Artificial intelligence is so ubiquitous that even tech agnostics would benefit from at least a working knowledge of its concepts. At the same time, AI’s ever-growing complexity means practitioners need to know the wheat from the chaff when it comes to practical application how-to’s.
To that end, we asked three AI experts to pick some of their favorite books about artificial intelligence.
The panel includes Jana Eggers, CEO of Nara Logics, a machine-learning-powered recommendation engine; Garrett Smith, founder of Guild AI, an open-source machine-learning engineering platform; and Alex Castrounis, an AI consultant and author of AI for People and Business: A Framework for Better Human Experiences and Business Success.
Their selections range from a highly technical consideration of AI’s so-called black box problem to a historical overview of machine learning; from a sober counterpoint to the field’s deep-learning fixation to a thoughtful critique of algorithm bias.
Approachable AI Books for Non-Technical Readers
AI for People and Business by Alex Castrounis
It’s becoming imperative for business leaders to understand artificial intelligence and machine learning at an appropriate level in order to build great data-centric products and solutions. Given that, I wrote AI for People and Businesses for executives, managers and non-technical folks that are interested in leveraging AI within their organization, and to fill a gap that I saw in the AI literature.
I also wrote it for practitioners interested in a business perspective around AI, to give them frameworks they can use to explain complex AI concepts to their company’s leadership. Because sometimes there's a bit of a struggle there. At the end of the day, I think it will help people understand exactly what AI is and help them learn how to identify opportunities with AI. It's really focused on developing and executing a successful AI vision and strategy as well.
And AI is hard to simplify because it's inherently not simple. If you want, you can dive all the way down into vector calculus and matrix and linear algebra and statistics — the list goes on. But it's all about what level of granularity is right for what target audience. This book really simplifies all those very complex things in ways that benefit executives and managers. (Castrounis)
You Look Like a Thing and I Love You: How Artificial Intelligence Works and Why It’s Making the World a Weirder Place by Jenelle Shane
This kind of tells you, Okay, here’s where we are. We're in a nascent state and we need to understand what that entails — where it's strong and where it's not.
The book makes where AI is [in terms of evolution] more real. In my AI talks, I use a lot of examples that come from Amazon, looking at the [curious] recommendations you sometimes get for products and the challenges with that. I'm not picking on Amazon; I chose it because it's something people can relate to.
And that's what Jenelle does; she makes AI relatable. So people understand better where the technology is and some of the challenges that we might be coming across. Because people imagine AI as this beautiful, wonderful magic black box that's smarter than them — and it's not. Jenelle helps ground that for readers, so that they're less scared of it and hopefully engage more with it. It’s a fun, easy read. (Eggers)
Inspired: How to Create Tech Products Customers Love by Marty Cagan
It’s not specifically about AI, but rather about how to deliver technology. It's a great book for everyone from engineers to executives to management. I've given this book to all of them. Engineers have [read it and] been like, “I never understood why it was so hard to work on my teams, and I've been part of the problem!” Or, “I've hated our designer all this time, and now I understand them and what their role is and what my role is!”
Cagan puts together a good framework for how to define and deliver products. The focus is technical products, but it's good for products in general. You can read it quickly. If you have a three-hour flight, you can skim it and still pick up a lot. Marty's very smart and has been in the industry for a long time.
And it was a personal journey for him: He started out as an engineer himself and was on a product that wasn't successful. And he thought, But I delivered exactly on the MRD, or marketing requirements doc, so why did it fail? It's either them or it's me. Who was it? It's a combination. (Eggers)
How the Mind Works by Steven Pinker
It’s not an AI book but it does have a section on building artificial intelligence.It’s just a preposterously good book just in general. It's almost on par with The Selfish Gene type of overarching, broad view of the evolutionary effect on the human brain. How the Mind Works is a very nice high-level view over human brain function, not necessarily directly applicable to artificial intelligence, but it is a very, very good book. And Pinker does talk about AI there. (Smith)
Advanced AI Books
Accelerate: The Science of Lean Software and DevOps: Building and Scaling High Performing Technology Organizations by Nicole Forsgren, Jez Humble and Gene Kim
If Inspired is about how to define the best product as a team, this is about how to deliver it. It's really the DevOps equivalent of product definition. Once you get to the right product, how do you then continually deliver it? And that's especially critical for AI, because you have more change streaming in from both data and the algorithm.
