Russian roulette is "non-ergodic", in that the population average outcome is different from the average outcome of repeating the game. On a population basis, 5/6 people "win". But if you play many times, one "loss" is enough to ensure you can't play again. As it turns out, most systems we encounter every day are also non-ergodic. Mistaking them for ergodic—assuming that averages indicate what we should do—can cause bad outcomes.
Luca Dellanna's book, Ergodicity, is incredibly short. In fact, it's closer to the format of a long web essay. It spends most of its time exploring how non-ergodic systems impact our lives, and how they should influence decision making. One part that stood out was how incentives can be different whether a situation is ergodic or not. A company with replaceable employees (manual labor, perhaps) can treat them as mostly ergodic, while for each employee, the employment is non-ergodic. Additionally, natural selection is great for populations, but bad for individuals — so, learn to be more like a population than individual.
Overall, it's a useful and valuable read. The length is good — he doesn't fall into Malcolm Gladwell traps of filling pages with anecdotes that tell the same story, but also provides enough real-world context to build a solid mental model of the concept.