What do you think?
Rate this book
Pelican Books #34
Artificial Intelligence: A Guide for Thinking Humans
A sweeping examination of the current state of artificial intelligence and how it is remaking our world
No recent scientific enterprise has proved as alluring, terrifying, and filled with extravagant promise and frustrating setbacks as artificial intelligence. The award-winning author Melanie Mitchell, a leading computer scientist, now reveals AI’s turbulent history and the recent spate of apparent successes, grand hopes, and emerging fears surrounding it.
In Artificial Intelligence, Mitchell turns to the most urgent questions concerning AI today: How intelligent—really—are the best AI programs? How do they work? What can they actually do, and when do they fail? How humanlike do we expect them to become, and how soon do we need to worry about them surpassing us? Along the way, she introduces the dominant models of modern AI and machine learning, describing cutting-edge AI programs, their human inventors, and the historical lines of thought underpinning recent achievements. She meets with fellow experts such as Douglas Hofstadter, the cognitive scientist and Pulitzer Prize–winning author of the modern classic Gödel, Escher, Bach, who explains why he is “terrified” about the future of AI. She explores the profound disconnect between the hype and the actual achievements in AI, providing a clear sense of what the field has accomplished and how much further it has to go.
Interweaving stories about the science of AI and the people behind it, Artificial Intelligence brims with clear-sighted, captivating, and accessible accounts of the most interesting and provocative modern work in the field, flavored with Mitchell’s humor and personal observations. This frank, lively book is an indispensable guide to understanding today’s AI, its quest for “human-level” intelligence, and its impact on the future for us all.
No recent scientific enterprise has proved as alluring, terrifying, and filled with extravagant promise and frustrating setbacks as artificial intelligence. The award-winning author Melanie Mitchell, a leading computer scientist, now reveals AI’s turbulent history and the recent spate of apparent successes, grand hopes, and emerging fears surrounding it.
In Artificial Intelligence, Mitchell turns to the most urgent questions concerning AI today: How intelligent—really—are the best AI programs? How do they work? What can they actually do, and when do they fail? How humanlike do we expect them to become, and how soon do we need to worry about them surpassing us? Along the way, she introduces the dominant models of modern AI and machine learning, describing cutting-edge AI programs, their human inventors, and the historical lines of thought underpinning recent achievements. She meets with fellow experts such as Douglas Hofstadter, the cognitive scientist and Pulitzer Prize–winning author of the modern classic Gödel, Escher, Bach, who explains why he is “terrified” about the future of AI. She explores the profound disconnect between the hype and the actual achievements in AI, providing a clear sense of what the field has accomplished and how much further it has to go.
Interweaving stories about the science of AI and the people behind it, Artificial Intelligence brims with clear-sighted, captivating, and accessible accounts of the most interesting and provocative modern work in the field, flavored with Mitchell’s humor and personal observations. This frank, lively book is an indispensable guide to understanding today’s AI, its quest for “human-level” intelligence, and its impact on the future for us all.
336 pages, Hardcover
First published October 15, 2019
About the author
Melanie Mitchell
7 books167 followersMelanie Mitchell is a professor of computer science at Portland State University. She has worked at the Santa Fe Institute and Los Alamos National Laboratory. Her major work has been in the areas of analogical reasoning, complex systems, genetic algorithms and cellular automata, and her publications in those fields are frequently cited.
She received her PhD in 1990 from the University of Michigan under Douglas Hofstadter and John Holland, for which she developed the Copycat cognitive architecture. She is the author of "Analogy-Making as Perception", essentially a book about Copycat. She has also critiqued Stephen Wolfram's A New Kind of Science and showed that genetic algorithms could find better solutions to the majority problem for one-dimensional cellular automata. She is the author of An Introduction to Genetic Algorithms, a widely known introductory book published by MIT Press in 1996. She is also author of Complexity: A Guided Tour (Oxford University Press, 2009), which won the 2010 Phi Beta Kappa Science Book Award.
She received her PhD in 1990 from the University of Michigan under Douglas Hofstadter and John Holland, for which she developed the Copycat cognitive architecture. She is the author of "Analogy-Making as Perception", essentially a book about Copycat. She has also critiqued Stephen Wolfram's A New Kind of Science and showed that genetic algorithms could find better solutions to the majority problem for one-dimensional cellular automata. She is the author of An Introduction to Genetic Algorithms, a widely known introductory book published by MIT Press in 1996. She is also author of Complexity: A Guided Tour (Oxford University Press, 2009), which won the 2010 Phi Beta Kappa Science Book Award.
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 30 of 358 reviews
October 21, 2019
As Melanie Mitchell makes plain, humans have limitations in their visual abilities, typified by optical illusions, but artificial intelligence (AI) struggles at a much deeper level with recognising what's going on in images. Similarly in some ways, the visual appearance of this book misleads. It's worryingly fat and bears the ascetic light blue cover of the Pelican series, which since my childhood have been markers of books that were worthy but have rarely been readable. This, however, is an excellent book, giving a clear picture of how many AI systems go about their business and the huge problems designers of such systems face.
Not only does Mitchell explain the main approaches clearly, her account is readable and engaging. I read a lot of popular science books, and it's rare that I keep wanting to go back to one when I'm not scheduled to be reading it - this is one of those rare examples.
We discover how AI researchers have achieved the apparently remarkable abilities of, for example, the Go champion AlphaGo, or the Jeopardy! playing Watson. In each case these systems are tightly designed for a particular purpose and arguably have no intelligence in the broad sense. As for what's probably the most impressively broad AI application of modern times, self-driving cars, Mitchell emphasises how limited they truly are. Like so many AI applications, the hype far exceeds the reality - when companies and individuals talk of self-driving cars being commonplace in a few years' time, it's quite clear that this could only be the case in a tightly controlled environment.
One example, that Mitchell explores in considerable detail are so-called adversarial attacks, a particularly AI form of hacking where, for example, those in the know can make changes to images that are invisible to the human eye but that force an AI system to interpret what they are seeing as something totally different. It's a sobering thought that, for example, by simply applying a small sticker to a stop sign on the road - unnoticeable to a human driver - an adversarial attacker can turn the sign into a speed limit sign as far as an AI system is concerned, with potentially fatal consequences.
Don't get me wrong, Mitchell, a professor of computer science who has specialised in AI, is no AI luddite. But unlike many of the enthusiasts in the field (or, for that matter, those who are terrified AI is about to take over the world), she is able to give us a realistic, balanced view, showing us just how far AI has to go to come close to the more general abilities humans make use of all the time even in simple tasks. AI does a great job, for example, in something like Siri or Google Translate or unlocking a phone with a face - but AI systems still have no concept of, for example, understanding (as opposed to recognising) what is in an image. Mitchell makes it clear that where systems learn from large amounts of data, it is usually impossible to uncover how they are making decisions (which makes the EU's law requiring transparent AI decisions pretty much impossible to implement), so we really shouldn't trust them with important outcomes as they could easily be basing their outcomes on totally irrelevant inputs.
Apart from occasionally finding the explanations of the workings of types of neural network a little hard to follow, the only thing that made me raise an eyebrow was being told that Marvin Minsky 'coined the phrase "suitcase word"' - I would have thought 'derived* the phrase from Lewis Carroll's term "portmanteau word"' would have been closer to reality.
