Kirkus
Copyright © Kirkus Reviews, used with permission.

Amid the flood of optimism about artificial intelligence, the significant dangers must be understood and assessed. Suleyman might seem like a strange person to write a book about the dangers of AI. He is the CEO and co-founder of Inflection AI, and, before that, he co-founded DeepMind (now owned by Alphabet), a company working at the leading edge of AI research. As the author shows, however, it is precisely because he is an expert that he knows enough to be fearful. He believes that within a few years, AI systems will break into the broad public market, placing enormous computing power in the hands of anyone with a few thousand dollars and a bit of expertise. Suleyman recognizes that this could bring remarkable benefits, but he argues that the negatives are even greater. One frightening possibility is a disgruntled individual using off-the-shelf AI to manufacture a deadly, unstoppable virus. Other scenarios range from disrupting financial markets to creating floods of disinformation. Suleyman accepts that the AI genie is too far out of the bottle to be put back; the questions are now about containment and regulation. There is a model in the framework established by the biomedical sector to set guidelines and moral limits on what genetic experiments could take place. The author also suggests looking at “choke points,” including the manufacturers of advanced chips and the companies that manage the cloud. The key step, however, would be the development of a culture of caution in the AI community. As Suleyman admits, any of these proposals would be extremely difficult to implement. Nonetheless, he states his case with clarity and authority, and the result is a worrying, provocative book. “Containment is not, on the face of it, possible,” he concludes. “And yet for all our sakes, containment must be possible.” An informative yet disturbing study and a clear warning from someone whose voice cannot be ignored. Copyright © Kirkus Reviews, used with permission.

Publishers Weekly
(c) Copyright PWxyz, LLC. All rights reserved

“An emerging cluster of related technologies centered on AI and synthetic biology... will both empower humankind and present unprecedented new risks,” according to this shrewd debut. Suleyman, cofounder of the artificial intelligence companies DeepMind and Inflection AI, chronicles the technological advances that led to today’s AI boom with anecdotes from his career, describing DeepMind’s 2012 work on an algorithm capable of teaching itself simple computer games and the company’s 2018 breakthrough developing a program capable of predicting unknown protein structures. According to Suleyman, four features distinguish these new technologies: the high speed at which they’re developing, the broad variety of uses for them, their ability to function relatively autonomously, and their capacity to affect “entire societies” (he mentions the possibility that “a single system could control autonomous vehicles throughout a territory”). Regulation is the key to dodging dystopia, he contends, outlining 10 steps for keeping AI under human control, including ensuring that all AI have a “bulletproof off switch” and requiring government-issued licenses to produce “the most sophisticated AI systems.” Suleyman’s account of DeepMind’s achievements can come across as self-serving, but anecdotes about other companies working on technologies capable of, for instance, interfacing directly with the human brain, underscore the mind-bending possibilities. It’s a sober take on navigating the perils of AI. (Sept.)