Software development has gone from annual releases to continuous deployment. Not everybody's there, but most people are somewhere on the spectrum. With AI, we have to accelerate. Because not only are algorithms changing, but they then impact the software and technology around them. And you have the data that impacts the AI. Data models are constantly changing because the data is constantly changing. You’re dealing with a much more complex ecosystem, so we really need to adopt those principles. It’s really DevOps on steroids, right? Or chaotic DevOps.
That's why this book is especially important for AI. If Inspired is the foundation, then Accelerate is what you really need to deliver AI. They complement each other — and they're critical for AI because AI is more nebulous. We have to get these definitions down and we have to get delivery down. (Eggers)
Technically Wrong: Sexist Apps, Biased Algorithms, and Other Threats of Toxic Tech by Sara Wachter-Boettcher
You need to understand bias and the problems we can create with these algorithms. There are several good titles on this now, [including] Weapons of Math Destruction and Algorithms of Oppression. I do think that Sara provides many different types of examples that are particularly related to technology and what's happening with the digital transformation, which is where a lot of AI is coming in.
With some bias, the problem is the data has the bias built in. Even if you're not putting the explicit tags of bias — gender, race, things like that — there's so much [that’s] built in and been reinforced because of what the human bias thinks already.
AI will pick up on our generalizations. That's where we need to be careful about what data we give it to learn on. How do we make sure that we're cognizant of what's baked in, even when it's not explicit? (Eggers)
Interpretable Machine Learning by Christoph Molar
This is also very technical, very much a textbook, but it talks about some areas that are quite a bit more directly important to our clients [at Guild AI]. It’s a guide for making black boxes explainable.
Probably the defining problem of our day is that, as you start to become more sophisticated and your models become more complex, the ability to understand those models — why they're doing what they're doing, why they're making the predictions that they're making — becomes much more difficult.
That’s part of the double-edged sword in AI. AI has a lot of promise, but as you start to move toward that promise, your risks go up proportionately — where models do things that are not just mysterious, but potentially quite dangerous depending on the application.
As AI develops, topics of interpretability and transparency are going to come up. And it's going to provide a very serious check to the advancement of AI. Our point of view is that the only way to really keep up with this is to use more math, more data science.
It's like an arms race. As the math becomes more complex in making predictions, the math needed to interpret and understand those models as humans becomes more important and more advanced. So characteristics that allow us to understand why they're saying what they say. And this book is one of very few that really covers that. There are a lot of papers on the topic but very few books. (Smith)
AI Books Focused on Deep Learning
The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World by Pedro Domingos
This book provides a wider framework than just deep learning, which is the hot thing now. Two things to bear in mind: People should know about the different tribes, as the author calls them, and they should also understand that most solutions are going to be ensemble systems, meaning it's not going to be one-tribe-takes-all. It's going to be a combination of several.
You see that even with what DeepMind did with AlphaGo, which used two tribes, arguably even three. So it's a good framework, and it's accessible. For technical people, it's probably going to open their eyes to some things they didn't know about, especially if they just got into AI in the latest craze. And it's also accessible to business people, meaning it's not too technical that they feel like they have to slog through it. It is a little more dry than my next pick, but will give you a spoonful of sugar to go with the shredded wheat — and I like shredded wheat, to be clear.
