Best Books on AI (and the Podcast Episodes to Hear First)

Some links below are affiliate links, so we may earn a commission if you buy through them, at no cost to you. Here’s how that works.

AI books go stale faster than milk. Anything about specific tools is out of date by the time it’s printed, and plenty of the rest is either breathless hype or doom. So I picked books built on ideas that hold up even as the models change.

Two are about using AI well at work. Two are about where the risks really are. One is a skeptic’s book, and you need at least one of those. One’s the big-picture thought experiment. And each comes with a podcast episode, which matters here more than anywhere, because these authors keep updating their thinking as things move.

Quick Glance

1. Co-Intelligence by Ethan Mollick

Pair it with: Coaching for Leaders #674, “Principles for Using AI at Work”. Listen to the episode

Mollick is a Wharton professor who seems to have tested every AI tool before most of us heard of it. His book is the most practical one here: treat AI as a strange new coworker, always invite it to the table, stay the human in the loop, and assume this is the worst AI you’ll ever use.

The “jagged frontier” idea is the gem. AI is shockingly good at some tasks and quietly bad at others that look similar, and you only find out which by using it a lot.

Read it if: you want to get better at using AI in your actual job.

Skip it if: you want theory or long-term predictions.

Why this episode: It’s a short, focused interview on applying his principles at work. A good one to send to your team.

2. AI Snake Oil by Arvind Narayanan and Sayash Kapoor

Pair it with: The TWIML AI Podcast #704, “AI Agents: Substance or Snake Oil”. Listen to the episode

Two Princeton computer scientists separate AI that works from AI that doesn’t. Their sharpest point: generative AI is real progress, but predictive AI (tools that claim to forecast who’ll commit a crime, succeed at a job, or drop out of school) is often snake oil sold to institutions that should know better.

It’s the skeptic’s book every AI reading list needs. Not anti-AI, just anti-nonsense.

Read it if: you buy, evaluate or make decisions about AI products.

Skip it if: you’re already deeply skeptical and want reinforcement. Read Mollick instead to stretch.

Why this episode: Narayanan applies the book’s thinking to AI agents, which is the hype cycle happening right now.

3. The Coming Wave by Mustafa Suleyman

Pair it with: 80,000 Hours with Mustafa Suleyman. Listen to the episode

Suleyman cofounded DeepMind and now leads AI at Microsoft, so this isn’t an outsider’s warning. His argument is that AI and synthetic biology will spread faster and wider than any technology before them, and that “containing” them is close to impossible but necessary anyway.

It’s sweeping and a bit grand in places. It’s also the best book for understanding how the people building AI think about the risks.

Read it if: you want the big picture from someone at the center of the industry.

Skip it if: you want practical advice for this quarter.

Why this episode: Rob Wiblin pushes Suleyman on uncomfortable questions, including whether his own company is speeding up the race he’s worried about.

4. The Alignment Problem by Brian Christian

Pair it with: 80,000 Hours #92 with Brian Christian. Listen to the episode

How do you get a machine to do what you mean rather than what you said? Christian covers the history of that problem, from biased hiring models to AI that games its own reward, through interviews with the researchers working on it.

It’s the best-written book here. Christian has a gift for making technical ideas feel like stories, and you’ll come out understanding machine learning far better than before.

Read it if: you want to really understand how AI systems learn and fail.

Skip it if: you want a quick read. It’s thorough.

Why this episode: It’s a long episode, and Christian goes deep on the research, which makes it a great companion for anyone who wants more after the book.

5. Human Compatible by Stuart Russell

Pair it with: Future of Life Institute AI Alignment Podcast with Stuart Russell. Listen to the episode

Russell co-wrote the standard AI textbook, so when he says the field’s basic approach is flawed, people listen. His point: we build machines to pursue fixed objectives, and a powerful machine pursuing the wrong objective is dangerous. His fix is AI that stays uncertain about what we want and keeps checking.

It’s more technical than Suleyman and more focused than Christian. Short, clear, and genuinely important.

Read it if: you want a serious expert’s proposal for making AI safe.

Skip it if: you’re looking for workplace tips.

Why this episode: The interview was recorded right as the book came out, and Russell walks through the whole argument step by step.

6. Life 3.0 by Max Tegmark

Pair it with: Lex Fridman Podcast with Max Tegmark: Life 3.0. Listen to the episode

Tegmark, an MIT physicist, imagines the long-term futures AI could lead to, from utopias to takeovers, and asks which ones we actually want. The opening story about a team that secretly builds a superintelligence is worth the price alone.

It’s the most speculative book on the list, and some of it reads differently now than when it came out. It’s still the best one for thinking big about where all of this goes.

Read it if: you like big questions and thought experiments.

Skip it if: you want grounded, near-term analysis. Read Narayanan instead.

Why this episode: This was the very first episode of Lex Fridman’s show, recorded as part of an MIT course, and it’s a relaxed, wide-ranging conversation.

The Reading Order I’d Actually Use

Read Co-Intelligence and AI Snake Oil together. One makes you an enthusiastic user, the other keeps you from getting fooled. That combination is the right attitude for most people.

If you want to go deeper on the risks, read The Alignment Problem next, then Human Compatible. Save Suleyman and Tegmark for when you want the view from 30,000 feet.

FAQ

What is the best book on AI for beginners?

Co-Intelligence by Ethan Mollick. No technical background needed, and you’ll use it the same day.

Are AI books outdated quickly?

Books about specific tools are. The ones on this list focus on ideas (how AI fails, how to evaluate it, how to work alongside it) that hold up as models change.

Which AI book is most skeptical?

AI Snake Oil. It’s careful, evidence-based skepticism rather than doom.

Should I read The Alignment Problem or Human Compatible first?

The Alignment Problem. It gives you the background that makes Russell’s proposals easier to follow.