| Title | : | The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World |
| Author | : | |
| Rating | : | |
| ISBN | : | 0465065708 |
| ISBN-10 | : | 9780465065707 |
| Format Type | : | Hardcover |
| Number of Pages | : | 352 |
| Publication | : | First published September 8, 2015 |
In the world's top research labs and universities, the race is on to invent the ultimate learning algorithm: one capable of discovering any knowledge from data, and doing anything we want, before we even ask. In The Master Algorithm, Pedro Domingos lifts the veil to give us a peek inside the learning machines that power Google, Amazon, and your smartphone. He assembles a blueprint for the future universal learner--the Master Algorithm--and discusses what it will mean for business, science, and society. If data-ism is today's philosophy, this book is its bible.
The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World Reviews
-
Review moved to
gwern.net. -
I am really struggling to remember a book that has irritated me as much as this one, which is a shame because it's on a very interesting and significant subject. Pedro Domingos takes us into the world of computer programs that solve problems through learning, exploring everything from back propagating neural networks to Bayesian algorithms, looking for the direction in which we might spot the computing equivalent of the theory of everything, the master algorithm that can do pretty much anything that can be done with a computer (Turing proved a long time ago that there will always be some things that can't). As the subtitle puts it, this is the quest for the ultimate learning machine that will remake our world.
So far, so good. Not only an interesting subject but one I have a personal interest in as I had some involvement in artificial intelligence many moons ago. But just reading the prologue put my hackles up. It was one of those descriptions of how a technology influences every moment of your life, as the author takes us through a typical day. Except 90% of his examples have only ever been experienced by a Silicon Valley geek, and those that the rest of us have come across, like algorithms to make recommendations to you on shopping websites and video streaming sites, in my experience, are always so terrible that they are almost funny.
The pain carries on in part because of a kind of messianic fervour for the topic that means that the author seems convinced it is about to totally takeover the world - and like most fanatics, he presents this view while viciously attacking everyone who disagrees, from the likes of Marvin Minsky and Noam Chomsky to Black Swan author Nassim Nicholas Taleb. It's interesting that Domingos is totally dismissive of the early knowledge engineers who thought their methodology would take over the world, but can't see that his own pursuit of the 'master algorithm' (think of Lord of the Rings, but substitute 'algorithm' for 'ring') is equally likely to be a pursuit that is much easier to theorise about than to bring to success.
To make matters worse, Domingos repeatedly claims, for instance, that thanks to learning algorithms it's possible to predict the movement of the stock market, or to predict the kind of 'black swan' events that Taleb shows so convincingly are unpredictable. Yet I have never seen any evidence that this is true, it seems to go totally against what we know from chaos theory, and Domingos doesn't present any evidence, he just states it as fact. (Could you really have predicted the existence of black swans before they were discovered? How about blue ones?)
One other problem I have with the book is that the author isn't very good at explaining the complexities he is dealing with. I've seen many explanations of Bayesian statistics over the years, for instance, and this was one of the most impenetrable I've ever seen.
I can't tell you to avoid this book, because I've not come across another that introduces the whole range of machine learning options in the way that Domingos does. But any recommendation has to be made through gritted teeth because I did not like the way that information was put across. -
Remembering Sophia...
While reading this book, the image of Sophia was constantly assaulting my mental screen!
I'm sure most of you remember that gorgeous social robot, that could blink, smile, raise an eyebrow, and was capable of 59 more facial expressions.
She has been a media darling, showing herself in magazines, newspapers, tv news, talk shows,... spreading her charm all over the world.
She could handle a clever conversation, make eye contact and even show some sense of humour!
Sophia dreams about helping humanity building a better world, and trembles with the thought of taking a shower 😜
She’s part of a project of genious-androids -- highly intelligent robots, whose main skill is solving all those mind cracking problems human brains find so hard to unravel!
According to their creators, in something like twentie years time, androids will be able to perform most of human jobs.
Scary?!...
My one and only fear is, by the time it happens, the ones who'll be ending like robots will be us, common mortals!
HeHe!... Just kidding!
Robots will always need us around! I'm sure they will demonstrate great abilities as collaborators, but couldn't possibly 100% substitute us.
The creation never surpasses its Master!
Since I'm crazy about happy endings, I have a strong feeling this Humanoids Story will grow into a happily ever after -- Sophia's dream will come true and we'll all live and share a better world.
Can't get worse, right?!...
Androids intelligence is possible, thanks to machine learning -- this book's core.
Hence, if you are curious about knowing all possible scopes of machine learning, when applied to human society, this book is for you 👍
I'm ending this review with a video, where you'll be watching a conversation between Sophia and Jimmy Fallon, on Tonight's Show. It’s really amazing:
https://m.youtube.com/watch?v=Bg_tJvC... -
A Era dos Humanóides
Que sensação experimentariam, se algum dia entrassem de urgência num hospital, para uma intervenção cirúrgica, e dessem de caras com um robot cirurgião?!
E se numa ida ao tribunal, a ocupar a cadeira do respeitável juiz, se encontrasse um humanóide?!
Para já, tais eventos ainda figuram na lista dos incríveis impossíveis, mas...não estaremos a encaminhar-nos para aí ?!...
