A Book to Make You Less Afraid of AI: Melanie Mitchell's "Artificial Intelligence"
Melanie Mitchell’s Artificial Intelligence came out in 2019. I read it in early 2020 and loved it— it made me wish I could write a book as good as this.
I’m preparing for an AI talk to a general audience in two weeks, so I re-read it to see if it was relevant after the ChatGPT breakthroughs.
It was still relevant. There were only a few minor points of things that might have happened faster than she had thought with ChatGPT.
And I liked it more this second time1.
If you are worried about AI, this book will help ease your mind. If you aren’t concerned, it will still educate you.
The book has three strong themes: the history of AI, details on how the algorithms work, and a level-headed discussion on human intelligence.
The history of AI shows how the goals laid out in 1956 could still be the same goals we have now. The field has made tremendous progress, yet the biggest goals of human intelligence remain elusive.
The history discussion includes the AI Winters, the different paths pursued, and how a tragic accident slowed research in neural networks. The book goes up to current times with the rebirth of AI in 2012 with Deep Learning breakthroughs.
The second theme is about the details of how deep neural networks and natural language processing work. She does this in a way that doesn’t require you to be an expert. I always think that lifting the covers eases the fear of the unknown.
She covers both positives (these algorithms can seem magical and do things that we once thought were impossible) and the negative (they are brittle— if something small changes, you may have to retrain, and the knowledge they gain from training a problem isn’t transferable to even similar problems).
The final theme is the one that runs throughout the book: a discussion on what it means to have human-like intelligence. She makes convincing arguments that these algorithms are far from having the kind of intuition and understanding humans have. She recently published an article at Science.org that touches on this topic.
The book is well worth a read. If you don’t have time for that, follow her on social media for a rational voice in the discussion of AI.
It inspired two new essays I want to write in the next few weeks. One of the essays will be about different ways to think about AI Doom. The other, longer one, will be a better way to think about how people talk about AI.