[The original title of this review was: "Great book, though a few inaccuracies & bold claims". See below for two updates on the review and why I changed the title.]
I'm a little over halfway with this recently published book, which I'm really enjoying so far - and I expect to enjoy it all the way to the end. A lot of great and unexpected insights here, and it seems that the authors did a good job explaining extremely complex algorithms and showing their applicability to real life (though it's hard for me to tell how good their explanations are to a novice, since I'm an expert in the field - I have two masters in Computer Science and working on my PhD, and was familiar with 90% of the algorithms described before opening the book).
My biggest quibble with this book (and the reason they lost a star) is that I noticed a few annoying/sloppy inaccuracies, which makes [made! - see below for updates] me ever so slightly doubt the accuracy and veracity of other areas of the book that I'm less familiar with. The other issue is the boldness of their (otherwise very interesting) conjectures.
For example, the authors misunderstand and misquote the 2-minute rule from David Allen's Getting Things Done, claiming the rule tells you to perform any less than 2-min task immediately when it occurs to you - and essentially simplifying the entire GTD system into the 2-min rule, which is in fact a tiny part of GTD (pg. 105-106). In fact, however, Allen does not suggest that at all - that would distract you from whatever you're currently engaged with, i.e. require a context switch (the costs of which the authors discuss at length). Instead, you should write that task down and add it to your intray, just like any other task. The 2-minute rule is applied later, while clearing your intray (which can be anytime in the next 48 hours). The point of the 2-minute rule is that the time spent on adding this task into your otherwise-extremely-flexible GTD system, and then tracking it in said system, would take longer than two minutes. This type of tracking is akin to what the authors refer to as "meta-work", and thus performing the 2-min task at inbox clearing time saves you an equal or greater amount of meta-work later. This is completely in line with the type of scheduling suggestions that the authors discuss. I'm not familiar with the other popular advice books the authors quote in the scheduling chapter or in the others chapters (e.g. the empty-your-closet type books they discuss in chapter 4), so I don't know if there are other such mischaracterizations, but it makes me suspect there might be. And I get that they're trying to differentiate their own advice from "all the other pop books out there", but if they're going to explicitly cite other books, they should try not to misrepresent them.
Also, when discussing the Gittins rule and the multi-armed bandit problem, they say that a machine with a 0-0 record has "a Gittins index of 0.7029. In other words, something you have no experience with whatsoever is more attractive than a machine that you know pays out seven times out of ten!" (pg. 40). However, their own table on the same page clearly shows that a machine with a 7-3 record has a Gittins index of 0.7187, making such a machine ever so slightly superior to a 0-0 one. After some more reading I realized that what they meant was that a machine with a 0-0 record and *uncertainty* is better than a *certain payout* of 70% (i.e. guaranteed to payout 7 out of 10), but that was not what the text implied.
To be clear, these inaccuracies in and of themselves aren't huge - but they planted a seed of doubt in my mind [which is not as big anymore - see below] as to whether there were other such misrepresentations or inaccuracies in the book that I simply hadn't caught, and detracted from my enjoyment of the book.
The other concern I have with this book is that several chapters end with provocative suggestions that aren't actually empirically-backed. These conjectures are cool, but I'd have liked to see scientists be more careful about making such bold claims, or at least couching them in the need for more research to establish whether they were entirely true. One example here was the discussion about the decline of aging supposedly being a result of simply having a larger history to remember (pgs 103-104). This is a fascinating conjecture, and one that deserves to be studied properly, but they are basing it on some research work that was not age-related. I suspect the authors may be on to something, at least in the context of "normal aging" cognitive decline as opposed to, say, alzheimer-related decline. However, as stated in the text, the conjectures are stated a bit too strongly for my tastes ("But as you age, and begin to experience these sporadic latencies, take heart: the length of the delay is partly an indicator of the extent of your experience.", pg 104). I'd hate to see anyone making decisions based on them - potentially missing an earlier diagnosis, say, of alzheimer's, because the authors claimed that cognitive decline is totally normal.