A verdade é que já se conhecem softwares capazes de diagnosticar doenças e robots que colaboram em cirurgias; estes são braços e mãos capazes dum alcance e precisão inatingíveis por um ser humano. Porém, carecem de autonomia, pois ainda são manipulados por humanos!
E também já há por aí um robot advogado — o famoso DoNotPay, cuja especialidade consiste em anular multas de estacionamento, e que já venceu uns bons milhares de casos.
E não podemos olvidar o projecto de robots geniais da Hanson Robotics, que foram especialmente criados para ensinar, entreter e servir, e assim cooperar com os humanos num projeto “Mundo Melhor”!
As expectativas apontam para que dentro de dezenas de anos, muitas das funções desempenhadas por humanos, passem a sê-lo por robots.
Pois!... Ao que parece, a grande Revolução da Inteligência Artificial, está prestes a rebentar!
Para já, temos a semente, mas ainda há por aí algumas cartas para dar!...
E o seu sucesso depende directamente do Algoritmo Mestre — o super herói deste livro! 😉
Para conhecê-lo de perto, só lendo "a revolução do algoritmo mestre" — este mesmo livro, claro!...
“Todo o conhecimento — passado, presente e futuro — pode ser deduzido de dados por um único algoritmo de aprendizagem universal.
Eu chamo a este algoritmo o Algoritmo-Mestre.”
Este fabuloso algoritmo é capaz de aprender o que quer que seja, a partir dos dados fornecidos:
“Tudo o que temos de fazer é fornecer-lhe a quantidade suficiente do tipo certo de dados, e ele há de descobrir o conhecimento correspondente. Se lhe dermos um vídeo, ele aprende a ver. Se lhe dermos uma biblioteca, ele aprende a ler. Se lhe dermos os resultados de experiências de física, ele descobre as leis da física. Se lhe dermos dados de cristalografia do ADN, ele descobre a estrutura do ADN.”
Preparem-se para uma vida de papo para o ar!!!... 😉👍 -
This review is a combination of 3- and 5 star reviews, so on average a 4 star rating.
I give these two ratings depending on who is reading this review. If you are a total novice in the world of computer science, or science in general for that matter, then this will likely be a 5 star book. It does a great job of introducing not just the concepts in machine learning, but also statistical ideas like variance, over-fitting, and even principal components. The key word there however is concepts. If you are a person with some programming or scientific research experience, such as myself, then you’ll likely find this book incredibly frustrating, though eventually rewarding.
Domingos has written The Master Algorithm as a primer to the various camps of machine learning, because, as I now know, there are many ways of approaching the concept of getting a machine to think like a human. There are for example the symbolists, who allow the computer to develop rules which it applies to a set of data to come to a conclusion, e.g. whether a voter with a certain voting history will vote Republican or Democrat. By contrast there are the connectionists, who create neural networks modelled on the brain, Bayesians who view machine learning as nothing more than another application of Bayes’ theorem, and evolutionaries who allow competing, mutated programs to duke it out in an arms race of algorithmic performance. The book excels as a top-down look at the kingdom of machine learning, and offers some interesting insights into how these various camps of thought can be combined into a titular ‘master algorithm’ capable of re-discovering the sum total of human knowledge given the raw data… and then some. The last few chapters offer a mind-expanding look at how machine learning, and a master algorithm, could fundamentally reshape society and the human race itself. Curing cancer, it turns out, would be one of the minor achievements of machine learning.
So far so good. A useful overview (likely useful even for those who have studied machine learning and perhaps have not explored beyond the walls of their camp of thought) and an authoritative look at the potential future of machine learning. So why as a scientist did I find this book frustrating?
Apart from the style of prose, which I didn’t particularly care for (count the number of books Domingos unnecessarily name-drops in the text), my major problem with the book was the lack of mathematical detail. As the book is aimed at a general audience this is to be expected, but it is immensely frustrating to see a concept explained in a hand-wavy way using a thousand words, when that same concept could be exactly explained in a few equations and sentences of explanation. After reading this book I immediately needed to read the actual papers that are discussed (and, it should be noted, are usefully acknowledged) so that I could really understand the techniques discussed.
As I said in the beginning, this won’t be a problem for the majority of readers, and if you don’t find this criticism particularly vexing then this book will likely be a 5 star read. It may well change your perception of computer science and uncover an interest that you never knew you had. However a word of caution to those with mathematical training – keep going through the waffly bits. It gets more interesting.
Much, according to Domingos, like the field of machine learning itself. -
An algorithm to quantify them, find them all, drive them into social media and tie them as customers.
Please note that I have put the original German text to the end of this review. Just if you might be interested.
The advantage of the best algorithm will lie in its autonomous development and improvement, to which nothing can catch up with. Just as a beginner has no chance against a professional with decades of training. However, the AI takes only moments or maybe days to surpass the level of a human master in more and more disciplines. If this is broken down to fractions of a second in the foreseeable future, any AI can not only absorb the theoretical, dry knowledge. But the application, research, and development alone, 24/7 and proceed forever.
Human geniuses that will influence the development are an unpredictable factor. Critical elements for the functioning of many modern technologies were only possible through single individuals. And just a few hundred people worldwide understand this one special aspect. Before the Eureka moment, decades often passed, in which an already existing concept was therefore not feasible. Or it failed because of the technical prerequisites, an unknown factor or an overlooked mistake.