There have been good books on the basics of AI already, and excellent ones on the problems that 'deep learning' and big data systems throw up. But without a doubt, Mitchell's book sets a new standard in giving an understanding of what's possible and how difficult it is to go further. It should be read by every journalist, PR person and politician before they pump out yet more hype on the AI future. Recommended.
Not only does Mitchell explain the main approaches clearly, her account is readable and engaging. I read a lot of popular science books, and it's rare that I keep wanting to go back to one when I'm not scheduled to be reading it - this is one of those rare examples.
We discover how AI researchers have achieved the apparently remarkable abilities of, for example, the Go champion AlphaGo, or the Jeopardy! playing Watson. In each case these systems are tightly designed for a particular purpose and arguably have no intelligence in the broad sense. As for what's probably the most impressively broad AI application of modern times, self-driving cars, Mitchell emphasises how limited they truly are. Like so many AI applications, the hype far exceeds the reality - when companies and individuals talk of self-driving cars being commonplace in a few years' time, it's quite clear that this could only be the case in a tightly controlled environment.
One example, that Mitchell explores in considerable detail are so-called adversarial attacks, a particularly AI form of hacking where, for example, those in the know can make changes to images that are invisible to the human eye but that force an AI system to interpret what they are seeing as something totally different. It's a sobering thought that, for example, by simply applying a small sticker to a stop sign on the road - unnoticeable to a human driver - an adversarial attacker can turn the sign into a speed limit sign as far as an AI system is concerned, with potentially fatal consequences.
Don't get me wrong, Mitchell, a professor of computer science who has specialised in AI, is no AI luddite. But unlike many of the enthusiasts in the field (or, for that matter, those who are terrified AI is about to take over the world), she is able to give us a realistic, balanced view, showing us just how far AI has to go to come close to the more general abilities humans make use of all the time even in simple tasks. AI does a great job, for example, in something like Siri or Google Translate or unlocking a phone with a face - but AI systems still have no concept of, for example, understanding (as opposed to recognising) what is in an image. Mitchell makes it clear that where systems learn from large amounts of data, it is usually impossible to uncover how they are making decisions (which makes the EU's law requiring transparent AI decisions pretty much impossible to implement), so we really shouldn't trust them with important outcomes as they could easily be basing their outcomes on totally irrelevant inputs.
Apart from occasionally finding the explanations of the workings of types of neural network a little hard to follow, the only thing that made me raise an eyebrow was being told that Marvin Minsky 'coined the phrase "suitcase word"' - I would have thought 'derived* the phrase from Lewis Carroll's term "portmanteau word"' would have been closer to reality.
There have been good books on the basics of AI already, and excellent ones on the problems that 'deep learning' and big data systems throw up. But without a doubt, Mitchell's book sets a new standard in giving an understanding of what's possible and how difficult it is to go further. It should be read by every journalist, PR person and politician before they pump out yet more hype on the AI future. Recommended.
Shelved as 'not-released-tbr'
May 24, 2020
October 29, 2019
This book should be widely read, especially by those with a technological or philosophical interest in artificial intelligence, which should be most people. It provides a succinct history of this ambitious thought-provoking field, and a beautiful overview of the current state of the art. It should be accessible to anyone unafraid of a little mathematics. Since it is such a quickly evolving field, this latter aspect may grow out of date rather quickly.
But most importantly, this book is a well-argued reality check: a bucket of cold water. (You know what I mean by that analogy, but can a computer?) AI is a technological endeavour, and like other big sci-fi dreams - deep space travel, cheap clean energy, transhumanism - there is an enormous gap between our current capability (impressive though it is) and our vividly imagined end point. It's a gap that's easy to dismiss while breathlessly fretting over superintelligence and singularities, but that gap is filled with some extremely difficult challenges that we currently have little idea how to approach, let alone solve.
Mitchell's prologue sets up the book as a quest to understand what is really going on in AI research, spurred by a colourful example of the starkly opposing views found in debates within and around the field. She recounts a visit by legendary AI researcher and author Douglas Hofstadter to Google's headquarters to give a talk. Hofstadter wrote "Godel, Escher, Bach: An Eternal Golden Braid" - a famous meditation on AI and the nature of human consciousness, which was a significant influence on many geeks, including both your humble reviewer and the author of this book. Mitchell sought out Hofstadter as a mentor as a result of reading it. Anyway, Hofstadter spoke to the younger hotshot Googlers of his aching anxiety that when general AI comes, it will reveal human consciousness as something not so special. Something explainable and easy to simulate. This was an unexpected departure from the usual AI-will-destroy-us/no-it-won't dichotomy. The Googlers, for the most part, also shared the faith that general AI was in some sense imminent, but that it would be a boon to society, and the existential angst didn't factor into their thinking.
The topic of AI inspires a lot of this sort of philosophising - as it should - and much has been written on it. But like all such high-concept scientific pondering, it's healthy to rest it on a bedrock of hard technical reality. This book provides that reality. How far have we really got on the quest to build our own replacement?
Mitchell begins with sketch of the history of AI research from its birth in the 1950s, and outlines its key figures and main ideological branches. These are broadly classified as symbolic (programmed facts and rules for inferring new facts, analogous to conscious reasoning) and subsymbolic (biologically inspired structures which learn patterns and rules via lots of example data, analogous to subconscious learning). Early on the symbolists were the dominant sect, but the pendulum has swung dramatically in the other direction in the last decade or two thanks to the runway success of "deep learning".
For a more detailed survey of the many approaches to the challenge of AI, I recommend "The Master Algorithm" by Pedro Domingos, who is quoted a number of times in this book.
Mitchell then begins her coverage of the current state of the industry with a deep dive into its hottest algorithm (you hated that pun, but could a computer?). Deep neural networks are now being used everywhere, from image recognition, to automated translation, to self-driving cars. Their success has lead many to speculate that this is significant step toward the holy grail of "general AI" - that is, a system that can learn a wide range of domains and function in them all simultaneously.
But Mitchell identifies a number of fatal flaws in current techniques that will limit how far they can be taken. And there are no obvious or easy solutions.
Deep neural networks are very good at learning patterns from training data, but even with "big data" quantities of examples there will always be the "long tail" of rare or one-off exceptions that will cause deep neural networks to fail spectacularly. Just think of the sorts of strange things that might happen in the road in front of a self-driving car. These can cause bizarre and unpredictable outputs because the system has no "common sense" understanding of the world to fall back on.
Similarly, such networks after vulnerable to malicious attack. Carefully designed and small changes to input data can induce incorrect responses, sometimes tuned to the attacker's wishes. Interestingly the changes can be so subtle that a human observer wouldn't notice the difference. This implies that the algorithms aren't really understanding the input, at least not in the way we do.
And one of the most frustrating shortcomings of deep neural networks is their inability to "explain" their results. They may give correct answers but we don't know how they reached that answer, at least not in any conceptually meaningful way.
Mitchell's explanation of the ideas behind the recent breakthrough at playing the came of "go" - that is, DeepMind's AlphaGo - was a very satisfying example of how various algorithmic techniques can be combined to yield spectacular results, albeit in narrow domains. AlphaGo combined deep neural networks with reinforcement learning, and Monte Carlo tree searching.
