As generative AI tools like Doximity, OpenEvidence, and UpToDate’s Expert AI make medical information faster and easier to access, recalling isolated facts will become less differentiating. In medical school, I lived on Anki flashcards, trying to cram as much as possible for boards and shelf exams. Now, many factual questions can be answered by AI in seconds, often with links to primary sources.

Physicians still need a strong knowledge base. We need enough knowledge to recognize what matters, ask the right questions, and catch an answer that doesn’t make sense. But our value will increasingly come from knowing how to apply that information to the patient in front of us.

The class I dreaded most in medical school involved standardized patients, but it ended up being one of the most useful. I would evaluate trained actors portraying different clinical scenarios. The open-ended nature of these encounters made me anxious, especially when someone started crying or yelling (this did happen). You can memorize the differential for chest pain easily. It’s much harder to make flashcards for delivering a cancer diagnosis or responding to a distressed patient.

These encounters gave me early repetitions in clinical judgment. I had to review the available data, form an initial hypothesis, and then revise it as I interviewed and examined the patient. At the same time, I had to read their body language and level of distress.

Consider a patient with asthma who presents with shortness of breath. They’re pausing every few words to breathe, but I don’t hear wheezing. Then they mention a 20-hour plane ride, and I notice that the left leg is more swollen than the right. The initial label of an asthma exacerbation no longer fits. The patient may have a pulmonary embolism and needs urgent evaluation.

Clinical judgment is the ability to combine medical knowledge, incomplete data, the physical examination, patient context, and the consequences of being wrong… and then make a decision.

We start building that skill in medical school and develop it through years of seeing patients under different conditions. Repeated experience can eventually turn deliberate reasoning into faster pattern recognition. The best clinicians still know when to slow down, question their first impression, and ask for another perspective.

AI will become increasingly useful in this process. It can retrieve evidence, organize a differential diagnosis, identify risk factors, and suggest a treatment plan. For routine conditions, it may eventually produce recommendations that look a lot like clinical judgment.

The concern is that AI may also remove some of the repetitions we need to develop judgment. If a trainee asks AI for the differential before forming one independently, they may reach the correct answer without strengthening the reasoning that produced it. Used well, AI can provide feedback and expose blind spots. Used too early, it can substitute for the cognitive work that training is supposed to develop.

Intermediate-risk pulmonary embolism is a good example of where that distinction matters. The patient has evidence of right-heart strain but remains hemodynamically stable. Anticoagulation may be sufficient, while catheter-based treatment or thrombolysis could help—or expose the patient to unnecessary bleeding.

The data informs the decision, but it does not make it. A pulmonary embolism response team considers the patient’s trajectory, imaging, bleeding risk, comorbidities, preferences, and the hospital’s available resources. AI can help organize the evidence and identify relevant risk factors. It cannot yet replace the bedside assessment, multidisciplinary discussion, contextual understanding, and accountability required to make that decision.

Trainees and practicing clinicians will need to be deliberate about where AI enters this process. A useful approach is to think first, use AI second, and then compare. Form an assessment, ask the tool to challenge it, and examine the differences. Training is the safest time to build that habit because supervision and feedback can expose mistakes before those mistakes become independent practice patterns.

In summary, AI will make medical information easier to access, but information alone does not determine what a patient needs. Physicians will remain responsible for interpreting uncertainty, recognizing context, and accepting the consequences of a decision. As AI becomes more capable, those judgment skills will become more valuable and more important to protect during training.