An individual can really change the world in some ways. Hopefully the birth of the next century genius will take place in a country that has a decent educational system. That is an incentive for the educational systems of this world. So far, luminaries in their field, concerning the coming social influence of programmers, were nothing.
Different basic concepts, secret algorithms, and various development paths expand the possibilities. Chance and research flow into each other and a broad range of approaches are being developed. Many different varieties can be made out of the 5 basic concepts probabilistic inference, support vector machines, backpropogation, genetic algorithm and inverse deduction. There can be a breakthrough at any given time in a software architecture that puts it above all others. Or mergers from different concepts and intelligence applications that complement each other. In the right balance and combination of the strengths and weaknesses of the systems, the master algorithm is likely to be hidden.
The imitation of natural data processing systems, both hardware and software, is a cornerstone. And to compensate the weaknesses of these systems with the advantages of AI. Once an AI has understood the functioning of mind and brain, personality and emotion will be broken down into a logically comprehensible pattern. Both from the point of hardware (neurochemical processes) and software (ego, emotions). Humans are conditioned pattern recognition machines with reproducible and predictable functional mechanisms.
Misanthropic ones can look at it this way: Endeavored to mitigate damage, one must also ask the question, how long and under how many more victims people continue to make many decisions. The hitherto man-made system shows its dysfunctionality in the history and the current state of the environment. To back on a new, better horse cannot even cause such destruction in worst case scenarios. It would be just too illogical and stupid, deeply human attributes. Too low standard for an AI.
In that sense, it would be necessary for the algorithms to take people by the hand and give them an unimportant decision now and then as a reward. Otherwise, the creators should stay away from the levers of power to avoid debacles as in the past.
What literally corresponds to deus ex machina is the lack of traceability of decision-making. Due to the complexity and the many thousands of specialists who work on such entities, nothing can be deciphered anymore. In principle, it is an already self-existent and functioning AI that one just can not understand anymore. And humans push all threads, decision-making processes and infrastructures in its hands. Historically, a consistent continuation of the tradition of oracles, omnipotence and blind obedience.
This raises the question of the influences that could manipulate decision-making. Once again the quanta and all the levels that we do not understand yet. One knows that there is a great deal of barely understood topics lurking out of theoretical physics. Nevertheless, one accepts the decisions of an AI that makes inexplicable modes of calculation under unknown influences for essential decisions.
To strive for a sci-fi scenario: nothing would be more attractive than to manipulate the oracle of a civilization that has no real control over it. For better or for worse. Nobody would notice anything.
Ein Algorithmus sie zu quantifizieren, sie alle zu finden, in die sozialen Medien zu treiben und als Kunden zu binden.
Der Vorteil des besten Algorithmus wird in seiner autonomen Weiterentwicklung liegen, die man so nicht mehr einholen kann. Wie ein Anfänger einem Profi mit jahrzehntelangem Training nicht das Wasser reichen kann. Nur dass die KI nur Augenblicke oder höchstens Tage braucht, um das Niveau eines menschlichen Meisters in immer mehr Disziplinen zu übertreffen. Wenn das in absehbarer Zukunft auf Sekundenbruchteile herab gebrochen ist, kann jede KI nicht nur das theoretische, trockene Wissen absorbieren. Sondern die Anwendung, Forschung und Weiterentwicklung alleine, 24/7 und für immer voran treiben.
Auch menschliche Genies, die die Entwicklung noch beeinflussen werden, sind ein unvorhersehbarer Faktor. Wurden doch Schlüsselelemente für das Funktionieren vieler moderner Technologien nur durch Einzelpersonen möglich. Und nur ein paar Hundert Menschen weltweit verstehen diesen Teilaspekt. Vor dem Heureka Moment vergingen oft Jahrzehnte, in denen ein bereits vorhandenes Konzept deswegen nicht umsetzbar war. Oder es scheiterte an den technischen Vorraussetzungen, einem unbekannten Faktor oder einem übersehenen Fehler.
Ein einzelner kann in mancherlei Hinsicht wirklich die Welt verändern. Es steht zu hoffen, dass die Geburt der nächsten Jahrhundertgenies in einem Land statt findet, dass ein seiner würdiges Bildungssystem hat. Das ist einmal ein Anreiz für die Bildungssysteme dieser Welt. Bisher waren Koryphäen auf ihrem Gebiet, in Relation zu dem kommenden gesellschaftlichen Einfluss von Programmierern, nichts.
Verschiedene Grundkonzepte, geheime Algorithmen und unterschiedliche Entwicklungswege erweitern die Möglichkeiten. Zufall und Forschung fließen ineinander über und ein breiter Reigen an Ansätzen wird immer weiter entwickelt. Es kann jederzeit bei einer Softwarearchitektur einen Durchbruch geben, der sie über alle anderen erhebt. Oder Fusionen aus verschiedenen Konzepten und Intelligenzanwendungen, die sich gegenseitig ergänzen. In der richtigen Balance und Kombination der Stärken und Schwächen der Systeme dürfte der Meisteralgorithmus verborgen sein.
