Free preview: This is Chapter 1 of AI in Healthcare: Promise, Practice, and the Human Stakes, shown in full. The whole book is free with a free account.
It is a clear-eyed look at where AI is genuinely helping patients and clinicians and where it is quietly failing them, across diagnosis, drug discovery, fraud, bias, privacy, jobs and regulation.
Chapter 1: The Wrong Kind of Wrong
Every industry says it is being transformed by artificial intelligence. Healthcare actually has a claim to that word that most industries do not, because the cost of getting it wrong here is not a bad recommendation or a wasted ad budget. It is a missed diagnosis, a denied treatment, or a life.
That single fact changes everything about how AI should be judged in this field. When a streaming service recommends a film you dislike, you scroll past it. When a hospital algorithm misses a warning sign in a deteriorating patient, someone can die. That is not a rhetorical flourish. It is the reason this book needs to exist.
A shopping app that occasionally suggests the wrong product is mildly annoying. A fraud detection system that flags a legitimate claim is a paperwork headache. A diagnostic tool that misses an atypical pneumonia because the patient never ran a fever is a different category of problem entirely. This book does not treat healthcare AI as a single story of either triumph or danger, because it is neither. It is a fast-moving set of tools, spreading through clinics, insurance offices, and research labs at a pace regulators are still struggling to keep up with, with real benefits and real failures documented side by side.
To give a sense of scale before going further: as of early 2026, the US Food and Drug Administration’s own public database listed roughly 1,500 AI and machine-learning-enabled medical devices authorized for use in the United States, with the most recent additions arriving weeks before this book was written. That number was six in 2015. The growth is not a projection. It has already happened.
What this book asks is a more useful question than “is AI good or bad for healthcare.” It asks where it is actually working, where it is failing, who benefits, who is put at risk, and what a smart patient, clinician, or healthcare worker should actually do about it.
The short version, which the rest of the book will earn: it works in narrow, well-defined tasks where the training data matches the real world, such as reading mammograms, flagging early sepsis, and predicting which discharged patient is likely to return. In those cases the evidence is genuinely encouraging. It fails in predictable, documented ways: atypical presentations that do not match the textbook, patient populations underrepresented in the training data, and situations where a clinician’s trust in the tool outpaces the tool’s reliability. The distance between those two realities, between the promise and the practice, is where this book lives.
Keep reading
Open the full book. A free account is needed to read the remaining chapters, and it takes about a minute: create a free account or log in.