Book: L’Incertitude apprivoisée, by Roger Vandomme

A data scientist for more than thirty years, Roger Vandomme occupies an in-between space too often neglected: making the complex clear without betraying it.

Structured in three movements — understand, apply, reflect — the book makes uncertainty the guiding thread of a long genealogy, from Pascal to the large language models.

At a time when mastery of AI is becoming a matter of power, a book useful to anyone who must decide, fund or regulate without being able to remain wholly ignorant of the subject.

Roger Vandomme, L’Incertitude apprivoisée. Origines, fonctionnement et usages de l’intelligence artificielle, 2026, 243 pages.

The literature on artificial intelligence divides roughly into two camps: technical manuals, which assume an already-equipped reader, and popular essays, which often give in to the ease of the catastrophist narrative or the enchanted promise. Roger Vandomme — a data scientist for more than thirty years, a consultant for sectors as varied as banking, defence and telecommunications, a former teacher at the Canadian Forces College and the University of Toronto — set out to occupy an in-between space: to make the complex clear without betraying it, in his own words.

A journey in three movements

The book is structured in three parts of unequal length but coherent logic. The first — and most developed — deals with understanding the foundations: what it is to decide under uncertainty, what it is to know, how a model is built, what algorithms exist, how neural networks work. The second part addresses practical uses: tools, professions, fields of application, launching a project, generative and agentic AI. The third, more philosophical, examines the fears, the ethical dilemmas and the existential questions this technology raises.

This division into three movements — understand, apply, reflect — is the mark of a teacher. Roger Vandomme does not merely want to inform: he wants to take the reader from one point to another. The introduction sets the frame frankly: « 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. »

The heart of the book: uncertainty as guiding thread

The title is not a marketing device. Uncertainty is truly the thread running through the whole book. Roger Vandomme begins with a chapter on decision to show that artificial intelligence was not born from a vacuum but from a fundamental anthropological need: to reduce uncertainty in order to act better. From Pascal and Fermat inventing the calculus of probabilities out of games of chance, to Kahneman describing the cognitive biases of human decision-making, the book traces a long genealogy that gives AI its historical depth.

Rather than beginning with a definition of AI, the author begins with the question AI is supposed to solve: how to decide in an uncertain world?

This entry through decision is one of the book’s original features. In doing so, it places the technique within an intellectual continuity and reminds us that the algorithm is not a recent invention but the culmination of centuries of formalised reasoning. The chronological fresco in the appendix — which runs from the agricultural revolution and the empirical discovery of seed-to-harvest correlations to the LLMs of 2024 — gives this ambition concrete form.

Read also: Leo XIV facing artificial intelligence: Babel or Jerusalem

A pedagogy through metaphor and example

The technical part is probably the most successful. Roger Vandomme explains neural networks, backpropagation, decision trees, random forests and transformers without ever assuming any mathematical prerequisites. His metaphors are well chosen: crossing a street at a red light to illustrate what a decision algorithm is; the confusion matrix explained through medical screening — when should one favour precision, when should one favour recall? The example of spam versus undetected cancers is limpid and memorable. Random forests are presented as a « democratic vote » among hundreds of trees, each of which has seen a different portion of the data: it is accurate and accessible.

The section devoted to deep learning and neural networks is particularly well handled. In it, Roger Vandomme retraces the AI winters — the periods of disillusion that followed each wave of enthusiasm, from Minsky and Papert’s critique in 1969 to the second winter of 1987-1993 — to show that the current triumph is not self-evident but the result of patient accumulation. This perspective tempers the ambient hysteria: if today’s enthusiasts are often right, their predecessors too believed they held the revolution in their hands, before the limits revealed themselves.

Closing reflections that open onto the essential

The book ends with a chapter of « final reflections », written just as the manuscript was sent to the publisher, in February 2026, from Toronto and Nice. In it, Roger Vandomme notes Yann LeCun’s departure from Meta to found AMI Labs in Paris, the developments of Claude and of Anthropic’s agents, and the rise of agentic AI. This chapter has the virtue of reminding us that any book on AI is dated the moment it appears — and that this is precisely why the foundations matter.

« A model is only a map, sometimes a very precise one, but it is never the territory. A performance is not a truth. »

The conclusion is perhaps the densest page of the book. This distinction between the map and the territory — borrowed from Korzybski without being named — sums up Roger Vandomme’s project. AI makes uncertainty calculable, visible and shareable, but it does not abolish it. It shifts the responsibility of deciding without doing away with it. And « the real danger is not that AI becomes conscious. It is that it be used without conscience. »

Read also: War in the age of AI: when algorithms transform the battlefield

In a context where mastery of AI is becoming a matter of power, and where political and military decision-makers are often the least well equipped to understand its mechanisms, a work that takes the time to explain what algorithms really do has a usefulness that goes well beyond its initial target audience. It is a book to put into the hands of anyone who must decide, fund or regulate, without being able to afford to remain wholly ignorant of the subject.

Roger Vandomme, L’Incertitude apprivoisée. Origines, fonctionnement et usages de l’intelligence artificielle, 2026, 243 pages.

Martin Capistran is a lawyer and holds a doctorate in law.