Likewise, the survey of the successes and failures in the domain of natural language processing were fascinating, and provided the clearest example of the roadblock that's looming ahead in many of these subfields: that is, that current AI systems don't really understand the world. They don't understand what the words they are processing refer to. That difficult to articulate mapping between a word like "drink" and the abstract concept of a drink that we humans so easily grasp.
The final part of the book has a couple of chapters on these critical missing puzzle pieces. One of the key skills we employ so effortlessly as humans, but which has proven so difficult to train into a computer, is that of making analogies. We instantly recognise objects or situations as instances of a abstract idea (drinks, arguments, friends, accidents) and draw upon our experience with similar instances and applying them to new situations. We easily spot the pertinent "sameness" between two instances of a thing, and understand which differences are relevant and which aren't. As such we can generalise our knowledge powerfully. We haven't figured out how to get a machine to do this well.
Mitchell rounds out the book with a "self interview", in the style of Godel Escher Bach, where she asks herself many of the "big questions" related to AI and uses this to summarise her conclusions. She makes a compelling case using insider knowledge, copious examples, quotes from other leading thinkers, and an entertaining wry wit. "And if any computers are reading this, tell me what "it" refers to in the previous sentence, and you're welcome to join in the discussion."
But most importantly, this book is a well-argued reality check: a bucket of cold water. (You know what I mean by that analogy, but can a computer?) AI is a technological endeavour, and like other big sci-fi dreams - deep space travel, cheap clean energy, transhumanism - there is an enormous gap between our current capability (impressive though it is) and our vividly imagined end point. It's a gap that's easy to dismiss while breathlessly fretting over superintelligence and singularities, but that gap is filled with some extremely difficult challenges that we currently have little idea how to approach, let alone solve.
Mitchell's prologue sets up the book as a quest to understand what is really going on in AI research, spurred by a colourful example of the starkly opposing views found in debates within and around the field. She recounts a visit by legendary AI researcher and author Douglas Hofstadter to Google's headquarters to give a talk. Hofstadter wrote "Godel, Escher, Bach: An Eternal Golden Braid" - a famous meditation on AI and the nature of human consciousness, which was a significant influence on many geeks, including both your humble reviewer and the author of this book. Mitchell sought out Hofstadter as a mentor as a result of reading it. Anyway, Hofstadter spoke to the younger hotshot Googlers of his aching anxiety that when general AI comes, it will reveal human consciousness as something not so special. Something explainable and easy to simulate. This was an unexpected departure from the usual AI-will-destroy-us/no-it-won't dichotomy. The Googlers, for the most part, also shared the faith that general AI was in some sense imminent, but that it would be a boon to society, and the existential angst didn't factor into their thinking.
The topic of AI inspires a lot of this sort of philosophising - as it should - and much has been written on it. But like all such high-concept scientific pondering, it's healthy to rest it on a bedrock of hard technical reality. This book provides that reality. How far have we really got on the quest to build our own replacement?
Mitchell begins with sketch of the history of AI research from its birth in the 1950s, and outlines its key figures and main ideological branches. These are broadly classified as symbolic (programmed facts and rules for inferring new facts, analogous to conscious reasoning) and subsymbolic (biologically inspired structures which learn patterns and rules via lots of example data, analogous to subconscious learning). Early on the symbolists were the dominant sect, but the pendulum has swung dramatically in the other direction in the last decade or two thanks to the runway success of "deep learning".
For a more detailed survey of the many approaches to the challenge of AI, I recommend "The Master Algorithm" by Pedro Domingos, who is quoted a number of times in this book.
Mitchell then begins her coverage of the current state of the industry with a deep dive into its hottest algorithm (you hated that pun, but could a computer?). Deep neural networks are now being used everywhere, from image recognition, to automated translation, to self-driving cars. Their success has lead many to speculate that this is significant step toward the holy grail of "general AI" - that is, a system that can learn a wide range of domains and function in them all simultaneously.
But Mitchell identifies a number of fatal flaws in current techniques that will limit how far they can be taken. And there are no obvious or easy solutions.
Deep neural networks are very good at learning patterns from training data, but even with "big data" quantities of examples there will always be the "long tail" of rare or one-off exceptions that will cause deep neural networks to fail spectacularly. Just think of the sorts of strange things that might happen in the road in front of a self-driving car. These can cause bizarre and unpredictable outputs because the system has no "common sense" understanding of the world to fall back on.
Similarly, such networks after vulnerable to malicious attack. Carefully designed and small changes to input data can induce incorrect responses, sometimes tuned to the attacker's wishes. Interestingly the changes can be so subtle that a human observer wouldn't notice the difference. This implies that the algorithms aren't really understanding the input, at least not in the way we do.
And one of the most frustrating shortcomings of deep neural networks is their inability to "explain" their results. They may give correct answers but we don't know how they reached that answer, at least not in any conceptually meaningful way.
Mitchell's explanation of the ideas behind the recent breakthrough at playing the came of "go" - that is, DeepMind's AlphaGo - was a very satisfying example of how various algorithmic techniques can be combined to yield spectacular results, albeit in narrow domains. AlphaGo combined deep neural networks with reinforcement learning, and Monte Carlo tree searching.
Likewise, the survey of the successes and failures in the domain of natural language processing were fascinating, and provided the clearest example of the roadblock that's looming ahead in many of these subfields: that is, that current AI systems don't really understand the world. They don't understand what the words they are processing refer to. That difficult to articulate mapping between a word like "drink" and the abstract concept of a drink that we humans so easily grasp.
The final part of the book has a couple of chapters on these critical missing puzzle pieces. One of the key skills we employ so effortlessly as humans, but which has proven so difficult to train into a computer, is that of making analogies. We instantly recognise objects or situations as instances of a abstract idea (drinks, arguments, friends, accidents) and draw upon our experience with similar instances and applying them to new situations. We easily spot the pertinent "sameness" between two instances of a thing, and understand which differences are relevant and which aren't. As such we can generalise our knowledge powerfully. We haven't figured out how to get a machine to do this well.
Mitchell rounds out the book with a "self interview", in the style of Godel Escher Bach, where she asks herself many of the "big questions" related to AI and uses this to summarise her conclusions. She makes a compelling case using insider knowledge, copious examples, quotes from other leading thinkers, and an entertaining wry wit. "And if any computers are reading this, tell me what "it" refers to in the previous sentence, and you're welcome to join in the discussion."
July 16, 2021
Dr. Mitchell, I salute you.
This is a primer that actually does its job.
In a field that tends to Sarah Connorish hysteria or I-am-become-Goddish euphoria, she finds a middle way.
She also manages to explain some of the nuts and bolts of how AI actually works. She does so in an intuitive, conceptual way without eye-glazing equations, absurd similes or hyperbole. A tour de force. Did she have a helping hand from the future?
This is a primer that actually does its job.
In a field that tends to Sarah Connorish hysteria or I-am-become-Goddish euphoria, she finds a middle way.
She also manages to explain some of the nuts and bolts of how AI actually works. She does so in an intuitive, conceptual way without eye-glazing equations, absurd similes or hyperbole. A tour de force. Did she have a helping hand from the future?
July 2, 2021
A realistic overview of the AI field, what it has accomplish so far and the long road ahead. Good for any level of expertise to get an overall idea of the techniques and specially their limits, without any of the hype that surrounds this field. There’s still a lot of work to do, AI isn’t going to take over the world in any near future but still... what a time to be alive!