Die Imitation natürlicher Datenverarbeitungssysteme sowohl bei Hardware als auch Software ist ein Grundpfeiler. Und die Schwächen dieser Systeme mit den Vorteilen der KI zu kompensieren. Hat eine KI erst die Funktionsweisen von Geist und Hirn verstanden, wird auch Persönlichkeit und Emotion auf ein logisch nachvollziehbares Muster herunter gebrochen werden können. Sowohl von der Hardware (neurochemische Prozesse) als auch von der Software (Ego, Emotionen) her. Menschen sind konditionierte Mustererkennungsmaschinen mit reproduzierbaren und vorhersagbaren Funktionsmechanismen.
Misantrophisch kann man es so betrachten: Lapidar und auf Schadensbegrenzung bemüht muss man sich auch die Frage stellen, wie lange und unter wie vielen Opfern weiterhin Menschen viele Entscheidungen treffen sollen. Das bisher von Menschen geschaffene System zeigt an der Geschichte und dem aktuellen Zustand der Umwelt seine Dysfunktionalität. Auf ein neues, besseres Pferd zu setzen, kann nicht einmal in Worst Case Szenarien derartige Zerstörungen anrichten. Es wäre schlicht zu unlogisch und dumm, zutiefst menschliche Attribute.
Insofern täte es Not, dass die Algorithmen die Menschen bei der Hand nehmen und ihnen hin und wieder als Belohnung eine unwichtige Entscheidung spendieren. Ansonsten sollten die Erschaffer aber weit entfernt von den Schalthebeln der Macht bleiben, um Debakel wie in der Vergangenheit zu vermeiden.
Was wortwörtlich deus ex machina entspricht, ist die nicht vorhandene Rückverfolgbarkeit der Entscheidungsfindung. Aufgrund der Komplexität, der vielen Tausend Spezialisten, die an solchen Entitäten arbeiten, lässt sich nichts mehr dechiffrieren. Es ist im Prinzip eine jetzt schon autark existierende und funktionierende KI, die man schlichtweg nicht mehr verstehen kann. Und der man alle Fäden, Entscheidungsprozesse und Infrastrukturen in die Hand drückt. Geschichtlich betrachtet eine konsequente Fortsetzung der Tradition von Orakeln, Allmachtsansprüchen und blindem Gehorsam.
Das wirft die Frage nach den Einflüssen auf, die die Entscheidungsfindung manipulieren können. Wieder einmal die Quanten und all die Ebenen, die wir noch nicht verstehen. Man weiß, dass da sehr viel sehr Unverstandenes in der theoretischen Physik lauert. Trotzdem akzeptiert man die Entscheidungen einer KI, die auf unerklärlichen Wegen unter unbekannten Einflüssen essentielle Berechnungen anfertigt.
Um ein Sci Fi Szenario zu bemühen: Nichts wäre attraktiver als das Orakel einer Zivilisation, dass dieses selbst nicht unter Kontrolle hat, zu manipulieren. Zum Guten oder zum Schlechten. Niemand würde etwas merken. -
The author states that "intuition is what you use when you don't have enough data". The author will show heuristically how intuition is slowly being taken out of analyzing big data and being replaced with algorithms which teach themselves how to make the data speak for themselves. "All learning starts with some knowledge" (a quote from Hume, that the author invokes), and from Hume we know that there is a problem with induction, no matter what the particular can not prove the universal. The trick is to get from the data (the particular) to the universal and the author explains in detail the five general ways we learn and shows how they work in practice. The five ways are Symbolic (think: rational thought), Connective (modeling like the Network in the brain), Bayesian (nothing is certain and all is contingent), Evolutionary (see "The Selfish Gene" by Dawkins), and by Analogy.
The key is to use some variations of the ways ('tribes') and have the method (algorithm) use the data to exploit the information that is within the data set and do it recursively (and as Douglas Hofstadter says "I am a Strange Loop"). The computers are becoming faster, cheaper and can manipulate ever larger and more easily accessible data sets, and the methods have become more refined and usable. For example, brute force Bayesian methods are not used since the whole decision tree necessary for learning complex solutions are never practical and are now replaced by naive Bayesian techniques (only some of the dependent states need to be computed) giving only a small loss in overall accuracy.
The overall point of the book is to show that there is evolutionary thinking going on in writing smart algorithms which are able to let the data speak for themselves and the computer scientists have a tool box of techniques which enable real objective knowledge to be extracted from the data.
I like the TV show Person of Interest. Everything that "The Machine" does on that show can be explained by the techniques discussed in this book. This author doesn't think the computer will ever be able to think or have its own "will". I think this book would be an excellent lead in to the Nick Bostrom book "Superintelligence: Paths, Dangers and Strategies". That book does think super AI will happen and a computer will develop a 'will'. This book, "Master Algorithm" is an excellent primer for someone who believes the "singularity is near" even though the author disagrees (It's odd this author thinks the super AI is not possible because the way he starts off the book by explaining the P=NP problem and how solving that could create a master algorithm which in my way of thinking would lead to a super AI). -
Um livro bem denso que usa a inteligência artificial como contexto para explicar algoritmos. Passando pelos vários tipos de algoritmos de machine learning: conexionistas, bayesianos, evolutivos, simbolistas e analogistas.
