Book Review: You Look Like a Thing and I Love You by Janelle Shane

You Look Like a Thing and I Love You by Janelle Shane

Computer artificial intelligence is an interesting topic because the possibilities seem full of endless promise. The actualities of AI are a lot more eye-opening in their successes and failures. Computer scientist Janelle Shane has written a book chronicling the past ten or fifteen years of work in the artificial intelligence field, focusing on how weird and unexpected things turn out.

The main challenge for computer AI is that it is designed by humans. Humans craft questions or challenges for a computer and the computer does its best to reply. If the question is not worded carefully enough, a solution is often inappropriate or unhelpful. The computer solves the problem of preventing people from choosing the left fork by killing all the people who try to take the left fork. Or another AI succeeded in never running into a wall by standing still. Often, AIs work in a simulation of reality designed by humans. The simulations often have flaws that the AI exploit in achieving a solution. Another AI was tasked design a body with arms, legs, torso, and head that could go from Point A to Point B. The AI stacked the body parts so that when the body fell at Point A the top end of it landed on Point B. That's a solution that works...in the simulation. Humans need to design questions, challenges, and environments carefully to get workable solutions.

Another problem for AIs is the sample data they are given. The author designed an AI to come up with pick-up lines (where she got the title for the book). The suggestions are hilarious because they look like a parody of real pick-up lines. She gave the computer lots of examples of pick-up lines and the data set led to some natural consequences (maybe pick-up lines are inherently ridiculous?). A more serious problem is found with AIs that screen employment applications--often they mimic the prejudices that go along with historical hiring practices. An application with "Harvard" or "Cambridge" on it will move up in the ranks; an application with an address from "the wrong side of the tracks" will move down. Computer scientists have recognized these flaws and tried to adjust with mixed results. Flawed data leads to flawed conclusions whether you are an AI or a human.

Computers are brilliant at working through hundreds and thousands of iterations to get better and better at performing tasks. The future is indeed full of possibilities; it's important to know what is actually possible.

Recommended.