Algorithms to Live By is the equivalent of a breezy summer read, a guilty pleasure that anyone in the analytics field can gobble up and consume with great delight. The book’s subtitle, “The Computer Science of Human Decisions,” sounds dry but the book is anything but. By delving into detailed descriptions of the common human dilemmas we all face, the authors provide buckets of fascinating insight into how computers can help us deal with our daily preoccupations in a productive manner.
The authors start things rolling in the first chapter with a engaging analysis of a class of problems they refer to as “optimal stopping.” Here’s an example: Let’s say you’re single and want to optimize your search for a suitable spouse. You’ve been on a number of dates and have now found someone who seems like a distinct possibility. But how do you really know that this person is the best you can get? Are there any guidelines that can be followed to tell you when you should stop looking and conclude that, with all probability, this is the best match that can be obtained? It turns out that there is. A study of probabilities tell us that after expending 37% of your time in the search, you should propose to the next person who seems better than anyone you have previously seen. The first 37% of dates are strictly for data gathering purposes. The leap to an spousal attachment can occur anytime thereafter. This, however, assumes that anyone you propose to will accept your offer. If the probability of an acceptance is less than 100%, that also affects the calculation.
Turning to more computer intensive problems, the authors next discuss issues of sorting and searching. The point is driven home that Google searches rely on efficient and effective sorting techniques. What is called a search is really a sort. The ubiquitous sorted list is essential to our ability to parse through search engine results. As is done throughout the book, the authors offer apt analogies that present arcane concepts in terms the average reader can understand. In this case, they compare the general problems of sorting with sport teams that devise different types of tournaments, such as single elimination and round robin, as a means to rank themselves from best to worst.
For readers interested in traditional statistics, the authors include a compelling chapter on Bayes’ rule. For anyone who has struggled with Bayesian probabilities in a statistics course, the authors do a wonderful job of explaining the concepts in clear and simple terms, and also cover related offshoots such as Laplace’s Law and the Copernican Principle. From there, the text moves on to issues of overfitting data and techniques such as Constraint Relaxation and Lagrangian Relaxation.
Final chapters turn to algorithms utilized in internet and network communications, and then to the fascinating world of game theory. The subtle point is made that, whereas previous topics have dealt with basic human flaws such as our limited ability to process information and our race against the irrevocability of time, game theory extends these issues to an explicit consideration of conflicts between man and society. Returning to the optimal stopping analysis of finding a suitable mate discussed in the first chapter, the authors relate game theory’s prisoner’s dilemma to a consideration as to why more people don’t leave marriages after they’ve made the commitment. Game theory suggests that marriage alters the structure of the relationship game, shifting the equilibrium between the partners to an arrangement with a different set of payoffs. Equally fascinating insights are revealed in the worlds of poker, auctions and even religious belief.
Algorithms to Live By is an engaging read that covers the gamut of topics covered in more technical books on statistics, computer algorithms, networking and game theory, but does it in an intuitive manner that relates quantitative methods to the human condition.
Algorithms to Live By: The Computer Science of Human Decisions
by Brian Christian and Tom Griffiths
Henry Holt, April 2016
351 pages, $28.00