November 29, 2021
Solid read. Though crazy that parts of it are already dated given it was published in just 2019! The AI field has been moving so crazy fast
Would recommend if you're comfortable with basic math and are interested in how a lot of AI algorithms work, and what's real vs what's hype. Also a useful read for understanding the perpetual "hype" behind AI, and how companies (like IBM in the past) have used it for marketing
Would recommend if you're comfortable with basic math and are interested in how a lot of AI algorithms work, and what's real vs what's hype. Also a useful read for understanding the perpetual "hype" behind AI, and how companies (like IBM in the past) have used it for marketing
November 29, 2020
Quite interesting. Recommended.
November 16, 2022
A clearly written book for the amateur, if AI can ever be clear! Mitchell does a good job of describing the history of AI, though some of the academic back-slapping becomes irritating at times. The final chapters focus the book's theme: how advanced is AI and should we exclude it or welcome it? The media is often filled with hype and thoughts of a very brave new world where intelligent machines approach human capabilities. In reality, AI is nowhere near human intelligence/general intelligence.
Mitchell arrives, finally, at a telling conclusion: humanity should fear machine stupidity rather than supra intelligence. Machines working with incorrect human data or experiencing tail backs are dangerous. A book that gives food for thought.
Mitchell arrives, finally, at a telling conclusion: humanity should fear machine stupidity rather than supra intelligence. Machines working with incorrect human data or experiencing tail backs are dangerous. A book that gives food for thought.
December 16, 2022
It was very interesting but even though she dumbed it down some parts were still hard to grasp for my tiny little brain.
October 18, 2022
Human imagination has always been centuries ahead of human innovation and nowhere else is this dichotomy more glaring than when it comes to the field of Artificial intelligence. We have all seen movies, shows, or read books about what the world would like with AI embedded into its fabric. Some of these depictions are over embellished, fanciful and down right fucking terrifying. We have heard brilliant researchers, scientists, entrepreneur's simultaneously tell us mixed messages ranging from "There is nothing to be afraid of, AI will never happen." to "The Human species will be upended by AI.". Not to mention the overly ambitious ones who prophetically claimed "AI would replace humans in 20 years" 60 years ago. With so many boisterous opinions it can be difficult to distinguish the signal from the noise.
Melanie Mitchell's book is a valiant effort to help distinguish the signal from the noise. She goes through the history of AI, how news headlines from 60 years ago are not that different from the ones we see today, that is not to imply the field of AI is static by any means. In fact it has grown tremendously with the exponential pacing of Moore's law, making GPU chips possible which makes the matrix multiplication computation feasible, neural networks are able to do the processing they could not handle a decade and some change ago. It's not that the field is not racing forward, it's more so that it has become harder to interpret what exactly each new hurdle jump means for the future of the field. Often times progress is met with a defining headline that over embellish the nature of the accomplishment leading to more confusion from spectators. Let's not forget that companies also receive a healthy stock bump when they walk out their latest AI jump. This can mean billions for the company in publicity to sell services who don't quite live up to the hype. Not to mention academics crave attention more than a Coachella raver craves ecstasy.
Even with this incentive coming into the light, AI has made rapid improvements in numerous fields that are very pronounceable from natural language processing, computer vision and speech processing etc. The author goes into depth about the differences between types of learning such as supervised learning and reinforcement learning. It goes through the journey of AI from being a moderately good checkers player in the 60's to beating a chess champion in the late 90's and then to dominating the most sophisticated games such as go with Googles AlphaGO. Making tremendous strides in what is a closed system, a system where the rules do not deviate from expectations too often and statistical Monte Carlo simulations based on past moves/game results can make the system infinitely smarter. Some researchers thought it would take 20-30 years for AI to beat a go player based on its exponential number of moves. You see how the headlines from this game to the average spectator can be explosive and misguided.
I have a hard lined stance on AI, as companies, governments and rogue factions pour billions into AI to one day make trillions. The pace of AI is moving like a formula one race car at speeds that are impossible to stop and think about the implications of each new finding. AI researchers talk in unison as if they're all on the same page, adhere to the same ethics and want the same outcome. This is impossible. As we speak almost every country has a team of brilliant researchers working in a lab making sure never to disclose what they're researching, their findings or achievement's.The countries that are not doing this are naive Every movie scenario with killer AI implications are not fanciful, they're in fact very realistic. If you go back in the totality of human history, no human or country that had the ability to exert tremendous power on an opposing force to get what they wanted by ill means through a technology shied away from using that technology. AI researchers talk as if they all will make progress together, this is incredibly stupid. AI represents an ability to dominate an entire system without blowback or repercussion to the one harnessing in the short run.
As climate change hurdles on and countries begin to fight for resources such as water etc. AI could be looked as a way out, no doubt countries begun their AI divisions some time ago preparing for whats to come. In terms of hierarchy of fears, the fear of general intelligence or super intelligence are not the top two on my list. The fear is giving basic computer algorithms control of our systems that are very stupid who might interpret things in ways that are unpredictable. Think of it likes this, developers (myself included) are very stupid, don't let fancy titles fool you, or jargon used to describe what they do like "I am a data scientist who does statistical modeling, spotting patterns", no developer in the history of software engineering has ever written a program without unintended side effects, that did not say "Why the fuck is it doing that?" after they ran their program.
Let me give you an example, take SpaceX while the liftoff of Falcon 1 version 2 was commencing. Between stages, the rocket was supposed to integrate new propulsion between altitudes like reaching into orbit. One line of code had an unintended side effect that caused an error. This line of code most likely passed code reviews, sitting in the code base like a ticking time bomb.
Innovation emerges from numerous iterations, which can sometimes be all failures and all usually bad ideas yet from these bad ideas, good ideas can emerge. With each iteration, the innovator's level of understanding moves forward and from that trial and error process, the innovator and the technology grow. The key point is that the growth is not in tandem. It is not shared proportionately, sometimes the technology can rapidly move ahead of the understanding of the initial invention. Smart people have a habit of thinking they're smarter than they actually are. It's very human to be over confident, this is a pattern we all have seen throughout history. Sometimes researchers however brilliant, whatever their credentials maybe can fall into this trap. Take these two arguments and then combine them with a mantra like "Move fast and break things", you can create a dangerous precedent. With AI, there are no bug fixes, there are no reverting changes back to the initial state, no reboot etc. This is why I believe AI will rapidly move forward and uncontrollably.
I believe unintended consequences of AI are inevitable. Researchers have talked about AI gaining the values of humanity, the morality etc in what is regarded as the "Alignment problem" between AI and humans. How to align our values and goals with AI. Again, this assumes that rogue factions/governments are all aligned with ethical researchers but lets go with that for now.
I don't see how a super intelligent being, looking at humans could ever think "oh yeah I want to work with this species." given the countless tragedies, genocides, wars, murders we have conducted... I mean should I say more?
AI is coming whether it takes 10, 20, 30... 100 years. It is coming and when it does, I am not sure what the use of humans are for?
Overall I greatly enjoyed this read. Mind you I am biased to all AI books 5 stars.
