Could artificial intelligence (AI) prescribe medications?

A new bill in America, called the “Healthy Technology Act of 2025,” could allow Artificial Intelligence (AI) – meaning very advanced computer programs – to prescribe medication. This would be possible if the state where it applies agrees and if the AI technology is approved by the U.S. Food and Drug Administration (FDA).

Currently, some doctors already use AI to help them with administrative tasks (such as taking notes) or to offer suggestions in medical decision-making. However, experts say more research is needed before an AI program can write prescriptions on its own, if this will ever be possible. For now, there is no AI technology that can do this safely and effectively.

How could AI help with prescribing medication?

Researchers are exploring two main directions:

1. Predictive tools: These analyze a patient’s electronic health record or genetic information to estimate the probability that a particular treatment will work.
2. “Digital twins”: AI creates computer models of real patients, on which different medications can be virtually tested to see which would be most suitable.

The discussion about AI prescribing medication also arises because many people face difficulties in accessing quality, affordable healthcare services.

Why are doctors worried?

Even though AI has potential, there are several reasons for concern:

• Accuracy: AI can make mistakes that seem correct (called “hallucinations”). A mistake in prescribing medication could have serious consequences. For example, a study showed how an AI program misdiagnosed a patient based on keywords.
• Lack of physical examination: AI analyzes data but cannot physically examine a patient – it cannot listen to the lungs, palpate the abdomen, or observe other important physical signs.
• Understanding nuances: The decision regarding medication often involves a complex discussion between doctor and patient about risks, benefits, and personal preferences. It is hard to imagine how a machine could manage these nuances.
• Bias (partiality): AI programs are created by humans and learn from data collected from humans. If this data is incomplete or reflects existing biases (e.g., related to race or gender), AI can adopt them and make biased recommendations.
• Data quality: To train AI correctly, very detailed and complete (“deeper”) datasets are needed, including genetic, socio-economic factors, etc., not just large amounts of data (“broader”).

Regulatory issues

Even if the law were to mandate FDA approval, experts are concerned. Current approval standards for AI-based medical devices might be less strict than those for drugs. Many AI devices are approved through a procedure that only requires them to function similarly to older devices, which might be insufficient for a program making large-scale prescribing decisions.
Furthermore, AI programs can change and evolve as they learn from new data. How their performance will be monitored long-term, post-approval, is still unclear.

Risks and liability

There is a risk that the technology could be launched on the market before it is truly ready, which could harm patients and slow progress in the field. An important and unanswered question is who would be liable if AI prescribes the wrong medication and the patient suffers.

What are the chances of the bill passing?

Similar proposals were introduced in the U.S. Congress in 2021 and 2023 but were not adopted. Even if the bill passes, it remains to be seen whether doctors would have enough confidence to use AI for such important medical decisions.

What might the future look like?

Some experts imagine that AI could be used in the future for prescribing in very limited, low-risk situations, under a doctor’s supervision. For example, AI could verify if a patient meets all criteria on a checklist for a specific simple condition, and a doctor would review and sign the prescription. AI could also connect to electronic health records to reduce prescribing errors, being aware of a patient’s allergies, even if the attending physician does not have access to the complete history.

When will we know if AI is ready?

To be sure, AI that prescribes medication would need to be tested in rigorous prospective clinical trials. These studies would need to directly compare patients treated based on AI recommendations with those treated traditionally and demonstrate that AI makes decisions faster, with at least equally good performance, or that it leads to better patient outcomes.

Source Medscape