differs from traditional telemedicine because it seeks to reproduce aspects of the doctor’s professional identity, including her voice, image, and the knowledge she has accumulated throughout her career.

Romanello Bueno noted that, as long as Quintana is supervising the interactions, validating clinical responses, and making the final decisions, the system can be understood as an expanded form of telemedicine. The scenario changes if the avatar begins to formulate diagnostic hypotheses or conduct consultations autonomously — a scenario that, according to him, would require its own regulatory framework.

Another aspect considered central is transparency. Both Romanello Bueno and Fagundes argued that patients and students need to clearly understand the role played by AI in each interaction. “The patient must know whether they are speaking with the doctor in real time, with an avatar that replicates her, with an automated agent based on her knowledge base, or with a team supervised by her,” stated Romanello Bueno. According to both experts, this clarity is essential to preserve patient autonomy and ensure valid informed consent.

The issue of liability also remains unresolved. In a system of this type, different parties may share legal obligations, including the supervising physician, the technology developers, and the institutions involved. According to Romanello Bueno, mechanisms for tracking and logging interactions will be essential for determining liability in the event of failures or harm related to the platform’s use.

Validation

One of the project’s greatest challenges is reducing the risk of AI “hallucinations” in a medical context. For this reason, the development includes clinical validation protocols and continuous supervision. Gastal explained that senior psychiatrists will monitor the consultations conducted by the physician with the avatar’s mediation to ensure clinical safety and the consistency of responses. They have clinical experience in the doctor’s areas of expertise and will assess whether the avatar’s behavior remains consistent with recognized care standards and with Quintana’s own clinical practice.

Validation will take place in progressive stages, beginning with simulated scenarios and selected patients, before any expansion. “Our focus is on rehabilitation. We want to restore capabilities to individuals whose minds continue to function but whose bodies no longer respond,” he said.

“In our project, I will always be behind every interaction, every assessment, or every action,” assured Quintana. “My avatar will be used solely as a means to carry out what I desire and command through my mind.”

From a regulatory standpoint, the developers state that the system is being designed in accordance with CFM Resolution No. 2454/2026 — issued by Brazil’s Federal Council of Medicine — which regulates the use of AI in medical practice in Brazil. The approach treats the avatar as a form of supervised telemedicine, with the physician retaining full responsibility for all decisions. “As long as she is alive, she will be practicing telemedicine,” said Gastal.

The technological infrastructure is already in operation for Quintana’s academic activities. For future stages that may involve other professionals or formal clinical validation, the developers state that the project will be submitted to a research ethics committee, in accordance with applicable regulatory requirements.

Valued at approximately 5 million Brazilian reais (roughly US $900,000), ExtensIA has received investments from Seguros Unimed, Unimed Campinas, and Unimed-BH and remains in the beta phase, with final testing expected to be completed by March 2027. The team expects to subsequently publish a framework to guide similar applications for other disabling diseases and across different professional fields, describing not only the technological architecture but also protocols for training, supervision, and replication of the technology.

Beyond medicine, the researchers cite potential future uses in engineering, education, law, and academic research. Quintana, however, sums up the project more simply: “Today we’re conducting a case study, and I’m serving as the test subject. Having the opportunity to participate in something so grand — something that can change the lives of people in situations similar to mine for the better — is what drives me,” she concluded.

But there are questions that the researchers themselves admit they do not yet know how to answer. “We’ll face a problem when she passes away,” said Gastal. The group has begun discussing topics ranging from digital succession and intellectual property to philosophical questions about consciousness and the permanence of human knowledge mediated by AI. “This touches on ethical, theological, philosophical, and human issues for which, honestly, we’re not yet prepared,” he said.

For Romanello Bueno, the discussion goes beyond technology and enters the realm of digital heritage. According to him, the wishes expressed by the doctor herself during her lifetime should be the primary criterion for determining what may or may not be done with the avatar in the future.

The expert considers it legitimate for a system of this type to be used in the future to preserve knowledge, memory, and educational content. However, the continuation of clinical activities after the professional’s death would represent a much more complex scenario: “The intellectual legacy can remain; the false continuity of the person cannot.”

“Preserving knowledge does not mean preserving clinical competence,” Fagundes said. According to him, academic works may retain educational value after the death of their authors, but medical practice requires continuous updating, the incorporation of new evidence, and ongoing review of treatment protocols. In this context, an avatar can preserve an intellectual legacy but cannot replace the human process of continuous learning that characterizes the practice of medicine.

The ExtensIA project is funded by Seguros Unimed, Unimed Campinas, and Unimed-BH. Gastal is the academic director of Unimed College and the project coordinator.

This story was translated from Medscape’s Portuguese edition.