Pesado, mas didático e bem humorado. Não recomendo para o público geral. Recomendo muito para quem já tem um background em matemática ou ciências em geral, ou quem realmente quer entender como funcionam os métodos de análise de big data atuais, porque depende de vários conceitos prévios. E recomendo especialmente para quem trabalha com bioinformática, pois achei a didática do autor sensacional para explicar métodos de análise como bayesiana, MCMC, cadeias de Markov, scale vector e afins. Talvez sirva para várias outras áreas com as quais não sou familiar. E mesmo para quem já tem um bom conhecimento na área, pode ser relevante, dado o tanto que o autor conhece, explica e relaciona.
Bastante atual, com vários exemplos interessantes e relevantes de companhias e problemas onde os algoritmos são aplicados. Além de um escopo muito maior do que só algoritmos, a discussão é bem situada no presente. E a discussão final sobre o que fazemos com nossos dados pessoais e como (possivelmente) lidar com a relação perda de privacidade/ganho de personalização é bastante sóbria e relevante. Me lembra um pouco a densidade e compreensão do autor do
Superintelligence: Paths, Dangers, Strategies.
Só cansei de ler "cura do câncer" como o exemplo de objetivo. -
This book is filled with complex computer science concepts. I got halfway before getting lost in most of the content. It is hard to grasp unless you've studied computer science. Essentially, this book is about the quest to create 'The Master Algorithm,' which can create all other algorithms. There won't be a need for a human to create specific applications for each desired need, like an app that is only capable of playing chess. The Master Algorithm will be able to learn how to create other apps. An analogy to the Master Algorithm is the hand, which can use tools to create everything else that humans want.
-
Confesso que parti para "The Master Algorithm" (2015) com várias reservas: a primeira prendia-se com a dificuldade de trazer um assunto desta complexidade para uma discussão leiga; a segunda tinha que ver com a minha desconfiança sobre a possibilidade efetiva de se criar um algoritmo único, de tudo capaz. No final do livro tenho de dizer que Pedro Domingos, professor na Universidade de Washington, fez um belíssimo trabalho, não só o livro é acessível como nos abre o apetite para o tema. O que mais gostei, e acaba sendo o cerne do livro, foi da descrição das metodologias que estão a ser seguidas para que a máquina possa aprender, não por serem exóticas mas antes pelo contrário, por responderem por métodos que nós próprios, humanos, também temos vindo a utilizar para construir conhecimento.
Continua no VI em:
https://virtual-illusion.blogspot.com... -
Inane verbiage with no educational content. Just constant fawning and endless lists of potential and current applications of learning algorithms. The writing is beyond tiresome. Here's just one paragraph:
"You’ve reached the final stage of your quest. You knock on the door of the Tower of Support Vectors. A menacing-looking guard opens it, and you suddenly realize that you don’t know the password. “Kernel,” you blurt out, trying to keep the panic from your voice. The guard bows and steps aside. Regaining your composure, you step in, mentally kicking yourself for your carelessness. The entire ground floor of the tower is taken up by a lavishly appointed circular chamber, with what seems to be a marble representation of an SVM occupying pride of place at the center. As you walk around it, you notice a door on the far side. It must lead to the central tower—the Tower of the Master Algorithm. The door seems unguarded. You decide to take a shortcut. Slipping through the doorway, you walk down a short corridor and find yourself in an even larger pentagonal chamber, with a door in each wall. In the center, a spiral staircase rises as high as the eye can see. You hear voices above and duck into the doorway opposite. This one leads to the Tower of Neural Networks. Once again you’re in a circular chamber, this one with a sculpture of a multilayer perceptron as the centerpiece. Its parts are different from the SVM’s, but their arrangement is remarkably similar. Suddenly you see it: an SVM is just a multilayer perceptron with a hidden layer composed of kernels instead of S curves and an output that’s a linear combination instead of another S curve."
Had enough yet? There are chapters full of this drivel. This is actually representative even of parts of the book that aren't an acid trip. Just like the rest of the book this passage mentions concepts without explaining any of them and mixes everything together without any reason or structure. Who is this book for? There are in-jokes dotted around like the Jennifer Aniston one or concepts like centaurs (in relation to chess) and neither they nor their backstories are ever explained yet it's clearly aimed at the general reader. -
Good overview of machine learning. The master algorithm seems like an overwhelming concept at first, but the book is very accessible for anyone who has a basic education in comp sci. However it's pretty clear that the intended audience is nerdy high school boys considering a career in machine learning, despite the author saying this book is written for everyone/anyone.
Also, I can't help but find his depiction of the post-master algorithm world creepy. Call me old fashioned. -
Most of this book was great because it read like a short summary of what is taught in an introduction to cognitive science class. While spouting about how Bayesian stats decidedly kick frequentist stat's ass, to which I agree, the author showed how to look at the world itself, and everything in it, through a more Bayesian lens. He hammered home the central point that nothing can be understood in isolation and must rather be understood through its connection to the things around it. One more book on my list that belongs on the complexity/emergence/network shelf. Any book in that category, if it's any good at all, will be among my favorites. Adding to that, this author did a great job of covering one of my favorite subjects, Markov chains.
Every time I dictate into my phone and realize that Apple has corrected "thus" to DOS and can recognize any computer programing term but not the science terms I dictate, I laugh to myself at the coders who create the Markov chains. They are seriously biased in favor of programmers. But in the end, that is because they are programmers:)
This book does an exceptional job of explaining Bayes, providing a brief history of how it came to be used, and showing how it is at work naturally, all around us-- neurons, self driving cars, electrons, etc.