The author’s correct in that there are tribes and the tribes don't often mix, but I think we need to encourage the tribes to mix more. I challenge with the whole “master algorithm” [idea] because there's not going to be one. Like I said, it's going to be an ensemble. Getting that across, and how to mix and match them [is important]. But I do think it's a great initial framework. (Eggers)
Rebooting AI: Building Artificial Intelligence We Can Trust by Gary Marcus and Ernest Davis
I see this book as being kind of a shot across the bow of the deep learning/connectionist camp, which has sort of taken over the discussion around artificial intelligence … There are several different traditional ML camps; connectionism is neural networks — same idea.
Rebooting AI argues, let's take stock of artificial intelligence, our goals and what useful AI would look like, and ask ourselves, How close to this does deep learning … really get us? The thesis basically is: It gets us down the road in some ways, but in a whole host of areas it doesn't get us anywhere we need to get.
And all the attention applied to deep learning right now is, in the authors’ view, somewhat distracting from other areas that could yield fruit. They're trying to encourage a broader view of AI, revisiting some of the more classical AI camps and disciplines — looking at work that's 40 and 50 years old in some cases as being integral to the advancement of artificial intelligence. It’s a very good book that helps temper the euphoria over deep learning. (Smith)
Deep Learning by Ian Goodfellow, Yoshua Bengio and Aaron Courville
This is one of the best textbooks I've ever read, period. It focuses on deep learning, but it covers the fundamentals of machine learning. It just does a very good job of being very information-dense but also very accessible.
It’s very technical, so it's probably not for everybody. It’s definitely not in the category of popular topics in AI. It's an advanced textbook that would be taught in a graduate-level course, and [readers] would need a number of mathematics prerequisites to understand it. You can read it and get [something from it], but to actually treat it as a textbook, you're in full-on grad-program mode at that point. (Smith)
Neural Networks and Deep Learning by Michael Nielson
A free online book that’s very easy to read and understand, specifically about neural networks and deep learning. It includes a lot of helpful images, visualization and even some videos. And I really like the author's writing style and voice.
I do a lot of speaking engagements and training workshops and I often get the question: I’m trying to get into AI or machine learning; what do you recommend for me? And unfortunately there's not really a one-size-fits-all answer because it really depends on people's learning styles. Some people like videos, some people like podcasts and some people learn better hands-on or reading a book. But I’m a book guy — the one-stop-shop organization and depth of focus. It's hard to piece together that much knowledge and information, just scouring the web for like articles.
And I really like how Nielson (pictured upper left) writes. I even pinged him once, when I first read it a long time ago, just to tell him that I love the casual writing style. (Castrounis)
AI Book Focused on Machine Learning
The Hundred-Page Machine Learning Book by Andriy Burkov
It’s an excellent overview of machine learning, written for practitioners. It covers most areas of machine learning that a practitioner should know about. There’s a good amount of theory and math without being overly technical or mathematically rigorous. I do think that all practitioners should have this on their bookshelf. And it also benefits non-practitioners that want to take a bit of a deeper dive into various aspects of machine learning as well.
I really like how succinct and summarized it is. It’s like a tour of machine learning that could serve as an intro or intermediate book and even as a desktop reference. It’s definitely practitioner-focused, unless you're a non-practitioner who really wants to learn more of the nitty gritty of machine learning.
Every chapter is a similar length, with summaries on all the topics associated with machine learning plus the math behind it, but not like the rigorous derivation of all the equations. This is a step [away] from that — more of a succinct summary/desktop reference nature. But it's a very good book. (Castrounis)
Machine Learning Yearning by Andrew Ng
A great book for practitioners that's similar to The Hundred-Page Machine Learning Book in its broad coverage of machine learning and its application to artificial intelligence. But it's written in a much more how-to- or cookbook-style approach than that book. It's sort of like, if you want to do this, this is how you do it; if you want to do that, this is how you do that.
It's also written in a very logical order that closely mimics the process, key considerations and trade-offs that data scientists and machine learning engineers follow when working on machine learning projects, end to end. It's somewhat unique in that respect. And it's written by Andrew Ng, who's obviously at the forefront. (Castrounis)
