Bookshelves are groaning from the weight of many volumes dissecting the emergence of AI. Some like Kaifu Lee’s AI Superpowers have already been summarized and critiqued by me. (Looi, 2020) However, The Age of AI features a lead author, Henry A. Kissinger, who appears unlikely to pen such a study, if only because he is 98 years old! Indeed his expertise is statecraft and geopolitics. Joining him are two authors more typically aligned with the subject, Eric Schmidt (Google) and Daniel Huttenlocher (MIT). Moreover, it is possible that Kissinger’s assistant, Meredith Potter, did most of the research, fact-checking, and drafting.
Yet, it is Kissinger’s perspective that lends the book its broad context in two important ways that differentiate it from other analyses: the meaning of AI in the evolution of human reason and the impact of AI on geopolitical matters. A slim book, it mostly raises questions and does “not pretend we have all the answers”. (Kissinger, 2021 p 6)
The authors begin citing two AI breakthroughs: AlphaZero, a chess playing program, and an MIT project using AI to find a new antibiotic. Both AIs showed something unexpected: the solutions did not correspond to how humans have been solving similar problems. For example, AlphaZero developed strategies that no human grandmaster had ever contemplated, including sacrificing valuable pieces, humbling them by showing that it had uncovered insights that had eluded a lifetime of study by grandmasters.
The MIT AI program found a molecule to inhibit a particular bacteria by “identifying attributes that had eluded human conceptualization or categorization”. (Kissinger, 2021 p 9) The researchers achieved this by training the AI to find molecules with structural patterns that had previously been effective at fighting certain bacteria. The intelligence didn’t know why they were effective, just that they were. But it also found new molecular relationships that endowed them with bacteria-fighting ability; thus the AI “detected aspects of reality humans have not detected, or perhaps cannot detect.” (Kissinger, 2021 p 11) Similarly, GPT-3 is an AI that can generate human-like text given some initial inputs.
With the help of AIs, we are developing a new way of interpreting reality that in some cases is different from human…