Roger Vandomme, a data scientist for more than thirty years and a consultant for sectors including banking, defence and telecommunications, has written "An Approximate Intelligence : Where AI Began, How It Works, What It Means," published in 2026, 243 pages. The book is structured in three movements — understand, apply, reflect — with the first and most developed part covering foundations such as deciding under uncertainty, building models, algorithms and neural networks. The second part addresses practical uses, including tools, professions, fields of application, launching a project, and generative and agentic AI. The third part examines fears, ethical dilemmas and existential questions. The introduction states: « At a time when machines are learning at great speed, our collective responsibility is to learn to think with them and sometimes, if necessary to preserve our critical spirit, to learn to think against them. » Uncertainty serves as the book’s guiding thread, beginning with a chapter on decision to show that AI arose from a fundamental anthropological need to reduce uncertainty. The genealogy runs from Pascal and Fermat inventing the calculus of probabilities out of games of chance to Kahneman describing cognitive biases, giving AI historical depth. The author starts with the question AI is supposed to solve — how to decide in an uncertain world — rather than with a definition of AI.