The Master Algorithm:How the Quest for the Ultimate Learning Machine Will Remake Our World (Chinese version) Hardcover – December 19, 2016
|New from||Used from|
Enhance your purchase
Explore your book, then jump right back to where you left off with Page Flip.
View high quality images that let you zoom in to take a closer look.
Enjoy features only possible in digital – start reading right away, carry your library with you, adjust the font, create shareable notes and highlights, and more.
Discover additional details about the events, people, and places in your book, with Wikipedia integration.
The Amazon Book Review
Find your next favorite read with editors’ recommendations. Read it now.
Enter your mobile number or email address below and we'll send you a link to download the free Kindle App. Then you can start reading Kindle books on your smartphone, tablet, or computer - no Kindle device required.
To get the free app, enter your mobile phone number.
Frequently bought together
Top reviews from the United States
There was a problem filtering reviews right now. Please try again later.
The book is split into multiple chapters which start from discussing abstractly the master algorithm and then move on to some of the philosophical issues associated with using such algorithms. In particular the author discusses at the core of believing in pattern recognition algorithms is belief in inductive reasoning. The author discusses human learning and gets into some neuroscience and how neural networks are constructed. The reader gets a vague sense of Hebbian learning and how neuron weighting are at the core of neural networks. The author spends a lot of time discussing various approaches in machine learning and gives the reader an intuitive feel of Bayesian learning. The author was an originator in a particular algorithm called naïve Bayes which greatly simplified solutions to certain problems and so the author introduces his ideas to the reader. Other machine learning ideas are introduced like genetic programming and multivariable regression. The author also discusses other machine learning algorithms which turn data into a vector and then look for close neighbors of the vector to classify the input. The author also spends some time on how unsupervised learning would look. The book combines computer science ideas and intuition and tries to use a fictitious robot as the means to convey ideas about how a computer would learn. The author finally introduces his own master algorithm called alchemy which combines most of the models described in the book. The reader really gets little actual sense of what's going on in the algorithm as the author qualifies one needs a PhD in computer science.
The Master Algorithm is the first book I have seen which introduces some of the ideas being used in machine learning to a general audience. It does so quite well and most of the ideas are absorbable. At the same time there are a few too many instances where the author is self promoting talking about all of the brilliant ideas he has had which have reshaped the field and how other areas of AI research of the past or Kurzweil and his singularity concept are idiotic. Despite probably being right in much of his analysis its arguing with no one on the other side and unproductive. Also the flavor of the writing is odd - it turns into some fantasy literature at times as though that makes the subject more digestible and in fact makes it more irritating. I enjoyed reading aspects of the book and do think the parts on what different schools of machine learning focus on are well written for a non expert, unfortunately there are many other parts of the book which one wants to get through as quickly as possible.
I have read a few textbooks on machine learning (intro to Statistical Learning by Hastie etc) and so I would say that my knowledge of ML is at the "textbook overview" level. Since I am not a ML practitioner, I may not be the best judge of a book such as this one, it was a fairly difficult read, and I know I need to read the book a second time to get an even greater appreciation and understanding of the concepts covered by the author. That being said, it was a very enjoyable book. The book was very different from any ML book that I've read or checked out either at a bookstore or online. I think one needs to have some knowledge of ML to appreciate the book - concepts like supervised learning, unsupervised learning, Bayesian inference, support vector machines, neural networks, etc .... The book deserves a 5-star rating because it added a lot of value to my understanding of ML, and increased my desire and curiosity to learn more about the field of ML ....
Thus, I chose this book both to see why we aren't there yet, and to see if I can still even follow the current arguments.
Like The Soul of a New Machine, this book is written to both be accessible to a literate general audience and rigorous enough for specialists. In my opinion, it entirely succeeds, although my son is better than me at math, so I skimmed some of the math-heavy discussions, as my interest is more in the philosophical issues and predictions about coming developments. In that, I was completely satisfied, as just when I was thinking of docking a star because the discussion was getting too deep into the weeds of particular implementations of machine learning, the author returned to a very interesting discussion of what it all means for our current and future society, complete with useful predictions about what jobs are still likely to be around in twenty years, and the inevitability of voters eventually approving a guaranteed basic income for those no longer able to be employed due to robots and computers doing ever more of the work previously done by humans.
I was glad to see the author also explored the dark side of all this - the potential for it to control people rather than to free them, and his suggestions on how to make sure the changes achieve good for all rather than for only a few. I was particularly interested in his conclusion that we needn't worry about machines ruling over us because even the best machine learners lack will, doing only what they have been programmed to do. I also like his idea that although machine learning may be the next step in evolution, it will be one taken with rather than instead of us, with humans branching out along with computers in ways not predictable yet, but likely to be beneficial.
