AI in Medicine

A four-week elective for fourth-year medical students

Author

Sean Davis

Published

October 7, 2026

Preface

This is the student-facing book for a four-week, ungraded elective on AI in medicine for fourth-year medical students. Each chapter is one module: one to two hours of reading, watching, and doing that stands alone, so a missed week costs one chapter, not the course. Nothing in it needs an install, a paid account, or a real patient.

How to read a chapter

Every chapter follows the same structure, and the headings are the same in every one:

  1. An opening scenario. A clinical situation you will recognize, before any jargon. About five minutes.
  2. Why this matters. What changes in residency in three months. Five minutes.
  3. How it works. The idea, in prose, with figures and tables. Twenty to thirty minutes.
  4. Watch or listen. One video or podcast episode, with a note on what to listen for and where you can stop. Fifteen to thirty minutes.
  5. Do. The activity, written so you can finish it alone. Wherever it touches the literature, you choose your specialty. Thirty to forty-five minutes.
  6. Check. A rubric you score yourself against. Five to ten minutes.
  7. What this changes for you. The standing question, what does this change about how you’ll practice? Written down, not optional, and the scenario from the opening comes back with a better answer.
  8. Go deeper. Papers, talks, and videos. Claims nobody has yet checked are flagged [VERIFY] and left flagged until someone has.

A chapter’s subtitle gives its time budget and the competencies it serves (AIM-1 through AIM-9 and AIM-X; the coverage table below says what each one is). Three rules hold everywhere: no patient information enters any tool, ever; verify before it counts; you sign the note.

Three ideas recur across the chapters, and each is named the same way wherever it appears. Proxies fail: almost every AI failure in medicine is a stand-in that looked reasonable. Efficiency gains get eaten by volume: per-unit savings are cancelled out by growth in use. The published number is the optimistic one: the most prominently reported statistic is nearly always the best case. A chapter that uses one of these says so.

The chapters, in the default order

Eleven chapters for eight sessions. The default four-week assembly runs them in this order, and the book’s contents are grouped by week the same way; the three spare chapters can be swapped in by cohort interest. Each one stands alone, so any order works. The syllabus says what a week looks like and when things are due. One assignment runs across all four weeks, the weekly resource log, where your own reading goes; several chapter activities feed it. Whoever runs the course reads the For instructors appendix.

Week 1, orientation and stance

  1. How these systems work: the eGFR is a model you already trust, and the chatbot is the same loop with words.
  2. Centaur or cyborg: what you delegate to the machine, and what you keep.

Week 2, judgment

  1. Critical appraisal: checking AI tools, their answers, and the evidence about both.
  2. Bias and equity: the proxy problem, from pulse oximetry to cost-as-need.

Week 3, practice

  1. Agentic coding: build a clinical tool in an hour, then find what is wrong with it.
  2. Talking to patients about AI: the most important session in the course.

Week 4, consequences and closing

  1. Will AI take your job? or What AI costs the planet, the cohort’s choice.
  2. Ethics, liability, and the device line: who answers for the note, and the closing question.

Spare, or for a hands-on cohort

Competency coverage

The course tags every chapter to a stable competency ID so it can be mapped to national standards when they arrive. The IDs and the coverage rules are in curriculum/competencies.md in the repository. In short: AIM-1 foundations; AIM-2 critical appraisal; AIM-3 bias and equity; AIM-4 ethics, law, and regulation; AIM-5 clinical use; AIM-6 talking to patients; AIM-7 workflow and systems; AIM-8 health data science; AIM-9 research and scholarship; AIM-X professional identity, the closing question.

Table 1: Which chapter serves which competency, from each chapter’s own subtitle and objectives.

The rules say AIM-1 through AIM-6 must each appear at least once in any four-week assembly, with AIM-4 and AIM-5 recurring. The default order above meets them: every core competency appears, AIM-4 four times, AIM-2 three times, AIM-5 twice. AIM-8, health data science, is not the main subject of any chapter; the foundations chapter covers training and test sets, area under the curve, and dataset shift while making its own point, and that is the depth this course goes to.

The tutor and the answer keys

A chat tutor over these chapters is planned. When it ships, it will answer questions from the rendered text and the sources it cites, and it will hold the answer keys for each chapter’s Check. The answer keys are written by the same process as the chapters and are never published; the tutor uses them to grade your reasoning, not to give you the answer. Until it ships, the Check rubrics in each chapter are the whole of what is available, and they are written to be scored alone.

How this book was made

This book was built with AI. The author designed the course and the chapter pattern, wrote the first chapter, and then had ten AI agents draft the remaining chapters in parallel from the module plans, each with citations checked against PubMed, an answer key, and a review as six invented readers. The author read the result, and the chapters then went through a plain-language pass, a second six-persona review, a cross-reference pass, and a pass replacing metaphor with direct statement, each also run by AI agents under the author’s direction. The author edited and committed all of it. The full account, including what the AI did and what a person checked, is in the appendix How this book was made and the AI ledger. The course asks you to disclose and verify; this is the course doing the same.

The design behind the book, competencies, constraints, and schedule, is in the curriculum/ directory of the repository.