An AI reading list — from practical primers to sci-fi short stories
The best reading on AI, as recommended by the experts
Experts are already building a future world brimming with artificial intelligence, but here in the present most of us are still trying to figure out what AI even is. This is a technology that will influence many aspects of our lives, from jobs to entertainment to health care, but that also engages with fundamental questions about what it means to be human. Questions like, “what is the nature of creativity?” and “how do we define consciousness?” Posing the question “how can I understand AI?” is nearly as daunting as asking “what is the meaning of life?”
But as with that tricky life question, a sense of overawing complexity doesn’t mean we shouldn’t try.
In order to help, The Verge has assembled a reading list: a brief but diverse compendium of books, short stories, and blogs, all chosen by leading figures in the AI world to help you better understand artificial intelligence. It’s an eclectic selection that ranges from practical primers to Golden Age sci-fi, and while reading everything listed below won’t get you a job at Google (though it certainly couldn’t hurt), it will give you much-needed context for this confusing and exciting time.
So read, enjoy, and get to know the captivating world of AI a little bit better.
★ Profiles of the Future, by Arthur C. Clarke
Recommended by Greg Brockman and Ilya Sutskever, co-founders of OpenAI
“Profiles of the Future changed our beliefs about how rapidly AI might affect the world. We used to think of technological change as a gradual, slow process — the sum of many small innovations that, when zoomed out, create only the illusion of rapid technological change.
Profiles made us realize there are some highly important exceptions. While later chapters describe Arthur C. Clarke’s predictions about the future, early chapters analyze others’ predictions about technologies like airplanes, space travel, and nuclear power before their development. In each case, the technology was predicted by a small number of optimists amongst a very large, vocal set of genuinely accomplished experts who were confident that a particular dramatic technological advance would never be achieved (at least not on a practical timescale). As a result, even to most experts, massive technological change appeared to come ‘out of nowhere.’
How will long term progress in AI look? Will it follow a predictable trajectory, with the field having a clear view of the upcoming progress in the next 5-10 years, or will we stumble upon a surprising yet disproportionate advance in AI that will transform the world rapidly? The perspective in Profiles means these questions are worth pondering.”
★ The Book of Why, by Judea Pearl and Dana Mackenzie
Recommended by Rumman Chowdhury, Responsible AI lead at Accenture
“An AI book with no robots, no doomsday scenarios, and no grandiose predictions of the future? How refreshing. This book’s humble and engaging writing style belies a deep hypothesis: the fundamental roots of our current systems of predictive modeling are wrong. According to the authors, we lack a language of causality; that is, quantifiable proof that one thing causes another. This is a fundamental weakness embedded in the history of statistics and tarnishes how we ask questions and seek answers.
The dirty secret of the AI and machine learning methods we use for prediction is that they cannot actually tell us with certainty whether some factor caused another, instead relying on millions of repetitions to give us high-value correlations. Many of our issues of biased outcomes in AI systems stem from an incomplete or poor understanding of interrelated variables (race and zip code, or socioeconomic status and education, for example). While still considered controversial (see Pearl’s debate with statistician Andrew Gelman on Twitter), The Book of Why presents a new narrative that questions and redefines the building blocks of our AI systems.”
★ “Franchise,” by Isaac Asimov
Recommended by Tim Hwang, director of the Harvard-MIT Ethics and Governance of AI Initiative
“Asimov’s Robot series is perhaps the cliche reference that gets rolled out when talking about the social impact of artificial intelligence. It’s mostly a convenient excuse to repeat well-worn tropes about The Three Laws of Robotics and point out — sagely — that the dreams of building intelligent machines are long-standing.
But, the cliche misses the mark. In the Asimov oeuvre, it is the stories featuring the massive, impersonal Multivac — rather than the Robot series — that best capture the present day reality of machine learning. In contrast to the walking, talking robots of the Robot stories, Multivac is an unwieldy server farm that requires specialized expertise to operate and frequently produces outputs uninterpretable to the technicians that run it.
One story I’ve found myself revisiting over and over again is Asimov’s ‘Franchise,’ published as a short story in the August 1955 edition of If magazine. In it, a future America (2008), decides to reduce voting to a statistical model that extrapolates the outcomes of all elections based on a set of questions answered by one, extremely representative person.
‘Franchise’ deftly captures the weirdly recursive nature of prediction, and the personal stresses of being the focus of algorithmic analysis. Importantly, the story illustrates the real and tricky balance between predictability and legitimacy. Even if we could do a perfect job predicting voting behavior, or recidivism, or employment performance, what does it mean for this to be an automated process versus a human one? Give it a read.”
★ Weapons of Math Destruction, by Cathy O’Neil
Recommended by Kate Darling, Research Specialist at the MIT Media Lab
“At first, I wanted to recommend a speculative science fiction book. But sometimes our current reality is a more interesting dystopia. In January 2019, US Congresswoman Alexandria Ocasio-Cortez was ridiculed for claiming that algorithms can be biased. No matter your political affiliation, I think everyone can benefit from a basic understanding of the pitfalls in contemporary AI systems. This book, illustrated with fascinating (and terrifying) real-world examples, is a great primer on the algorithms and data that we’re using, the delegation of power to systems that can make or break people’s lives, and the completely disastrous ways that we get it all wrong. Cathy O’Neil is a mathematician and data scientist who went from academia to the world of Wall Street quants and later joined the Occupy Wall Street movement. Her acclaimed book covers the problems with algorithms in the finance industry, but also in the areas of criminal justice, employment, education, and many more. Many of the AI systems we’re currently deploying and are likely to use in the near future run into the issues that O’Neil highlights. This book should be required basic reading for anyone interested in artificial intelligence implementation.”
