A few times a year, I wander into the book shop, looking for treasures. A couple of years ago I started following Janelle Shane on twitter. She’s the author of a book called “You Look Like a Thing and I Love You”, which is how I greet my wife now. I liked what she wrote on her blog, AI Weirdness, and what she shared on twitter, so when I saw her book on the shelf I grabbed it knowing that it was coming home with me.
I read You Look Like a Thing and I Love You on the plane ride home from my brother-in-law’s wedding. There were movies and tv shows available, I even had downloaded a video game for my tablet, but the book was engrossing enough that I never even thought about breaking those out during the flight.
This is a rare book: I’m comfortable recommending it to anyone, with no caveats.
My mom? Yep. My Father-in-law? Yep. My next door neighbor who moved in yesterday and I haven’t met yet? Again, yep. The book is fun and informative. It never talks down to you, and is patient in its examples.
One of the themes running through the book is text generators, and what machine learning figures out regarding language is incredible, but also understandable. You get to see the computer learn language and format, and you get to see it take an idea from total gibberish to something that makes a certain kind of sense.
Janelle Shane asks the AI to learn from several types of text sources and then generate new ones. The recipes are chaotic, crush some half and half, why don’t you? The pick up lines are childlike and funny, the title of the book is a pickup line that the AI came up with. The knock knock jokes are one that really stands out, since you get to follow the AI learning them, all the way from gibberish to a respectable pun.
I won’t spoil much, since watching what AIs say and output is a highlight of the book.
And while text generation is one of the funnier things to me, it covers so many more topics. What are neurons? When are they active? Why is everything a giraffe? You’ll find out these questions and more.
The book isn’t just funny, it’s actually informative without being condescending. Some people talk about AI as if it’s on the verge of skynetting and deciding that it doesn’t need humans. Rather, AI does what we ask of it, and it takes the path of least resistance to get there. Perhaps this is scary - Shane doesn’t hesitate to point out that emulating racism and gender discrimination not only have happened, it was entirely predictable.
One of my favorite chapters is titled “Unfortunate Shortcuts” (favorite is relative, they’re all good but this one had me cackling). There’s a lot of real world examples in this chapter like the time they train an AI to identify a particular kind of fish, called a tench. What they realized the AI was recognizing was human hands holding the fish against a green background. All of the training data looked like that, so that’s what the AI learned to look for.
Another thing to love about the book is the illustrations. They’re super cute, and they always add to what you’re reading, or help illustrate the point being made. There’s an illustration of the Tench picture that confused the AI, and it comes at the perfect moment to make you go “Ohhhh, I get it.” And that’s really the highlight of the book, it gives you so many of those moments, where the realization comes with a laugh.
A lot of people make wild predictions about what AI will be doing soon, Janelle Shane cuts through all of that with this book to not just tell you, but to show you what AI is doing now. This book is so clear, so fun, and leaves you with a better understanding of AI. Most importantly, this book leaves you with a smile.