Medical AI Needs More Than Accuracy to Work at the Bedside
Jean and Abgrall say medical AI can improve diagnosis and prevention, but workflow fit, training and transparent patient disclosure matter as much as model performance.

Why it matters
- Jean and Abgrall report that 55% of professionals using AI inform patients, while 80% of patients want clearer benefits-and-risks information.
- A 2013 study cited by Jean and Abgrall found 25–46% of women skipping breast-cancer screening cited compression pain.
- Clinicians rank compatibility with existing practices after a clear regulatory framework.
- Medical AI applications include disease-risk estimation, personalized monitoring and treatment, drug development and hospital operations.
- Doctors need AI training during medical school and throughout their careers, according to Jean and Abgrall.
By Elena Vasquez
Aurélie Jean and Sébastien Abgrall report that only 55% of professionals using AI inform patients, even though 80% of patients want clearer information about its benefits and risks.
Jean and Abgrall describe AI’s rapid spread through almost every industry and its growing precision. In medicine, they say the technology already helps patients manage their care and gives clinicians information that was once difficult, if not impossible, to obtain.
That promise will fade if doctors and patients are reluctant or reject the tools. The authors argue that comfort, trust and compatibility must shape design from the start, not after deployment.
They write: “Patient comfort is a measure of public-health effectiveness.”
What makes medical AI useful in health care?
The authors describe AI as a set of methods for modeling and simulating phenomena, understanding underlying mechanisms, answering questions and solving problems.
In health care, Jean and Abgrall list improved precision in examinations and diagnoses, disease-risk estimates, personalized monitoring and treatment, accelerated drug development, and help with emergencies and daily hospital operations.
These tools already help patients take a more active role in managing their health. They also give doctors access to information that was previously difficult, if not impossible, to obtain.
The authors say this reach can move medicine beyond treatment and toward prevention. AI can support earlier risk detection, more tailored care and a broader view of a patient’s health.
When does comfort become a design requirement?
An accurate model still has little value if patients feel uneasy or doctors find it cumbersome. Jean and Abgrall treat comfort as a practical condition for effective use.
Some AI solutions require an additional examination that patients have never experienced. That procedure must respect the patient’s dignity and consider their physical and psychological well-being.
Jean and Abgrall say each procedure must meet those needs because patients who feel at ease are more likely to understand the test and consent to it. Comfort is therefore part of safety and informed consent, rather than a cosmetic extra.
They cite mammography machines as an example of design responding to discomfort. Their surfaces now sit at human body temperature, making examinations less unpleasant.
When possible, AI should work with examinations already part of routine care. A new procedure may be necessary, but the authors say changes should remain minimal.
Every additional step takes time to explain and makes the examination longer. It can also feel more daunting, especially when the patient does not know what to expect.
How should AI fit doctors’ work?
Health-care professionals place a clear regulatory framework first, then rank compatibility with existing practices among the most important factors when medical teams develop or buy AI tools.
Doctors should not have to switch between systems or spend excessive time reading and interpreting new results. Jean and Abgrall warn that otherwise, clinicians may reject or misuse the tools.
Friction can also create dependence. Over time, the authors warn, overreliance on AI could erode medical skills.
A useful tool should support clinical judgment rather than quietly replace it. That requires developers to understand how doctors work, what information they need and where an automated result could create risk.
Who should shape a medical AI tool?
Jean and Abgrall recommend involving patients and doctors from the early stages of product design. Developers need to understand the needs and constraints of the people who will use their tools.
That collaboration can improve a model’s fairness and explainability. The authors say it can also prevent algorithmic discrimination before a product reaches patients.
Before developing or buying a tool, health-care providers should define the problem it needs to solve. Doctors should help align it with medical practice, guide models toward fairer answers and set limits for different scenarios.
To perform those roles and keep up with changing technology, doctors need AI training during medical school and throughout their careers. Training should cover the tool’s limits as well as its capabilities.
What should patients receive from AI-assisted care?
Patients need to make sense of the information AI provides. A numerical result without context can leave them unable to weigh benefits, risks or alternatives with their doctors.
Access to real-time health data could support shared decision-making with medical staff. Jean and Abgrall say it could also help patients take a more active role in their own care.
Developers need regular follow-up with doctors and patients long after a tool has been introduced. Their experience can reveal problems that standard performance metrics miss.
Regulations already require certain forms of monitoring. The authors say user feedback adds a different kind of evidence, one focused on comfort, comprehension and practical use.
They call this a comfort-by-design approach. Feedback should shape development and improvement after deployment, rather than remain a one-time survey.
Why should institutions measure patient comfort?
Citing a 2013 study, Jean and Abgrall report that 25–46% of women who skip breast-cancer screening cite compression pain as the reason. That finding makes comfort part of public-health effectiveness.
They recommend tracking patient satisfaction alongside diagnostic sensitivity. Institutions could then judge whether advances in medical technology encourage more people to get screened.
Health-care institutions should consider the patient’s experience from the first appointment through receipt of the results. They should also be transparent about how they use AI.
The authors return to the disclosure gap: only 55% of professionals using AI inform patients, while 80% of patients want clearer information about its benefits and risks.
Transparency matters because trust affects whether people consent, attend follow-up care and act on results. A tool that performs well in testing can still miss patients who never feel comfortable using it.
Can medical AI expand preventive care?
With hospital budgets under pressure, AI tools could make prevention more accessible beyond traditional health-care facilities. The authors see a route to earlier detection outside conventional clinical settings.
Many companies already introduce preventive health programs for employees. Jean and Abgrall say AI could help expand those programs and broaden access to early detection at controlled costs.
Beyond early detection, AI could give doctors a more holistic picture of a patient’s health. It could correlate physiological metrics, cross-reference test results and identify links between pathological mechanisms.
The authors say this approach could help assess complex conditions that a single test cannot fully capture. It may also reveal patterns that clinicians miss when they examine one measurement or result in isolation.
What will determine medical AI’s reach?
The authors’ standard has two parts: model performance and the comfort of the patients and doctors who use the tools. The second requirement receives as much attention as the first.
They conclude: “Only by meeting both can medical AI make prevention accessible to all.”
The proposal comes from two people working across technology and care. Aurélie Jean is CEO and co-founder of Infra, an AI deep-tech startup in women’s preventive medicine. Sébastien Abgrall is director-general of the Pôle Santé Bergère in Paris.
For Jean and Abgrall, the practical test is whether patients understand and consent to AI-assisted care, and whether doctors can use it without disrupting routine practice.
Source: feed.koreatimes.co.kr
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
708 articles
Related articles
- Google DeepMind Unveils AI Co-Clinician Research Initiative
- Apple Enlists 200,000 Americans to Test Trust in Health AI
- Apple Launches Apple Health AI Study App for Up to 200,000 Users
- OpenAI Releases MentalHealthBench, an Expert-Built AI Mental Health Benchmark
- OpenEvidence Gives Clinicians Free Access Intelligence Suite