Quibbles and concerns notwithstanding, I'm definitely enjoying the book and I think it's a great addition to the new genre of what's being called by some "science-help". It's also a good read for people who are tired of the same-old, and thirsty for some advice that's off the beaten path.
The rest of the book was as good as I expected.
Additionally, I sent this review to the lead author (Brian Christian) in case he wanted to address these issues. I was delighted to receive a very thoughtful response from him! They will be fixing the Gittens rule description in the paperback edition, to make it clearer to the reader. The author respectfully disagreed with me on the other two issues (GTD 2 minute rule & cognitive decline).
Given what I saw in the email, I'd say the intentions behind the book definitely merit 5 stars (even though I still disagree on their presentation of those two topics). However, I'll leave the original title & rating of 4 stars as it stands for the original hardcover edition, and for consistency's sake. As I originally said, the book stands as an excellent addition to the genre, and also likely as a great first exposure into Computer Science if you've never had any.
Apparently, this review is now listed as the top most helpful review on Amazon (cool!). The book has been so successful that the first author (Brian Christian) recently informed me that the book is now on its third printing, which means that the Gittins index issue mentioned above is now fixed in the current and future editions. As for the other issues I had, they are more subjective in nature, and not large enough in and of themselves to merit the original (harsher) title of the review. Again, for completeness' sake and to avoid rewriting history, I leave the original review as its stands and the original title is listed below the new title, with only a few comments in brackets leading readers to these updates in the bottom.
Despite being an East-coaster, I'm a member of the Long Now Foundation, which--when I'm asked to describe it--I usually say is like TED, but with a long term view and way better substance. The Long Now gives regular talks, and then puts those talks up in video and audio form for others, who couldn't be in attendance. I subscribe to the podcast in iTunes, and listen to it--along with other podcasts--on my way to and from work.
A few months ago, Brian Christian was the guest speaker, and gave a talk centered around the subject matter of his latest book: Algorithms to Live By. The talk was fascinating, and contained a nice mixture of computer science, statistics, and humor to win the crowd over, and Christian managed to do so without coming across as too "pop science."
I purchased the book that same week, and between juggling work responsibilities and twins, managed to carve out about an hour each night to read through it. There were chapters that held my interest, and chapters that didn't, but overall the book was a fantastic mix of how various computer science problems are also real work problems, and algorithms that solve one can be applied to the other as well.
The first thing that catches you in the book is the discussion of optimal stopping, and how given a decision that needs to be made, you should begin making your choice after 37% of the options have been mulled over, assuming any of the next decisions/options are better than the ones that came before. This is illustrated with the secretary problem, and you can see why the authors led with this example not just in the book, but also in the Long Now talk. It seems both crazy and fascinating to have a difficult decision boiled down to such a hard percentage. The authors then go over different variations of the problem, and show how slight alteration can bring the best outcome.
The authors (Christian and Tom Griffiths) then follow this up with a rapid succession of entertaining problems such as exploit/explore to determine whether you should go with something that you know, or try something new, as well as chapters on sorting, caching, and scheduling, giving messy desk people hope by showing that a stack of files on a desk where something searched for is retrieved and then placed on top of the pile will eventually result in the most optimized sorting methodology for the job, and reminding older, forgetful people that accumulation of knowledge can result in greater time to sift and retrieve that information, renaming so-called brain farts to caching misses.
The chapter on Bayes' rule is where things start to get a little bogged down, but only in the beginning. Eventually, the chapter turns into an explanation on forecasting, showing which various predictive methodologies should be used for which various distributions--even equating the Erlang distribution to politics.
The back half of the book isn't as tight or as entertaining as the parts that came before it, but overfitting was a great read to be perusing while Nate Silver was being hammered for his polling methodology in the most recent election, and the chapter on networking gave a great, easy-to-read introduction to how information networks differ from telephony. The authors then conclude the book with game theory, discussing the tragedy of the commons, and how, as a society, we could pursue better options in order to ensure mass participation in important initiatives.
As somebody who studies and works in computer science and mathematics, I can say that casual readers will likely get lost in some sections, but powering through or re-reading will get you on to the more entertaining sections. This is a great book that works as a science popularizer without injecting fluffy prose/concepts or dumbing the material down.