7 rules for how physicians should use AI responsibly

DR GPT™ perspective

Seven rules written for a working clinician rather than a committee. If you read one thing of mine about clinical practice, read this one.

The rule that does the most work is the first: treat AI output as a draft, not a decision. It sounds obvious and it is violated constantly, because a well-formatted note reads like a finished thought. Polish is not accuracy. The better the output looks, the harder it is to audit, which is the opposite of what most people assume when they buy a tool.

The rule people skip is the third: look for what is missing. Every clinician knows how to evaluate what is on the page. Almost nobody is trained to evaluate what is absent. A model will sound completely confident while omitting a recent lab, misreading a medication, or failing to notice that the patient in front of you does not match the pattern it learned. Reading for absence is a skill, and it is the one this decade of medicine will be judged on.

The rule with legal teeth is the fifth: document honestly. Write down what the AI produced and what you decided, separately. Not because it is good hygiene, but because when the question of responsibility is eventually asked, the note is the only thing that will speak for you.

What I would add now, which the piece does not say plainly enough: none of these rules survive a system that gives a clinician eleven seconds and no authority to disagree. Rules are for individuals. Oversight is a property of the system around them. If your department cannot name the person who owns a given AI output, the seven rules are decoration.

Originally published on KevinMD. Read the full essay at the link below, or see how I build these arguments for a room as a speaker.