In my opinion, he should have stopped before writing the last chapter. He basically ruined his beautiful book by merging it with the Selfish Gene theory and going on a rant about how humans are special. Does he know that any and every time we humans thought we were special, it turned out we were wrong? For some reason, he thinks humans will always control AI. He doesn't seem to be of the opinion that AI itself will be a mashup of human and machine, with no differentiation. He also went on to trash any advances that arise from Moore's Law, which he said was on it's last leg. Tell that to Stephen Hawking who just used tech derived from Moore's Law to create AI nanoships that will literally sail to Alpha Centauri to search for life on the earth-like planet in that system. -
From a content standpoint, The Master Algorithm by Pedro Domingos is a great crash course for anyone who is interested in learning more about machine learning. But from an “ease of comprehension” standpoint, this book is far from the layman’s journey that Domingos claims it is.
I found myself able to follow roughly 70% or so of the technical content of this book, and there were definitely some times that it was a bit too technical for me to completely grasp what he was trying to say. The writing at times can become circular in nature and to use a nerd joke… feels like a bit of a “circular reference” J. Based on these two components (content vs. comprehension), I believe that the Master Algorithm is a 2.5 to 3.0 star book that is worth reading if you are remotely interested in A.I., machine learning, and computer science. You just have to be comfortable with some of the complex concepts that he covers (maybe that is just me)
More about the book itself… The main point that I grabbed onto and took away is that there are five distinct methods/tribes for machine learning that will have to come together to create the Master Algorithm. The five tribes/methods are as follows:
1. Symbologists (inverse deduction);
2. Connectionists (backpropogation);
3. Evoluntionaries (genetic algorithms);
4. Bayesians (probablistic inference);
5. Analogizers (support vector machines))
Each of these five tribes has a piece of the puzzle that will ultimately help to create the Master Algorithm. And with this algorithm, true and grand machine learning will be able to flourish in a way that we have yet to see. The Master Algorithm is the unifier of machine learning: it lets any application use any learner, by abstracting the learners into a common form that is all the applications need to know.
But as Domingos states, the future of machine learning, although bright, is very much dependent on cracking the code of this Master Algorithm. Domingos implores each of us, novice and expert alike, to play our part in the development of this future and I suggest anyone interested in this material go to the websites that are in the book and take a look. There is some pretty cool stuff out there! -
Bill Gates put this book on his list of recommended reads this year. It interested me because of my work on an ed tech tool to help young adults read better and enjoy their reading experience more. To get the tool right, we have had to integrate artificial intelligence and to make it better we will need machine learning. We have experts on the team for that, but I like to know what's going on around me. I can't pretend I now fully understand machine learning, but Domingos did an excellent job surveying the field so I could get a gist of it.
He lays the foundations for machine learning and then identifies 5 approaches that are being pursued now and details the formal languages they rely on:
Symbolists- Logic
Connectionists- Neural Networks
Evolutionaries- Genetic Programs
Bayesians- Graphical Modeling
Analogizers- Specific Instances
Throughout these explanations he explores the challenges:
Complexity monster
Overfitting problem
Curse of dimensionality
Exploration/Exploitation dilemma
My background in educational philosophy and epistemology helped me follow along with ease and enthusiasm as he discussed the concepts that were related to learning theories, but I could only just barely hang on for the rides through mathematical foundations.
Points of interest:
Machine learning helped Obama's campaign make strategic ad buying decisions in 2012.
Machine learning's version of the nature v nurture debate is whether or not the brain or evolution are better models for the master algorithm.
Mastering Tetris is a great step towards solving 1000s of problems because it is a basic NP completeness problem. -
While coached as a guidebook to help find "The Master Algorithm," the one AI algorithm "that will rule them all" (his words, not mine), this book is much, much more. At times written whimsically, and at times treating very advance material in a way that non-sophisticated readers can understand, the book is part history lesson, part cultural commentary, and part description of the scientific process. I work exactly in the field of Artificial Intelligence and Machine Learning, and I am definitely a member of the target audience. Maybe I'm just an ML fanboy, but I found the treatment of everything, including that of the author's own work, to be at just the right level to keep the non-specialist interested, while informing the specialist about those other areas, where he is not a specialist.
The book finishes with musings about what it means to be a source of data for the corporations (and governments) that would use that data for good and/or evil. This alone makes the book worth reading. Such candor from those in the field of machine learning is really refreshing. -
L'intelligenza artificiale è un bel paradosso: ne usufruiamo quotidianamente (spesso senza esserne consapevoli), eppure per certi versi ci fa paura. D'altra parte l’autonomia decisionale di una macchina effettivamente genera interrogativi importanti. L'algoritmo definitivo è, per dirlo in due parole, un processo attraverso cui le macchine potranno imparare fondamentalmente come gli studenti imparano dall’insegnante. E si tratterà di studenti molto intelligenti, dato che non bisognerà nemmeno spiegare ma basterà fornir loro degli esempi. Impareranno osservandoci, come i bambini imparano osservando gli adulti. E poi ci saranno macchine che impareranno da altre macchine, aggiungendo conoscenza a conoscenza.