Melanie Mitchell's book is a valiant effort to help distinguish the signal from the noise. She goes through the history of AI, how news headlines from 60 years ago are not that different from the ones we see today, that is not to imply the field of AI is static by any means. In fact it has grown tremendously with the exponential pacing of Moore's law, making GPU chips possible which makes the matrix multiplication computation feasible, neural networks are able to do the processing they could not handle a decade and some change ago. It's not that the field is not racing forward, it's more so that it has become harder to interpret what exactly each new hurdle jump means for the future of the field. Often times progress is met with a defining headline that over embellish the nature of the accomplishment leading to more confusion from spectators. Let's not forget that companies also receive a healthy stock bump when they walk out their latest AI jump. This can mean billions for the company in publicity to sell services who don't quite live up to the hype. Not to mention academics crave attention more than a Coachella raver craves ecstasy.
Even with this incentive coming into the light, AI has made rapid improvements in numerous fields that are very pronounceable from natural language processing, computer vision and speech processing etc. The author goes into depth about the differences between types of learning such as supervised learning and reinforcement learning. It goes through the journey of AI from being a moderately good checkers player in the 60's to beating a chess champion in the late 90's and then to dominating the most sophisticated games such as go with Googles AlphaGO. Making tremendous strides in what is a closed system, a system where the rules do not deviate from expectations too often and statistical Monte Carlo simulations based on past moves/game results can make the system infinitely smarter. Some researchers thought it would take 20-30 years for AI to beat a go player based on its exponential number of moves. You see how the headlines from this game to the average spectator can be explosive and misguided.
I have a hard lined stance on AI, as companies, governments and rogue factions pour billions into AI to one day make trillions. The pace of AI is moving like a formula one race car at speeds that are impossible to stop and think about the implications of each new finding. AI researchers talk in unison as if they're all on the same page, adhere to the same ethics and want the same outcome. This is impossible. As we speak almost every country has a team of brilliant researchers working in a lab making sure never to disclose what they're researching, their findings or achievement's.The countries that are not doing this are naive Every movie scenario with killer AI implications are not fanciful, they're in fact very realistic. If you go back in the totality of human history, no human or country that had the ability to exert tremendous power on an opposing force to get what they wanted by ill means through a technology shied away from using that technology. AI researchers talk as if they all will make progress together, this is incredibly stupid. AI represents an ability to dominate an entire system without blowback or repercussion to the one harnessing in the short run.
As climate change hurdles on and countries begin to fight for resources such as water etc. AI could be looked as a way out, no doubt countries begun their AI divisions some time ago preparing for whats to come. In terms of hierarchy of fears, the fear of general intelligence or super intelligence are not the top two on my list. The fear is giving basic computer algorithms control of our systems that are very stupid who might interpret things in ways that are unpredictable. Think of it likes this, developers (myself included) are very stupid, don't let fancy titles fool you, or jargon used to describe what they do like "I am a data scientist who does statistical modeling, spotting patterns", no developer in the history of software engineering has ever written a program without unintended side effects, that did not say "Why the fuck is it doing that?" after they ran their program.
Let me give you an example, take SpaceX while the liftoff of Falcon 1 version 2 was commencing. Between stages, the rocket was supposed to integrate new propulsion between altitudes like reaching into orbit. One line of code had an unintended side effect that caused an error. This line of code most likely passed code reviews, sitting in the code base like a ticking time bomb.
Innovation emerges from numerous iterations, which can sometimes be all failures and all usually bad ideas yet from these bad ideas, good ideas can emerge. With each iteration, the innovator's level of understanding moves forward and from that trial and error process, the innovator and the technology grow. The key point is that the growth is not in tandem. It is not shared proportionately, sometimes the technology can rapidly move ahead of the understanding of the initial invention. Smart people have a habit of thinking they're smarter than they actually are. It's very human to be over confident, this is a pattern we all have seen throughout history. Sometimes researchers however brilliant, whatever their credentials maybe can fall into this trap. Take these two arguments and then combine them with a mantra like "Move fast and break things", you can create a dangerous precedent. With AI, there are no bug fixes, there are no reverting changes back to the initial state, no reboot etc. This is why I believe AI will rapidly move forward and uncontrollably.
I believe unintended consequences of AI are inevitable. Researchers have talked about AI gaining the values of humanity, the morality etc in what is regarded as the "Alignment problem" between AI and humans. How to align our values and goals with AI. Again, this assumes that rogue factions/governments are all aligned with ethical researchers but lets go with that for now.
I don't see how a super intelligent being, looking at humans could ever think "oh yeah I want to work with this species." given the countless tragedies, genocides, wars, murders we have conducted... I mean should I say more?
AI is coming whether it takes 10, 20, 30... 100 years. It is coming and when it does, I am not sure what the use of humans are for?
Overall I greatly enjoyed this read. Mind you I am biased to all AI books 5 stars.
April 9, 2024
The field of AI has come a long way since this book was published. But the ideas that Melanie writes about will always stand the test of time and provide a complete picture of the journey that AI has taken so far and what we must expect of its future. She has been one of my favourite thinkers in this space and she writes in a cool, intuitive manner with a few jokes sprinkled in the way one would write about a mildly annoying friend. The book comes full circle and ends with a line that leaves you laughing and with hope for us pesky humans.
December 27, 2019
Melanie Mitchell provides an excellent summary of the current endeavours across fields in AI and an interim (bleak) assessment of the field's progress towards the holy grail of strong AI. Mitchell discusses recent milestones reached by AI (most based on approaches using some form of deep learning) taking a critical point of view, devoid of sensationalism as encountered in the media. Mitchell’s writing is lucid and engaging: many technical concepts are explained in clear and vidid language.
What’s more, Mitchell is one of Douglas Hofstadter most successful students. Thus, the book may also read as as follow up to Hofstader's "GEB: an Eternal Golden Braid". Frequent references to Hofstadter's views on AI and to GEB are dispersed throughout the book and substantiate this interpretation. A lot has happened since GEB was originally published, and Mitchell's book can provide a much needed synthesis and constructive discussion of both Hofstader's ideas and the current state-of-art approaches for re-creating intelligence.
The book covers many areas of AI, especially considering the astounding growth of the field, but there were sections that could be further developed: for instance, the section on creativity at the end was interesting and could be developed in a stand-alone chapter. Notwithstanding, Mitchell’s AI is bound to entertain readers across disciplines and can become an accessible and balanced entry point to research on AI for a wide audience.
What’s more, Mitchell is one of Douglas Hofstadter most successful students. Thus, the book may also read as as follow up to Hofstader's "GEB: an Eternal Golden Braid". Frequent references to Hofstadter's views on AI and to GEB are dispersed throughout the book and substantiate this interpretation. A lot has happened since GEB was originally published, and Mitchell's book can provide a much needed synthesis and constructive discussion of both Hofstader's ideas and the current state-of-art approaches for re-creating intelligence.
The book covers many areas of AI, especially considering the astounding growth of the field, but there were sections that could be further developed: for instance, the section on creativity at the end was interesting and could be developed in a stand-alone chapter. Notwithstanding, Mitchell’s AI is bound to entertain readers across disciplines and can become an accessible and balanced entry point to research on AI for a wide audience.
May 17, 2023
This is an excellent survey of AI up through 2020 or so, when this book was published. It obviously doesn’t directly address recent breakthroughs in generative AI (ChatGPT), but it sets the stage and provides helpful context. There’s just enough theory to get the reader familiar with key concepts without getting bogged down in esoteric diagrams and notation. And I enjoyed the blend of science, history, and philosophy (what is intelligence, anyway?). Most of all, I found the author’s skepticism refreshing. Mitchell isn’t a technologist; she’s honest about the risks of AI and she undercuts a lot of the hype. That said, I would love to read an updated edition or sequel that addresses ChatGPT directly. Maybe Mitchell has something in the works?