★ The Diamond Age: Or, A Young Lady’s Illustrated Primer, by Neal Stephenson
Recommended by Jeremy Howard, co-founder of fast.ai
“The ‘Primer’ in the title refers to a leather-bound book. There are three Primers in existence, each one owned by a little girl. The primer is the greatest work of its creator, the top software engineer at the world’s most successful software company. Because, you see, it’s not an ordinary book; it is truly interactive, showing the reader exactly what they need at every moment, described in a way that is designed to maximize their interest. One of the three girls that owns a Primer is the protagonist, Nell, who after finding herself homeless discovers that the Primer has been teaching her all the skills she needs to survive, and to thrive. We follow her journey, guided by the Primer, from a little girl that’s lost everything, to a young woman who may just change the world.
I first read Diamond Age 20 years ago, and this message has stayed with me: technology can be harnessed to give opportunities to those that otherwise would not have them. As with all new technologies, there is today a knee-jerk reaction against ‘screens’ for children. There is no well-designed modern research to support this reaction. If we deny the opportunity to leverage technology in education, then we limit the best education to only those privileged enough to have access to the best teachers.
Our mission at fast.ai is to help provide access to AI tools and education to all. Technology is vital to this mission. Without it, our users and students wouldn’t have access to our online lessons and community, or the cloud compute platforms we rely on. However, I haven’t yet seen AI used to create a highly customized educational experience like the Primer. The technology foundations are largely in place now; it just needs someone to put them together. When that happens, we may hear of real-world stories like Nell’s.”
★ Machine Learning for Humans, by Vishal Maini and Samer Sabri
Recommended by Demis Hassabis, co-founder and CEO of DeepMind
“It’s surprisingly hard to recommend books about the nuts and bolts of AI that aren’t either too technical or too philosophical — I predict we’ll see a lot more over the next few years. I’d recommend Machine Learning For Humans as a good introduction that doesn’t require much prior knowledge, plus it’s free online. We were so impressed with it here [at DeepMind] that we ended up hiring one of its authors!
Another way to get to grips with AI is to use a subject you are more familiar with as a gateway. For example, most people know the basics of chess even if they haven’t played it much. Two expert chess players, Matthew Sadler and Natasha Regan, have just written a book called Game Changer about one of DeepMind’s recent research breakthroughs, AlphaZero, which learnt chess from scratch just by playing against itself to ultimately become the world’s strongest player. It’s one of the most comprehensive analyses of an advanced AI program ever undertaken and gives you a fascinating insight into how AI systems like AlphaZero work.”
★ Sorting Things Out: Classification and its Consequences, by Geoffrey C. Bowker and Susan Leigh Star
Recommended by Meredith Whittaker, co-founder and co-director of the AI Now Institute at NYU
“This is an essential text for anyone grappling with issues of AI bias, fairness, and justice.
Whatever else they are, AI systems are systems of classification. In brief, they ‘learn’ what they know from data, and they use what they learn to classify what they ‘see.’ For example, an AI system for hiring might be taught what a ‘promising job candidate’ looks like by [inputting] videos of ‘successful workers.’ Show this AI system a candidate video, and it compares the video to its ‘successful worker’ composite, classifying the candidate as either promising, or not. Such systems are already in use, and the stakes are high: if, for instance, black women weren’t represented among the ‘successful workers’ training videos, then it’s unlikely the system would classify them as ‘promising’ and unlikely that a black woman would ever get hired.
Sorting Things Out engages with the politics and consequences of such classification practices, treating classification not as a reflection of ‘natural categories,’ but as a product of history, culture, and power in which ‘each category valorizes some point of view, and silences another.’ The book examines classification systems ranging from Apartheid South Africa’s racial passbook, which struggled to apply rigid racial categories to diverse human bodies, to the World Health Organization’s International Classification of Diseases, which requires a vast bureaucracy in its attempt to normalize cultural differences in the understanding of illness and health. By attending to these histories, the authors expose the contingency of categories we often take for granted, providing a foundational resource for understanding, critiquing, and contesting the AI systems that are currently automating classification across core social domains.”
★ The Master Algorithm, by Pedro Domingos
Recommended by James Vincent, AI and robotics reporter at The Verge
I’m obviously no luminary in the AI world, but as someone who covers this field for a living, I’ve read more than a few books to orient myself, so I do have some experience here. There are two titles in particular that hooked my interest early on and that I continue to recommend: The Master Algorithm, by Pedro Domingos, and Superintelligence, by Nick Bostrom.
Superintelligence is the book about the threat posed by artificial general intelligence, or AGI, written by Oxford philosophy professor Bostrom. It’s inspired some questionable pronouncements from tech leaders on the threat from killer robots (which deserve to be taken with a barrel of salt in my opinion), but is the best introduction I’ve read to the problem of making smart machines safe; a problem which applies whether they’re super-smart or actually quite dumb. And despite the gloomy topic, this non-fiction book is a surprisingly fun read, feeling closer to science fiction at times.
The Master Algorithm, meanwhile, is a broader read that provides an excellent introduction to the technical aspects of AI. It walks you through all the basic components and concepts, from evolutionary algorithms to Bayesian probability, while showing how machine learning as a field cross-pollinates with disciplines like neuroscience and psychology. Domingos occasionally, I think, overstates the raw power of AI (these aren’t magical systems; they’re often deeply flawed, as other books in our reading list illustrate), but even that is a good reminder of how the very potential of this technology can hypnotize.