Domingos ha dedicato tutta la sua vita a questo settore della scienza, la sua competenza è evidente.
Però forse pecca di tecnicismo, dato che su un argomento di questa portata mi sarei aspettato, oltre all'aspetto puramente scientifico, qualcosa di più in tema etico e sociale. Mi sono posto molte domande che sono rimaste senza risposta. Ad esempio: quando l'"algoritmo" sarà disponibile, chi lo controllerà? Quali individui ne saranno i detentori (e quali le vittime)? In che modo l’acquisizione e la gestione di questa conoscenza, e l’inevitabile invasione della vita pubblica e privata che ne deriverà, modificherà l’idea che abbiamo della nostra società, dei nostri figli e di noi stessi? Come evitare che il divario fra paesi ricchi e paesi meno ricchi venga ulteriormente incrementato, visto che l'algoritmo sarà appannaggio di pochi?
Domande su cui l'autore glissa in larga misura. -
Cos'è l'intelligenza artificiale? può essere tutto o niente. Già il nome stesso è in un certo senso sbagliato, perché richiama un'idea degli anni 1950 che ormai è completamente superato. Il guaio è che ci sono cinque approcci fondamentali diversi secondo l'autore Pedro Domingos, e finché non si troverà un modo per riunificarli - l'Algoritmo Definitivo del titolo - non si potrà sperare in un vero avanzamento. Questo libro è molto americaneggiante: uno come me che tende a essere sin troppo laconico mi sono spesso un po' scocciato dalle ripetizioni, e fosse per me avrei eliminato del tutto il capitolo 2 con le speranze riposte nell'Algoritmo Definitivo e il capitolo 9 con gli scenari previsti. Ma la parte centrale, con la spiegazione dettagliata delle cinque famiglie di algoritmi attualmente studiati, è una preziosissima risorsa non solo per chi vuole conoscere lo stato dell'arte nell'AI ma anche per imparare a conoscere i vantaggi e gli svantaggi, anche per l'attenta traduzione di Andrea Migliori.
-
"The statistician knows that prediction is hard, especially about the future, and the computer scientist knows that the best way to predict the future is to invent it, but the unexamined future is not worth inventing."
"... The greatest benefit of machine learning may ultimately be not what the machines learn but what we learn by teaching them."
Although I didn't agree with all of the points about the present and future of AI in this book and there were a lot of fanciful metaphors being tossed around, this was a very enjoyable read.
Taking computer science classes about machine learning means I'm knee deep in the math and code but this allowed me to take a step back and look at AI from a greater perspective. The author explains a few common algorithms so it's also good as a tour through ML for laypeople.
There just aren't enough books out there about this subject written in a non-textbook fashion. -
Bazı temel bakış açılarını ve yaklaşımları vermesi açısından güzel bir kitaptı.
Ama tamamen yeni bir master algoritma geliştirmeye ve bunun nasıl yapılabileceğine odaklı bir kitap.
Machine Learning'le yeni algoritmalar geliştirme seviyesinde ilgilenenler için ilgi çekici olabilir.
Machine Learning'i kullanmak isteyenler için ise her ne kadar bazı bölümleri yararlı olsa da geneli kafa karıştırıcı ve gereksiz olarak düşünülebilir. -
I definitely enjoyed and appreciated this a lot more on my second read. The key difference this time was that I've finally been digging into the machine learning world a bit and had more context on which to connect the ideas.
-
I have an interest in this topic. I would love to "crack the code" on something but this was repetitious and unreadable. If you are considering this and want more detail on my opinion, please review the other 1-star reviews as I agree with nearly all of them. Could not finish.
-
Pedro Domingos é professor de Ciências da computação na Universidade de Washington.
Formado no Instituto Superior Técnico em Lisboa, ganhou o prémio de inovação da SIGKDD, o mais prestigiado na área das ciências de dados.
Domingos termina a Revolução do Algoritmo Mestre com uma mensagem que encerra em si o grande propósito do mesmo, dar a conhecer aos leitores o significado de Inteligência Artificial.
”A aprendizagem automática toca a vida a cada um de nós e cabe-nos a todos decidir o que queremos fazer com ela. Com o entendimento sobre a aprendizagem automática, ficamos numa posição muito melhor para decidir sobre a partilha de dados, o futuro do trabalho, a guerra robotizada, as promessas e os perigos da IA; e quanto mais forem aqueles que tiverem este entendimento, mais provável será que evitemos armadilhas e encontremos os caminhos certos.”
As máquinas são criação do homem. Enquanto algumas pessoas ficam deslumbradas com a Inteligência Artificial das máquinas (por exemplo: eu), outras pessoas se sentem atemorizadas achando que as máquinas irão ter capacidades subversivas e atacar o homem, e até mesmo extinguir os humanos do planeta, tal como NÓS humano extinguimos algumas espécies de animais que coabitavam connosco.
Nada disso: “As hipóteses de uma IA equipada com o Algoritmo Mestre assumir o controlo do mundo são nulas. O motivo é simples: ao contrário dos seres humanos os computadores não têm vontade própria. São produtos da engenharia e não da evolução. Mesmo um computador infinitamente potente não seria mais do que uma extensão da nossa vontade, não sendo, nada a recear. Os Algoritmos de aprendizagem são compostos por três componentes: representação, avaliação e otimização. A representação circunscreve o que ele pode aprender, a otimização faz tudo ao seu alcance para otimizar a aprendizagem – nem mais, nem menos -maximiza a função de avaliação. A função da avaliação é determinada por nós. Um computador mais potente apenas irá otimiza-la melhor. Não existe o risco de perder o controlo, mesmo que seja um algoritmo genético.”
