Artificial Intelligence: A Guide for Thinking Humans

Mitchell (Complexity: A Guided Tour), a Portland State computer science professor, ably illustrates the current state of artificial intelligence, debunking claims about computers that match or surpass human intelligence. She begins with a meeting that she attended with Google’s AI team alongside her former PhD advisor, Douglas Hofstadter, author of Gödel, Escher, Bach, who revealed he was “terrified” that a “superficial set of brute-force algorithms could explain the human spirit.” Mitchell then examines various areas of AI research, including image recognition, question answering, game playing, and translation. Each example yields similar results; namely, that computers can be trained to master specific tasks—as with the vaunted Jeopardy! win for IBM’s Watson program—but not to learn new abilities in general or truly understand meaning. Responding to claims by AI developers, Mitchell suggests that machines can never “fully understand human language until they have human-like common sense.” Moreover, AI programs remain susceptible to errors and hacking, in part because they are surprisingly easily fooled. Taking care to keep the text accessible, Mitchell lightens things with amusing facts, such as how Star Trek’s ship computer remains the gold standard for many AI researchers. This worthy volume should assuage lay readers’ fears about AI, while also reassuring people drawn to the field that much work remains to be done. (Oct.)