May 3, 2020
I read this cover to cover in about two sittings, first book I've done like that in a while. There's a lot of pop science books out about AI and machine learning and a lot of them aren't very good. This is intelligent but not obscure, conversational to the point that it's almost gossipy it reads like a quanta article if they delve just a bit deeper into their subject. Whether it's the cheating controversies or the history of AIs for games to the speculative portions in the later part of the book, I found it all engrossing and perfectly succinct. When my friends express their fears about the singularity or artificial intelligent this will be the book I give to them.
October 20, 2019
Excellent summary and overview of the current state and challenges facing artificial intelligence. Should be readily accessible to any interested reader without requiring pre-existing knowledge of the field.
December 1, 2019
Well explained, balanced, nuanced evaluation of the current status of AI.
July 6, 2020
"In any ranking of near-term worries about AI, superintelligence should be far down the list. In fact, the opposite of superintelligence is the real problem."
December 27, 2019
René Descartes, a French philosopher, mathematician and scientist in elucidating his famous theory of dualism, expounded that there exist two kinds of foundation: mental and physical. While the mental can exist outside of the body, and the body cannot think. Popularly known as mind-body dualism or Cartesian Duality (after the theory’s proponent), the central tenet of this philosophy is that the immaterial mind and the material body, while being ontologically distinct substances, causally interact. British philosopher Gilbert Ryle‘s in describing René Descartes’ mind-body dualism, introduced the now immortal phrase, “ghost in the machine” to highlight the view of Descartes and others that mental and physical activity occur simultaneously but separately.
Ray Kurzweil, the high priest of futurism and Director of Engineering at Google, takes Cartesian Duality to a higher plane with his public advocacy of concepts such as Technological Singularity and radical life extension. Kurzweil argues that with giant leaps in the domain of Artificial Intelligence, mankind will experience a radical life extension by 2045. Skeptics on the other hand bristle at this very notion, claiming such “Kurzweilian” aspirations to be mere fantasies putting to shame even the most ludicrous of pipe dreams.
The advances in the field of AI have spawned a seminal debate that has a vertical cleave. On one side of the chasm are the undying optimists such as Ray Kurzweil predicting a new epoch in the history of mankind, while on the other side of the divide are placed pessimists and naysayers such as Nick Bostrom, James Barrat and even the likes of Bill Gates, Elon Musk and Stephen Hawking who advocate extreme caution and warn about existential risks. So what is the actual fact? Melanie Mitchell, a computer science professor at Portland State University takes this conundrum head on in her eminently readable book, ““Artificial Intelligence: A Guide for Thinking Humans.” A measured book, that abhors mind numbing technicalities and arcane elaborations, Ms. Mitchell’s work embodies a matter-of-fact narrative that seeks to demystify the future of both AI and its users.
The book begins with a meeting organized by Blaise Agüera y Arcas, a computer scientist leading Google’s foray into machine intelligence. In the meeting, the genius AI pioneer and author of the Pulitzer Prize winning book, “Gödel, Escher, Bach: an Eternal Golden Braid” (or just “gee-ee-bee’), Douglas Hofstadter expresses downright alarm at the principle of Singularity being touted by Kurzweil. “If this actually happens, “we will be superseded. We will be relics. We will be left in the dust.” A former research assistant of Hofstadter, Ms. Mitchell is surprised to hear such an exclamation from her mentor. This spurs her on to assess the impact of AI, in an unbiased vein.
Tracing the modest trajectory of the beginning of AI, Ms. Mitchell informs her reader about a small workshop in Dartmouth in 1956 where the seeds of AI were first sown. John McCarthy, universally acknowledged as the father of AI and the inventor of the term itself, persuaded Marvin Minsky, a fellow student at Princeton, Claude Shannon, the inventor of information theory and Nathaniel Rochester, a pioneering electrical engineer, to help him organize “a 2 month, 10-man study of artificial intelligence to be carried out during the summer of 1956.” What began as a muted endeavor has now morphed into a creature that is both revered and reviled, in equal measure. Ms. Mitchell lends a technical element to the book by dwelling on concepts such as symbolic and sub-symbolic AI. Ms. Mitchell, however lends a fascinating insight into the myriad ways in which various intrepid pioneers and computer experts attempted to distill the element of “learning” into a computer thereby bestowing it with immense scalability and computational skills.
For example, using a technique termed, back-propagation, errors are taken away at the output units and to “propagate” the blame for that error backward so as to assign proper blame to each of the weights in the network. This allows back-propagation to determine how much to change each weight in order to reduce the error. The beauty of Ms. Mitchell’s explanations lies in its simplicity. She breaks down seemingly esoteric concepts into small chunks of ‘learnable’ elements.
It is these kind of techniques that have enabled IBM’s Watson to defeat World Chess Champion Garry Kasparov, and trump over Jeopardy! Champions Ken Jennings and Brad Rutter. So with such stupendous advances, is the time where Artificial Intelligence surpasses human intelligence already upon us? Ms. Mitchell does not think so. Taking recourse to the views of Alan Turing’s “argument from consciousness,” Ms. Mitchell brings to our attention, Turing’s summary of the neurologist Geoffrey Jefferson’s quote:
“Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain—that is, not only write it but know that it had written it. No mechanism could feel (and not merely artificially signal, an easy contrivance) pleasure at its successes, grief when its valves fuse, be warmed by flattery, be made miserable by its mistakes, be charmed by sex, be angry or depressed when it cannot get what it wants.”
Ms. Mitchell also highlights – in a somewhat metaphysical manner – the inherent limitations of a computer to gainfully engage in the attributes of abstraction and analogy. In the words of her own mentor Hofstadter and his coauthor, the psychologist Emmanuel Sander, “Without concepts there can be no thought, and without analogies there can be no concepts.” If computers are bereft of common sense, it is not for the want of their users trying to ‘embed’ some into them. A famous case in point being Douglas Lenat’s Cyc project which ultimately turned out to be a bold, albeit futile exercise.
A computer’s inherent limitation in thinking like a human being was also demonstrated by The Winograd schemas. These were schemas designed precisely to be easy for humans but tricky for computers. Hector Levesque, Ernest Davis, and Leora Morgenstern three AI researchers, “proposed using a large set of Winograd schemas as an alternative to the Turing test. The authors argued that, unlike the Turing test, a test that consists of Winograd schemas forestalls the possibility of a machine giving the correct answer without actually understanding anything about the sentence. The three researchers hypothesized (in notably cautious language) that “with a very high probability, anything that answers correctly is engaging in behaviour that we would say shows thinking in people.”
Finally, Ms. Mitchell concludes by declaring that machines are as yet incapable of generalizing, understanding cause and effect, or transferring knowledge from situation to situation – skills human beings begin to develop in infancy. Thus while computers won’t dethrone man anytime soon, goading them on to bring such an endeavor to fruition might not be a wise idea, after all.
Ray Kurzweil, the high priest of futurism and Director of Engineering at Google, takes Cartesian Duality to a higher plane with his public advocacy of concepts such as Technological Singularity and radical life extension. Kurzweil argues that with giant leaps in the domain of Artificial Intelligence, mankind will experience a radical life extension by 2045. Skeptics on the other hand bristle at this very notion, claiming such “Kurzweilian” aspirations to be mere fantasies putting to shame even the most ludicrous of pipe dreams.
