Artificial intelligence. Those words often conjure extreme visions: a shiny techno-future of self-driving cars and hyperefficient production, or a dystopian doomscape of mass unemployment and robot overlords.
In her first book, You Look Like a Thing and I Love You, Janelle Shane assures us that AI is more like a toaster than like Skynet from Terminator. It’s a tool—one that is really good at some things and really, really terrible at others. This accessible guide to AI and machine learning cuts through the techno-hype and shows how AI is making the world a stranger place.
You may know Shane from her humor blog, AI Weirdness, where she trains AI to perform all kinds of silly tricks, such as coming up with pick-up lines—the source of the book’s title. You Look Like a Thing follows in the blog’s footsteps, with plenty of hilarious AI fails, neural-net-generated recipes for “Basic Clam Frosting” and robots doing the can-can. But the humor isn’t just a gimmick. Shane, who holds advanced degrees in engineering and physics, unpacks the whack, using incidents of AI weirdness and abject failure to reveal the peculiar logics of machine learning. Why do image-identifying AI obsessively find giraffes everywhere? What made Google Translate interpret nonsense syllables as biblical verses? How could a self-driving car mistake a truck for a road sign? In unraveling all the ways AI can go wrong, Shane illuminates its inner workings.
Where Shane really excels is in spotlighting the social consequences of AI. She explains how an algorithm widely used to decide whether to release prisoners on parole systematically identifies black inmates as higher-risk for reoffending than comparable white inmates. In another example, a resume-screening bot routinely penalized women applicants. The data that AI are trained on are often more important than the design of the neural networks—and these datasets unintentionally reflect the explicit and implicit biases of the cultures that produced them. After reading this book, you’ll never again assume that an algorithm is neutral.
This is why You Look Like a Thing and I Love You should be essential reading, even if you never end up training your own pet neural-net. AI is already busy all around us—determining what we see (and don’t) on the internet, deciding whether we qualify for a loan, finishing our sentences. If AI unleash a dystopia, it won’t be because the algorithms have outwitted us but because they’ve been pigging out on our data, amplifying our worst tendencies and biases.
AI is a tool, not a slippery slope to the singularity. If we don’t understand how this tool works, what it excels at, how it fails, then we can’t use it to shape a better world.
