Artificial intelligence is already entering psychiatry, but its most useful role is often less dramatic than the headlines suggest. AI can help structure medical records, summarise information, support documentation and make it easier for a person to prepare questions before a consultation. That can free up time for what requires human judgement.
Imagine a psychiatrist who needs to review many years of clinical notes. An AI tool might highlight previous diagnoses, changes in medication and important events. This may be a good starting point, but not a conclusion. A system may overlook why a medication was stopped, misunderstand a phrase or present uncertain information as fact. Generative AI can write very convincingly and still be wrong. Well expressed does not necessarily mean correct.
The problem becomes more apparent when AI moves from administration into clinical decisions. Imagine a system recommending one antidepressant over another but being unable to explain whether the recommendation is based on sleep problems, previous side effects, lack of efficacy or a hidden pattern in its training data. Or a risk system marking a person as “high risk” without being able to explain whether this reflects bereavement, recent contact with emergency psychiatry, stopping medication or bias in the data. In such situations, a number is not enough; the clinician must understand what the number means.
The same applies to safety. Diagnosis, medication prescribing, assessment of suicide or violence risk and management of acute crises have consequences that require professional responsibility. An AI system can identify a pattern, but it cannot assume responsibility for the decision that follows. At the same time, psychiatric information is particularly sensitive, and identifiable information should not be entered into public or unauthorised AI systems.
The interesting future is therefore not “psychiatrist or AI” but psychiatrist with AI, when the technology actually improves the work. AI can help sort, remember and analyse. Humans must continue to assess, understand, explain and take responsibility. AI may become a highly skilled assistant—but it should not be the on-call psychiatrist.