The advances in the field of AI have spawned a seminal debate that has a vertical cleave. On one side of the chasm are the undying optimists such as Ray Kurzweil predicting a new epoch in the history of mankind, while on the other side of the divide are placed pessimists and naysayers such as Nick Bostrom, James Barrat and even the likes of Bill Gates, Elon Musk and Stephen Hawking who advocate extreme caution and warn about existential risks. So what is the actual fact? Melanie Mitchell, a computer science professor at Portland State University takes this conundrum head on in her eminently readable book, ““Artificial Intelligence: A Guide for Thinking Humans.” A measured book, that abhors mind numbing technicalities and arcane elaborations, Ms. Mitchell’s work embodies a matter-of-fact narrative that seeks to demystify the future of both AI and its users.
The book begins with a meeting organized by Blaise Agüera y Arcas, a computer scientist leading Google’s foray into machine intelligence. In the meeting, the genius AI pioneer and author of the Pulitzer Prize winning book, “Gödel, Escher, Bach: an Eternal Golden Braid” (or just “gee-ee-bee’), Douglas Hofstadter expresses downright alarm at the principle of Singularity being touted by Kurzweil. “If this actually happens, “we will be superseded. We will be relics. We will be left in the dust.” A former research assistant of Hofstadter, Ms. Mitchell is surprised to hear such an exclamation from her mentor. This spurs her on to assess the impact of AI, in an unbiased vein.
Tracing the modest trajectory of the beginning of AI, Ms. Mitchell informs her reader about a small workshop in Dartmouth in 1956 where the seeds of AI were first sown. John McCarthy, universally acknowledged as the father of AI and the inventor of the term itself, persuaded Marvin Minsky, a fellow student at Princeton, Claude Shannon, the inventor of information theory and Nathaniel Rochester, a pioneering electrical engineer, to help him organize “a 2 month, 10-man study of artificial intelligence to be carried out during the summer of 1956.” What began as a muted endeavor has now morphed into a creature that is both revered and reviled, in equal measure. Ms. Mitchell lends a technical element to the book by dwelling on concepts such as symbolic and sub-symbolic AI. Ms. Mitchell, however lends a fascinating insight into the myriad ways in which various intrepid pioneers and computer experts attempted to distill the element of “learning” into a computer thereby bestowing it with immense scalability and computational skills.
For example, using a technique termed, back-propagation, errors are taken away at the output units and to “propagate” the blame for that error backward so as to assign proper blame to each of the weights in the network. This allows back-propagation to determine how much to change each weight in order to reduce the error. The beauty of Ms. Mitchell’s explanations lies in its simplicity. She breaks down seemingly esoteric concepts into small chunks of ‘learnable’ elements.
It is these kind of techniques that have enabled IBM’s Watson to defeat World Chess Champion Garry Kasparov, and trump over Jeopardy! Champions Ken Jennings and Brad Rutter. So with such stupendous advances, is the time where Artificial Intelligence surpasses human intelligence already upon us? Ms. Mitchell does not think so. Taking recourse to the views of Alan Turing’s “argument from consciousness,” Ms. Mitchell brings to our attention, Turing’s summary of the neurologist Geoffrey Jefferson’s quote:
“Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain—that is, not only write it but know that it had written it. No mechanism could feel (and not merely artificially signal, an easy contrivance) pleasure at its successes, grief when its valves fuse, be warmed by flattery, be made miserable by its mistakes, be charmed by sex, be angry or depressed when it cannot get what it wants.”
Ms. Mitchell also highlights – in a somewhat metaphysical manner – the inherent limitations of a computer to gainfully engage in the attributes of abstraction and analogy. In the words of her own mentor Hofstadter and his coauthor, the psychologist Emmanuel Sander, “Without concepts there can be no thought, and without analogies there can be no concepts.” If computers are bereft of common sense, it is not for the want of their users trying to ‘embed’ some into them. A famous case in point being Douglas Lenat’s Cyc project which ultimately turned out to be a bold, albeit futile exercise.
A computer’s inherent limitation in thinking like a human being was also demonstrated by The Winograd schemas. These were schemas designed precisely to be easy for humans but tricky for computers. Hector Levesque, Ernest Davis, and Leora Morgenstern three AI researchers, “proposed using a large set of Winograd schemas as an alternative to the Turing test. The authors argued that, unlike the Turing test, a test that consists of Winograd schemas forestalls the possibility of a machine giving the correct answer without actually understanding anything about the sentence. The three researchers hypothesized (in notably cautious language) that “with a very high probability, anything that answers correctly is engaging in behaviour that we would say shows thinking in people.”
Finally, Ms. Mitchell concludes by declaring that machines are as yet incapable of generalizing, understanding cause and effect, or transferring knowledge from situation to situation – skills human beings begin to develop in infancy. Thus while computers won’t dethrone man anytime soon, goading them on to bring such an endeavor to fruition might not be a wise idea, after all.
December 6, 2023
Good background information in the field and coverage for a general audience.
I particularly liked her discussion on the Bongard Problems and her reference to GEB. The Bongard Problems reminded me (although Mitchell probably didn't intend it to) of Raven's Progressive Matrices test of Fluid Intelligence in humans. This connection of abstract reasoning in both cases (Bongard and Raven's) was an interesting parallel and although perhaps not intentional by the author, the value in consideration between the two and the related implications wasn't lost on me.
I do have to say although I appreciate the author's skepticism, she did seem a bit biased to me perhaps because of lived experience through multiple AI Winters and Summers, and having seen the marketing hype cycles occur over and over in the field.
And somewhat related in her latest presentation on November 15th, 2023 I think she admits to some level of cognitive dissonance (and bias?) perhaps as the latest news of Modern System (Transformer) capabilities continues to defy her expectations which can be seen here;
https://youtu.be/GwHDAfAAKd4?t=4439
(1:14:00)
I particularly liked her discussion on the Bongard Problems and her reference to GEB. The Bongard Problems reminded me (although Mitchell probably didn't intend it to) of Raven's Progressive Matrices test of Fluid Intelligence in humans. This connection of abstract reasoning in both cases (Bongard and Raven's) was an interesting parallel and although perhaps not intentional by the author, the value in consideration between the two and the related implications wasn't lost on me.
I do have to say although I appreciate the author's skepticism, she did seem a bit biased to me perhaps because of lived experience through multiple AI Winters and Summers, and having seen the marketing hype cycles occur over and over in the field.
And somewhat related in her latest presentation on November 15th, 2023 I think she admits to some level of cognitive dissonance (and bias?) perhaps as the latest news of Modern System (Transformer) capabilities continues to defy her expectations which can be seen here;
https://youtu.be/GwHDAfAAKd4?t=4439
(1:14:00)
December 12, 2023
i had to read this book for my psychology & ai class and i really enjoyed it! probably the first time i enjoyed reading a textbook for class. i thought the author explained all concepts very well and with good examples, and i also thought the length of the book and the chapters was really good
December 13, 2023
4 years and we got chatGPT and Bard this is wild
July 16, 2019
** I won an advance reader copy of this book for free through a Goodreads giveaway. **
This is a well-written and thought-provoking account of the history and potential future of artificial intelligence. The author writes in a style that allows both those who have a background in AI to gain new insight and those who have little to no background to follow along (I understand the basics but am by no means an expert and I had little to no trouble understanding). The author also does an excellent job of explaining misconceptions that exist about the field of AI and giving us a thoughtful look at the future of this field. Whether you are someone who is interested in going into the field or someone who fears the 'uprising', this book is an excellent place to start reading. Definitely recommended!
