When one thinks of algorithms, it is often in association with computers or machines. Not humans. It is also common to think algorithms are there to provide a simple, neat solution to complex problems only a machine could solve. Or that algorithms can, once fed enough information, predict one’s every action and solve every problem. The main premise of Algorithms to Live By is to disabuse one of such notions. Algorithms to Live By explores how regular people use algorithms without even realizing it in their day-to-day lives. By doing so, the authors hope to destigmatize the word and get people to see the concept differently. Though the book can be dry at times, the authors manage to write a book that is accessible to most people. And there are moments of insight that do make the book a fascinating read. As aforementioned, the book explores how people use algorithms in their day-to-day to accomplish tasks. They focus on several elements: explore/exploit, or when it is best to continue to look for something better or make a choice from what one already knows; sorting and tradeoffs; and scheduling being among the subjects of focus. What makes these sections interesting is that they often talk about tradeoffs that one would seem counterintuitive. An example of this is in the scheduling section. The authors mention how the placement of a task on a schedule may be influenced by how much one knows about the task: by its duration or difficulty. This may increase the difficulty of scheduling if one were to know every detail of every task that must be done for the day. They also mention that while some may be tempted to schedule tasks based on how easy they are, this may also come with downsides. Especially if one decides to prioritize harder tasks before easier ones, only to realize that its completion requires completing an easier task. They give an example of a NASA Mars rover being frozen due to this fact. The rover was programmed to prioritize high priority tasks first in its queue over low priority tasks. However, one of the low-priority tasks kept being pulled from the bottom of the queue to the top. This caused the rover to freeze. Thus, even well-thought-out systems can lead to problems. The above example with NASA shows another aspect of the book I like; the use of real world examples. The authors tell stories involving real world mathematicians and scientists struggling with these issues in their personal lives. This helps make the subjects feel personal and applicable to one's own life. In fact, I would argue that the only issue with the book is that these anecdotes seem to be an afterthought. This is due to the fact that the anecdotes become more prominent as the book progresses towards the end. Thus, the first few chapters can be somewhat dry in its presentation which may turn off a lay reader. Furthermore, the use of hypothetical scenarios in the earlier chapters feel like a pale imitation of the personal anecdotes of later chapters. All in all, this book was fairly enjoyable. While having some rough patches, the authors did try and succeed in making an accessible book.

For me, the book takes intellectual effort to absorb. As I was preparing to write this review, I was further impressed with the range of information presented by the authors. I am personally undertaking an investigation of machine learning, artificial intelligence, data mining, etc; The book fit into this investigation. If you have interests in this area (or areas), I think you'll find the book useful. It probably shouldn't have, but the parallels between common human problems and computer programming surprised me. As the book has had a large number of reviewers already, I will highlight some, but far from all, of the topics of each chapter so you may see if they make you curious. While the book speaks of algorithms to live by, the mathematics in the book is highly limited. Optimal stopping - how many people out of 100 possible candidates should one interview for a given position (including that of spouse)? 37%, Why? Read the book. The Explore/Exploit dichotomy - Should one ask the question "What's new" or "What's best"? Your answer may depend on your time horizon. As your time horizon shortens, "what's best" may be the better question. The book explains why. The book also looks at the multi-armed bandit as an example of the explore/exploit dichotomy. What's a multi-armed bandit? Think of the one-armed bandit in Vegas and multiply its arms. Mathematicians do so. Their conclusions may be useful. The trials of music critics also fit into the explore/exploit dichotomy. The authors explain why music critics find exploration a chore. Sorting - libraries are the metaphor for computer sorting. Human memory also requires sorting. Maybe the decline in memory as humans age may be due to the amount of information through which it must sort and not due to declining faculties. A five-year old has a lot less information to go through than a seventy-five year old. The authors consider sorting techniques with email, Yelp, and other common uses. There is much useful information. Caching - when is forgetting necessary? According to the authors, the first computer cache was developed for a supercomputer in 1962 ub Manchester, England. I wonder how "super" that computer was? Caching allows some information to be stored for repetitive use and uncached information to be kept in the background. Scheduling - many scheduling problems have "intractable" solutions. The authors suggest different solutions based on algorithms such as precedence constraints, earliest due date (one I personally use frequently, which I couple with a personal likely to get me in the most trouble the quickest test) and shortest processing time. The scheduling problem has received substantial effort from mathematicians. Bayes's Rule - how to use statistical inference to make useful predictions. Couple a well-defined problem with a range of prior outcomes and one can make accurate guesses. A .300 hitter comes to the plate against the same pitcher who has already struck the batter out twice and it may be a fair guess that the hitter is due for a hit. Overfitting - don't overthink and over complicate a problem. The authors advise against practicing the idolatry of data. A more complex theorem may well lead to less accuracy rather than more. On the level of incentive compensation, the authors quote Steve Jobs for being careful that you include only those elements in your incentive package that matter; you will get what you measure. Relaxatrion - the perfect is the enemy of the good. To get any useful answer from your mathematical model, it may be necessary to relax some of your constraints (insisting that your model never allow the traveling salesman to re-enter the same city twice may preclude any answer at all in a time period of less than the remaining life of the universe). Randomness - mathematicians sometimes realize that the best answer comes from sampling and not from strict calculations. This may explain why I get so many survey requests. Algorithms for prime numbers use this technique. And, apparently, thousands of years ago the Greeks were already looking for prime numbers. Networking - here the authors examine the "Byzantine generals" problem, which plays a part in explaining how computers communicate with each other. Game Theory - Alan Turing investigated the "halting problem" in the 1930s. What if you give your computer a problem and it just keeps going? Rock, paper, scissors is a game with which most are familiar. It, too, is part of game theory. When a game seems to have no satisfactory answer, maybe it's time to change the game. What happens when you have an "information cascade"? If any ot this interests you, I believe that you will enjoy the book. I recommend it highly.