Overall, a very good and very hopeful read, and comforting to still be able to follow along after a full career since grad school and years of retirement.
Top reviews from other countries
Dislike: First, it's not clear who the target audience is. The 'handwaving' opening and closing chapters seem aimed at the non expert. However, the core chapters demand a basic appreciation of Markov, Bayes and probability, which the average reader won't have. Second the language is idiosyncratic - particularly where the author refers to 'learning a model', by which he means 'training' or 'teaching' it logic and structure. While I recognize the lack of a good word for this, 'learning' doesn't seem right at all. I wondered if this was caused by translation from Portugese, but either way I think its the fault of Basic Books not doing their job as editors. The lack of good editing is a serious detraction and makes key sections far too difficult to read. This is noted by other reviewers who take a harsher line, but on the whole this detraction didn't diminish my overall rating.
Whilst I think there's stuff in the book that could be improved, it's hard for an author to make those kind of improvements without feedback. I can't help but think that the input of editors/agents/publishers should have made this a much better read. It seems to me that this book was crying out for a ghost-writer/co-author experienced in 'popular science' writing. More importantly, I think it desperately needed proof reading by people unfamiliar with machine learning. My guess is that drafts of this book were read primarily by people highly experienced in this subject, i.e. people who already understood the material being presented. I'm really struggling to see how someone at Basic Books could have read the book prior to publication and thought, “Yep, that was relatively easy to follow”.
The prologue claims the book is intended for a wide range of readers, pretty much from novice to expert. Personally, I would suggest that the only people who won't struggle with this book are experienced industry based practitioners and academics/researchers/post-docs in this area (and I guess also, readers who are happy to just skim sections they don't understand). To give some perspective, I have a grasp of ML basics, I currently work in a ML group (although my research is not explicitly ML), I've taken introductory and advanced masters units in the subject and have implemented some of the algorithms myself: Yet I struggled to understand much of this book – both the specific details and the broader overview.
I may reread some of this book in a couple of years when I have a bit more understanding and experience under my belt. Perhaps then I will appreciate the details and the author's perspective better. At my current level of experience, I found the book very hard going, a frustrating read and I don't feel like I learnt very much from it. For the time being, I will stick to machine learning text books.
Specific comments in no particular order (mostly negative - sorry):
- Key fundamentals of the subject are either, not explained (e.g. 'hypothesis' as an ML term), poorly explained (e.g genetic algorithms are explained devoid of any mention of 'selection'), or left till way too late in the book (e.g. supervised v unsupervised learning left till p203). ---This latter example is a real shame in my opinion, as I found this to be one of the best written chapters of the book.
- Many other areas did not get the foundational explanations needed prior to developing ideas further. Example: “The most important question in any analogical learner is how to measure similarity.” This crucial question is raised, then frustratingly left unaddressed because the subsequent text lacks any specific coverage of how this may be actually achieved.
- I felt that 'hypothesis', 'feature'/'attribute', 'label', 'example' should have been explicitly defined, in the context of what they mean for ML, right at the start of the book. This would have given the reader a much firmer footing before seeing further explanations using these terms routinely. Non ML readers will not appreciate, for example, the very specific use of 'hypothesis' wrt discussing ML algorithms.
- The first few pages of the prologue has been described by another reviewer as (IIRC) evangelistic nonsense – It's a harsh comment but it's pretty much on the nail. The author is right to highlight the role ML already plays in our world, but he's overstating the case (for most people) almost to the point of ridiculousness. Most people's lives aren't like this (even in computer science). It's not an inspiring start to the book.
- In the first 3 chapters 'very little' happens. I think this needs some significant condensing (or better, replacing with a chapter that covers fundamentals, gives some sense of supervised vs. unsupervised learning, and talks about how to measure similarity). As it is I had to wait to page 93 before we got started on ML proper.
- The book is crying out for some diagrams. Of the few that are used, some could have easily been better. The important diagram illustrating the “five tribes” is neither intuitive nor informative. And why, when illustrating SVMs, would you not show the margin in the diagram? – Thus immediately giving an intuitive idea of the role of the support vectors and also differentiating SVM from the simpler classifier diagram.
- There are places in the book where, if you're anything like me you'll be tearing your hair out for want of an explanation. In describing how nearest neighbour works, simply (and only) saying that: “It consists of doing exactly nothing” - is incredibly unhelpful, especially when the next few pages go into the algorithm in more detail, in the absence of a foundational explanation of how it works.
