Why you can't take medicines prescribed by AI

Dr. Google once ruled the roost. For any health problem, patients, instead of informing their doctor, would start tinkering around looking for the right diagnosis, treatment, and cure. A dangerous do-it-yourself approach that has almost become a trend. Now, Google is retreating, while Artificial Intelligence (AI) is advancing, with truly impressive numbers. According to a survey titled " Artificial Health," 94 percent of Italians search online for solutions to their physical and mental health problems, 43 percent through GPT and generative AI for health.

and one in seven young people changes the therapy and prescriptions signed by their doctor to comply with what the AI ​​says.

Few of these unfortunate patients, victims of the proliferation of Artificial Intelligence, are able to reflect on the fact that prescribing a drug requires a thorough clinical assessment, professional responsibility, and the ability to manage unpredictable situations—all things AI absolutely cannot guarantee.

The main problems with using AI for autonomous prescribing are:

• Incomplete information: AI may not know allergies, other therapies, pregnancy, pathologies, recent tests or other fundamental elements.
• Artificial Intelligence cannot visit the patient, have a clinical picture due to physical contact and knowledge of the patient, with his previous experiences which can be decisive for the purposes of new evaluations.
• Errors and “hallucinations”: An AI model can very reliably produce the wrong drug, dosage, or drug interaction.
• Clinical context: Two people with the same symptoms may have completely different diagnoses and treatments. Understanding the problem often requires a thorough physical examination and medical history.
• Interactions and contraindications: The safety of a drug can depend on age, kidney/liver function, other medications, and many other variables.
• Urgent situations: An AI may fail to adequately recognize a warning signal or may delay access to care.
• Responsibility: Prescribing entails a professional and legal responsibility that cannot simply be transferred to an algorithm.
• Update and reliability: Guidelines, approvals, and drug information change; a system must be verified against up-to-date clinical sources.
• Excessive automation: Even a very accurate system may cause the doctor or patient to automatically accept its suggestion without checking it.
• Dependence on pharmaceutical causes: Many scientific publications, which AI uses for diagnosis and treatment, and especially for prescribing medications, are funded by pharmaceutical companies. This makes AI-generated prescriptions even less reliable.