Frequentemente ouvimos falar em “erros informáticos”, pois bem… isso não existe… o que existe são erros humanos, e estes, claro, culpam as máquinas por fazerem o trabalho errado, mas elas apenas fazem o que lhes mandam, e quem manda somos nós, quem cria e implementa os algoritmos são os humanos.
Impressionante a quantidade de citações que retirei do livro, deixo apenas algumas, as que considero mais relevantes:
84.17% "A empresa Narrative Science possui um sistema de IA que é capaz de escrever resumos bastante bons de um jogo de basebol, mas não romances, porque - lamento, George F. Will - a vida é muito mais que um jogo de basebol.
O senso comum é importante não só porque a nossa mãe assim nos ensinou, mas porque os computadores não o têm."
Com a rápida e crescente evolução dos tradutores automáticos, cada vez mais eficientes, facilmente adquirimos na Internet qualquer livro, em qualquer língua, que os sistemas conseguirão traduzir para a nossa língua em poucos minutos.
72.5% "Á semelhança da memória humana, a aprendizagem relacional tece uma abundante rede de associações. Ela conecta objetos percecionados...
A aprendizagem relacional é a última peça do puzzle, o ingrediente final que nos falta para a nossa alquimia, o Algoritmo-Mestre".
70.83% "Se equiparmos o robô Robby com todas as capacidades de aprendizagem, que já vimos neste livro, ele será bastante inteligente, mas ainda um pouco autista. Há de ver o mundo como uma série de objetos separados, que ele pode identificar, manipular, e sobre os quais pode até fazer previsões, mas não perceberá que o mundo é uma rede de interconexões."
45.0% "O cérebro consegue aprender qualquer coisa, mas não consegue desenvolver um cérebro. Se compreendêssemos perfeitamente a sua arquitetura, poderíamos simplesmente implementá-la em hardware, mas estamos muito longe disso."
24.72% "Aristóteles afirmou que não existe nada no intelecto que não existisse primeiro nos sentidos. Leibniz acrescentou: «Exceto o próprio intelecto.» O cérebro humano não é uma folha em branco porque não é uma folha. Uma folha é algo passiva, algo no qual escrevemos, mas o cérebro processa ativamente a informação que recebe. A memória é a folha no qual ela escreve, e sim, começa por estar em branco."
19.44% "...no nosso trabalho, o que pode ser feito por um algoritmo de aprendizagem, e o que é que não pode, e - o mais importante - como é que posso tirar partido da aprendizagem automática para o fazer melhor? O computador é a nossa ferramenta, não o nosso adversário,"
9.44% "A Revolução Industrial automatizou o trabalho manual e a Revolução da Informação fez o mesmo pelo trabalho mental, mas a aprendizagem automática automatiza a própria automatização. Sem ela, os programadores tornam-se o estrangulamento que atrasa o progresso. Com ela, o ritmo do progresso aumenta.
... a revolução da aprendizagem automática causará extensas alterações económicas e sociais." Estima-se que daqui por 10 anos a IA seja vulgar nas nossas vidas: na nossa casa, no nosso trabalho e no lazer, tal como a eletricidade, a IA será imprescindível para nós.
October 25, 2017 – 0.0% "Qual é o melhor modelo para o Algoritmo-Mestre: a evolução ou o cérebro? Esta é a versão da aprendizagem automática do debate inato versus adquirido. E, do mesmo modo que a natureza e a educação se combinam para nos produzir a nós, seres humanos, talvez o verdadeiro Algoritmo-Mestre contenha elementos de ambas."
Afinal o que é o Algoritmo – Mestre?
É a consciência da máquina. (Estamos no bom caminho, mas ainda não a alcançamos)
Nota: este livro, de todo, não é de fácil leitura, é necessário que o leitor domine as ciências exatas. O livro aborda vários modelos matemáticos que estão na base da criação dos algoritmos e a linguagem matemática para quem não é da área torna o livro enfadonho e incompreensível. Talvez por isso o autor afirma ser um livro para profissionais e pessoas que adorem esta área do conhecimento.
Por fim uma nota para a escrita, na minha opinião está pouco cuidada, algumas lacunas, sei que o autor não domina a ciência da escrita, que provém de outra àrea de conhecimento, contudo falta nesta obra, duas a três leituras de aperfeiçoamento da escrita. -
Q:
A good learner is forever walking the narrow path between blindness and hallucination. (c)
No hallucination this time. Just a fun & very high-level escourse into the realm of coulda-woulda and what we have already in terms of algorithmification, generalized machine intelligence and the realms of their possible uses. -
In short it's totally inspiring :) some parts on statistics and probabilities are kind of not too active and in general math made as simple to understand as it possible. The book is much more on philosophy, computer science core concepts and people learning than on artificial intelligence and machine learning. Indeed everything to know to start and go further and no to be stuck with ML is here :) Strongly recommended! #AI
