I’m going back through the books I’ve read in 2021, so I have all my synopses in one place that’s not Goodreads.
You Look Like a Thing and I Love You by Janelle Shane
The author of the hilarious AI Weirdness blog delivers an overview of machine learning, what it’s capable of, and in particular, where it fails
Assumes no prior knowledge of machine learning, but doesn’t over-explain things like many popular science books are guilty of doing. Gives a realistic assessment of the limitations of machine learning algorithms, instead of the often hyperbolic descriptions that talk as if we’re already living in a sci-fi future. Has a few passages with the same types of lists as the AI Weirdness blog, with hilarious failures based on weird prompts. Simple cartoons of over-eager ML algorithms are throughout the book and never fail to be charming. I wasn’t aware how much image recognition algorithms want to see giraffes.
If you’re expecting a compilation of the blog, as I was, you’ll be disappointed, since there are only a few of the hilarious lists. On the other hand, if you were expecting a thorough description of how ML works, you’ll be disappointed, since it never quite went into enough depth for me. Although I’ve got a CS degree and several years of experience as a programmer, I’ve only got the barest understanding of the specifics of how ML is implemented. So when Shane casually mentions simulated robots teaching themselves how to hop on one leg or jump into the air, I can’t picture how that would actually work.
Great, charming, topical overview of the current state of machine learning and realistic expectations we should have for and concerns about this nebulous idea of “The Algorithm.”