- Whilst in general, I liked the writing style, at times I found phrases and expressions used by the author to be pointlessly obscure – thus confusing the reader further, rather than clarifying. Two examples: i) the heading “One if by land, two if by Internet” is probably baffling to the majority of people (who won't be aware of the original American War of Independence phrase it's derived from---and especially so for readers outside of the US) and so it doesn't help bring focus, or coherence or clarity to the subsequent text that it introduces. ii) “As Isaiah Berlin memorably noted, some thinkers are foxes---they know many small things---and some are hedgehogs---they know one big thing”. Metaphors are a great way of using an analogy to re-frame a difficult idea in familiar terms. They can thus render something complex or intractable as immediately intuitive. This doesn't work if the chosen metaphor is as unfamiliar or obscure as the concept it is supposed to explain!
- “S curve” ...this term used throughout the whole book. In a book at this level, what's wrong with just calling it a sigmoid?
- “We routinely learn MLNs...” ---It's a trivial point and I imagine I'll be accused of being a grammar pedant, but this occurs more than once in the book and I am sure other people will cringe at it. The correct word, if you don't want to use “teach”, is surely “train” ( or alternatively rearrange along the lines of “we let the algorithms learn” or “the algorithms learned”).
- Chapter 9. i) I wasn't particularly enthralled by the storytelling/fable approach the author used here. That's just my personal view---others may like it. ii) What particularly bothered me here was that suddenly it seemed that the book was no longer a general introduction/overview of machine learning and more a way to promote the author's own area of research. There is nothing wrong with an author presenting a partisan perspective, but I would have liked to have had a better sense that this was what was happening from the start, as the book till then had seemed like a (relatively) impartial broad coverage of the field (more experienced readers may have a different take on this). iii) The failure to convey an intuitive sense or simple understanding of the various algorithms in earlier chapters of the book meant that when they all were all bundled together in one chapter, I became totally lost. For me this chapter was a total train wreck.
- Chapter 10 presents an interesting but somewhat rose-tinted view of an ML future. The author is clearly much better informed than I, in this area, but I still think that this could have been a little bit more 'balanced'. I enjoyed the ideas, but the whole chapter neatly sidesteps the important “elephant in the room”--- that advanced MLs are inevitably likely to end up in the hands of the 'most powerful' (at least initially) – rather than the 'most benevolent' or 'philanthropic'. I also felt that there were a few occasions in this chapter where he was presenting his personal opinion as fact-–a personal bugbear of mine. Examples i)…. technology develops as an 'S-curve' (aaargh) rather than exponentially, ii) ...that we are at the end of Moore's law (again).
- I particularly liked the adapted quotes presented in this chapter and the subsequent epilogue: “any sufficiently advanced AI is indistinguishable from god” and “the unexamined future is not worth inventing” ... nice touches (but only in those cases where it doesn't confuse, if the reader doesn't 'get' the reference!)
- The book has a good reading list at the back (albeit with a few surprising absences), however, frustratingly the book does not cite sources in the main text for when specific topics (scientific work) are being discussed.
- The index is excellent.
It isn’t a technical textbook, so it doesn’t go into deep detail, but it can get a little too abstract, or sometimes some mathematical terms are explained in such a superficial way, that if you didn’t know them, you won’t understand his explanations, e.g. Markov Chains, Bayes Theorem, etc.
However, you get a general idea of the main differences between the five styles in machine learning, and the ways their algorithms try to find solutions to different problems. You won’t be able to program a neural network after reading the book, though!
The main point of the book is, like the title says, to find a Master Algorithm, since in his opinion, those 5 styles of Machine Learning are not enough to solve all the problems. They have a field where they shine, but they are not general solution finders. He proposes to combine the 5 styles to create one final style that will solve any problem. I was a little disappointed he didn’t propose a total new paradigm, but instead he proposes to combine and patch different parts of the other 5 styles, like a Frankenstein monster that will do the job.
I enjoyed author's jokes and small little stories it does enrich and fresh discussion about not always super cool topics (yes I speak about Bayesian). Overall I will recommend it to advance AI practitioners who want to design and share next episode of AI understanding world.
I will copy one of great thought from this book.
"People worry that computers will get too smart and take over the world, but the real problem is that they are too stupid and they have already taken over the world."
This is solely my impression, because author sometimes goes into great details. Nevertheless,amazing summary about current stage of AI.
Someone who already has a basic grasp of some of these concepts and wishes to read a book that gives a broad overview of the field ought to get a lot from it. But as a popular science book that a lay person can pick up, understand and enjoy? I doubt it.