This is a well-written and thought-provoking account of the history and potential future of artificial intelligence. The author writes in a style that allows both those who have a background in AI to gain new insight and those who have little to no background to follow along (I understand the basics but am by no means an expert and I had little to no trouble understanding). The author also does an excellent job of explaining misconceptions that exist about the field of AI and giving us a thoughtful look at the future of this field. Whether you are someone who is interested in going into the field or someone who fears the 'uprising', this book is an excellent place to start reading. Definitely recommended!
April 9, 2023
Probably the best non-fiction book I’ve read up to date.
The fact that this was published in 2019 (if I’m correct) makes it an even more compelling read. This book covers dilemmas and questions humans will have to face on the quest to reach artificial general intelligence. I really enjoyed the philosophical standpoints backed by psychological studies, as well as descriptions of how AI systems we use on daily basis actually work (there’s a lot of science but I still consider it a not so difficult read).
Given the recent AI-boom, I believe the author would have changed her opinions/predictions if she were writing it today. AI chatbots such as ChatGPT or Bing seem to be a huge step forward in AI development since 2019. For this reason, it is becoming more and more urgent to face many of the questions raised by the author, such as AI biases or AI morality.
The fact that this was published in 2019 (if I’m correct) makes it an even more compelling read. This book covers dilemmas and questions humans will have to face on the quest to reach artificial general intelligence. I really enjoyed the philosophical standpoints backed by psychological studies, as well as descriptions of how AI systems we use on daily basis actually work (there’s a lot of science but I still consider it a not so difficult read).
Given the recent AI-boom, I believe the author would have changed her opinions/predictions if she were writing it today. AI chatbots such as ChatGPT or Bing seem to be a huge step forward in AI development since 2019. For this reason, it is becoming more and more urgent to face many of the questions raised by the author, such as AI biases or AI morality.
February 12, 2020
A fresh, down to earth, review of the current AI craze. Melanie Mitchell is no stranger to the field (she did her Ph.D with Douglas Hofstadter, and, quite some years ago, published the best book about Genetic Algorithms you can find). She's quick to pinpoint the limits of current Neural Network (CNNs, RNNs, DQNs) centric approaches to AI, highlighting the need to overcome the "barrier of meaning" and search for models that could work with 'common sense' knowledge and the capacity for sophisticated abstraction and analogy making.
Fun read, she writes very well.
Fun read, she writes very well.
December 23, 2019
A wonderful introduction to AI, covering the latest techniques and resources in a truly accessible way. It deserves the title of “a guide for thinking humans” — I was impressed by the author’s ability to convey complex AI concepts in simple terms.
A bit that I particularly liked — the recurring recipe for AI research (ie, hype): define narrow problem, achieve human-level performance, make big claims for broader problem.
A bit that I particularly liked — the recurring recipe for AI research (ie, hype): define narrow problem, achieve human-level performance, make big claims for broader problem.
May 8, 2020
Wow, what a great read. No matter who you are, in this modern world it is highly likely that you are involved one way or another with AI. AI is so ubiquitous that at some point I want to dismantle the seemingly impenetrable barrier and peek inside a bit beyond the simplest definition. This book does exactly that, with such clarity.
And it offers more than that. Besides providing some basic notions of various algorithms in AI, Mitchell brings into the table an in depth discussion of the overall picture of the field, with her many years experience in the field as an academic researcher. She doesn't have a relentless confidence in AI and its wild promises. At the same time she doesn't talk as if the endeavor is inevitably doomed. The balance is hard to kept, and is kept.
I have been a bit skeptical of the current achievements of AI. Mostly I have been afraid of the misuse and the intransparency. Mitchell certainly mentions those aspects, but she further clarifies all the shortcomings without being dramatic or sensational. I realize I should be even more concerned. Most if not all applications are so fragile to attack (hacking), and I tremble to think of consequences. Yes, some applications are extremely useful, but in the end the utmost importance lies in how humans regulate the scope of the use of AI. I think everyone should read this book, especially policy makers, especially young applied computer scientists.
At the same time, the beauty of the book is that Mitchell helps to elucidate how "intelligent" humans are. Many things we consider easy turns out so gruellingly hard to analyze not to mention to recreate in a machine. I think of Kahneman and his Fast Thinking system. Oh yeah, we humans are trying to compete with nature that has been building intelligent life for a few billion years. Of course it is hard.
And it offers more than that. Besides providing some basic notions of various algorithms in AI, Mitchell brings into the table an in depth discussion of the overall picture of the field, with her many years experience in the field as an academic researcher. She doesn't have a relentless confidence in AI and its wild promises. At the same time she doesn't talk as if the endeavor is inevitably doomed. The balance is hard to kept, and is kept.
I have been a bit skeptical of the current achievements of AI. Mostly I have been afraid of the misuse and the intransparency. Mitchell certainly mentions those aspects, but she further clarifies all the shortcomings without being dramatic or sensational. I realize I should be even more concerned. Most if not all applications are so fragile to attack (hacking), and I tremble to think of consequences. Yes, some applications are extremely useful, but in the end the utmost importance lies in how humans regulate the scope of the use of AI. I think everyone should read this book, especially policy makers, especially young applied computer scientists.
At the same time, the beauty of the book is that Mitchell helps to elucidate how "intelligent" humans are. Many things we consider easy turns out so gruellingly hard to analyze not to mention to recreate in a machine. I think of Kahneman and his Fast Thinking system. Oh yeah, we humans are trying to compete with nature that has been building intelligent life for a few billion years. Of course it is hard.
March 2, 2021
Great introduction to the field of artificial for a general reader, pre-college teenagers, or even developers keen to explore new areas. It's a fair and balanced account of the algorithms and methodologies used in AI today and how they've evolved over the years. It's a wonderful antidote to all the people hyping their releases in the field and in media/journalism, and it will help you understand these cycles of hype and deflated expectations a lot better once you read it.
December 23, 2021
A gentle but holistic introduction to AI and progress in its various areas.
The author keeps a right balance between citing prominent speakers in the field and expressing her own stand. The book demystifies "AI" a lot by gently explaining how AI applications like natural-language processing, computer vision, game playing, etc. work. Her research interest sparks hopes in me about more sustainable progress in this field.
The author keeps a right balance between citing prominent speakers in the field and expressing her own stand. The book demystifies "AI" a lot by gently explaining how AI applications like natural-language processing, computer vision, game playing, etc. work. Her research interest sparks hopes in me about more sustainable progress in this field.
May 23, 2021
A nice summary of what has been achieved in the field of AI and the misconceptions and work still being done.
January 7, 2024
Istoria inteligenței artificiale împreună cu aspecte tehnice. Mai puțină paravlageala și mai multe lucruri concrete.
Mai întâi să vedem ce presupune AI și după aceea să ne îngrijorăm de el.
Mai întâi să vedem ce presupune AI și după aceea să ne îngrijorăm de el.
Displaying 1 - 30 of 358 reviews