Résumé de Algorithms to Live By

In Algorithms to Live By, computer scientist Brian Christian and cognitive scientist Tom Griffiths argue that the algorithmic problems computer scientists have studied for decades map directly onto everyday human decisions. Each chapter pairs a classic algorithm with a matching life problem: optimal stopping, illustrated by the secretary problem and the "37 percent rule" for deciding when to stop searching and commit, whether to an apartment, a job, or a spouse; explore versus exploit trade-offs, framed through the multi-armed bandit problem and Gittins index; sorting and search, where binary search and the cost of comparison explain how to organize closets, inboxes, and libraries; caching, in which Bélády's optimal page-replacement algorithm illuminates which items to keep readily at hand and the value of "thrashing" versus eviction; scheduling, drawing on Moore's algorithm and the "least slack" principle to show why shortening a to-do list beats reordering it; and Bayes's rule for updating beliefs with new evidence. The book also covers overfitting, tying it to the tension between simple and complex models and the value of early stopping; relaxation and Lagrangian multipliers as tools for loosening constraints; randomness and the surprising utility of randomizing (as in Markov chains for sampling the web); and game theory's Nash equilibria for bargaining and fairness, including the "I cut, you choose" procedure. Throughout, Christian and Griffiths stress that human intuition is often already near-optimal, but where it fails, computer science offers provably better rules. They also warn against over-applying algorithms, noting cases where human judgment outperforms mechanical rule-following. Drawing on experiments such as sorting studies and optimal-stopping trials, the book translates theoretical computer science into a rigorous, evidence-based guide to decision-making rather than a collection of motivational slogans.