REVIEW: You Look Like a Thing and I Love You: How Artificial Intelligence Works and Why It's Making the World a Weirder Place (nonfiction book) by Janelle Shane

You Look Like a Thing and I Love You is nonfiction on artificial intelligence. I bought my copy new.

Review:

This came out in 2019, after OpenAI released GPT-2 but well before ChatGPT's release. While I'd love to read an updated work by Shane (no amount of checking has made it poof into existence, alas), as far as I could tell this was still a really useful introduction to how artificial intelligence works and what its strengths and weakness are. Shane lays out what AI is and isn't, how it learns, the various ways it can run into trouble, the instances of disconnect between what humans ask AI to do and what it actually does, and more. 

I first became aware of this work after stumbling on some of Shane's hilarious machine learning blog posts on Twitter (way back when Twitter was Twitter). In fact, the title of this book comes from one such post on AI-generated pickup lines. Still, it sat on my TBR pile for years until ChatGPT came out and became a hot enough topic in academia to be mentioned several times during a Q&A session with a library job candidate.

While I appreciated Shane's humor and adorable little AI illustrations throughout, this also contained plenty of useful information written in a way that was relatively easy for someone without much of a technical background to understand. I'd have liked to see slightly more technical information than Shane provided (for example, I feel like I got a good general understanding of how AI training works, but I still can't picture what actually doing it looks like), but overall Shane's explanations were really clear and made good use of examples. One real-world example that stuck with me that illustrated AI's reliance on its training data and difficulties when asked to do a broader task than it was trained for (because AI does better with narrower tasks) was a self-driving car that had only been trained for highway driving. Its human driver had it take over while it was still in the city and it ended up hitting the side of a semi - it had only ever been trained to recognize semis from the back, so when it saw one from the side it interpreted it as best it could, decided it was an overhead sign, and didn't slow down for it.

I've already recommended this book to several of my fellow librarians as an accessible way to learn about AI and maybe get some ideas for how to talk about it to faculty and students.