Can AI help doctors without replacing their clinical judgment?

A mental health patient arrives for an appointment, and the clinician has no visibility into what has happened between visits. Artificial intelligence, increasingly present in every area of society, is emerging as a potential ally in changing this situation. Even so, its implementation in healthcare raises questions about the balance of responsibilities with the doctor and fears of losing autonomy and clinical rigor.

However, this may not be the best way to introduce artificial intelligence into today’s healthcare context. A more useful question would be: what can AI do to help professionals do their jobs better?

In a context of overburdened healthcare services, AI can be an ally in reducing certain burdens in the care process and expanding the professional’s capacity, always under appropriate clinical supervision and without displacing their judgment.

AI supports. The professional decides.

Although AI can automatically perform tasks involving information gathering, organization, and early detection, the professional’s role remains unchanged: data management capabilities help retrieve and present patient information in a useful and timely way, but the professional is the one who performs the following tasks:

• Interprets
• Provides context
• Decides
• Supervises
• Talks with the patient
• Assumes clinical responsibility

A clinical AI tool should not be evaluated solely on how well its model works, but on how it contributes to the professional’s work. The systematic review by Nuutinen and Leskelä (2023) specifically highlights the need to evaluate clinicians’ performance with and without support from machine learning systems, considering aspects such as the interface, the cases analyzed, and how performance is measured. In other words, strong results from an AI model do not, on their own, mean that it improves the professional’s decision-making.

AI can gather and structure information so that the professional can then apply their clinical judgment. In a supervised care process, technology can support certain stages, but it should not replace the professional in clinical decision-making or determine the steps of a treatment on its own.

Responsibility for clinical decisions must remain clearly defined and under the supervision of the healthcare professional.

Freeing up time for what requires clinical judgment

The focus should not be on how many things artificial intelligence can do within the clinical process, but on the value it brings to the professional and its ability to free up time for tasks that require clinical judgment.

Doctors do not necessarily need every possible data point. They need useful, structured information that is available when they need it.

Applying AI to clinical documentation tasks can improve efficiency and reduce professionals’ workloads. The review by Lee et al. (2024), which analyzed 36 studies, found improvements in documentation efficiency and accuracy, but also identified integration with electronic health records and error management as key challenges in translating these benefits into clinical practice.

For this reason, useful clinical AI should have certain characteristics:

• Be clinically validated.
• Be transparent about how it works.
• Be supervised by professionals.
• Be integrated into the care workflow.
• Ensure information security and protection.
• Communicate its limitations.
• Be designed to support decisions, not replace them.

The goal is not simply for professionals to work less, but for them to devote more of their time to what truly requires their clinical expertise: interpreting, deciding, and supporting the patient.

The future is not doctors versus AI

The question is not whether AI belongs in the consultation room. The question is how to design tools that add value without displacing what makes healthcare professionals indispensable: their knowledge, their judgment, and their ability to understand the patient.

That is the role clinical AI should play: expanding the professional’s capacity, not replacing it.

With this approach, Nexi Health aims to provide continuity in monitoring mental health patients between appointments. Bruna interacts with the patient before appointments and between visits, collecting clinical and emotional information about their progress, while Meripsy integrates and structures that information so the professional can access a longitudinal view of the patient. This way, relevant information generated between one appointment and the next does not remain isolated. Instead, it can reach the professional prepared for review, allowing them to devote more time to assessment, decision-making, and clinical support.

The question, therefore, should not be how much an AI can do, but which tasks it can safely take on so that the professional can devote more time to those that require clinical judgment.

References

• Lee, S., et al. (2024). Evaluating the impact of artificial intelligence (AI) on clinical documentation efficiency and accuracy across clinical settings: A scoping review. Cureus. https://doi.org/10.7759/cureus.73994
• Nuutinen, M., & Leskelä, R.-L. (2023). Systematic review of the performance evaluation of clinicians with or without the aid of machine learning clinical decision support system. Health and Technology. https://doi.org/10.1007/s12553-023-00763-1