Exam Room · Part V · The Therapeutic Armory

Artificial Intelligence Is Here

Chapter 13 Bedside Questions · 8 questions · about 8 minutes

What this chapter diagnoses

Clinical AI is already in the workflow and in front of patients, cleared without trials, sold on demonstrations, changing after approval, and deployed with no named indication, no local validation, no monitoring plan, no accountable physician and no stopping rule.

The clinical picture

AI as a new drug class that arrived before the trial: read the label, run a formulary review, dose it, watch for adverse effects, and write discontinuation criteria.

The examination

Answer as your executive team would, with the evidence in hand. Each question has a factual answer in a system you already own.

Do you have a chartered AI governance body with a physician chair and the authority to suspend a deployment?

Add a note: where is the evidence, or who owns it?

For each AI tool running today, can you show the indication, local validation, monitoring plan and discontinuation criteria on one page?

Add a note: where is the evidence, or who owns it?

Has any AI model been validated against your own patients rather than the vendor's? Which one, and when?

Add a note: where is the evidence, or who owns it?

Has subgroup performance been examined by race, ethnicity, language and payer? If not, you do not know whether you are running Obermeyer's algorithm.

Add a note: where is the evidence, or who owns it?

If you deployed ambient documentation, what is your audited fabrication rate, who audits it, and has it been tested in an interpreter-mediated encounter?

Add a note: where is the evidence, or who owns it?

Who is accountable for whether deployments build clinical skill or erode it? The first published deskilling signal appeared in three months.

Add a note: where is the evidence, or who owns it?

Has your AI strategy been reconciled with your workforce strategy and training pipeline, or are they managed in separate rooms?

Add a note: where is the evidence, or who owns it?

When did you last suspend an AI deployment because the data did not support continuing? If never, your governance is theoretical.

Add a note: where is the evidence, or who owns it?

A reflection tool from the book, not a validated diagnostic test or benchmark. Your answers and notes stay in this browser unless you choose to send them. Please don't enter patient information.

If no demonstration exists of how it performs on your patients, it has not been validated, it has just been installed.

The Diagnostician, Chapter 13