Guardrails: Guiding Human Decisions in the Age of AI

Everyone arguing about AI is staring at the wrong layer, this book insists: not data, not algorithms, but who gets to decide and under what constraints, because "the most-needed innovation is not technological, but social." The case studies do the persuading: a biased mortgage algorithm, Wikipedia's editing rules, a Cold War disinformation operation. A physician reading far outside his own field found that "the embedded stories provide a foundation for understanding." And a reviewer who studies guardrail design for a living calls it an "impressively detailed and useful book that is also short, fascinating, and readable," then boils the thesis down to "guardrails are not walls": rules that flex with circumstances where a wall only blocks the road.

Start with the title trap: a healthy share of this book's natural audience wants to know how chatbot content filters work, and an AI safety specialist had to spell out that all of that "is NOT the topic of this book." What those buyers actually get is "a dense, complex read, written with an academic audience in mind." Push through anyway and the payoff never arrives, because anyone hoping the authors "will lay out a one-size-fits-all model of regulation will be disappointed": the framework stays a framework, principles with no playbook attached. Which would explain the reception. The biggest block of Goodreads raters parks at three stars, and the page still reads "No one has reviewed this book yet": not one of them felt moved to say why.