Pages

Friday, April 21, 2017

Review: Deep Thinking

I tweeted this a few days ago:
I've now read Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins and can say that I was not disappointed. It's a thoroughly absorbing book, most of which I read in one sitting. It did confound my expectations in a way, though: somehow I'd expected there to be less chess in it. In retrospect, I realize that reflects the extent to which I've come to think of Garry Kasparov as a writer and activist no less than as a chess player; but of course chess was and always will be a central element of his career, and so it should be no surprise that computer chess, and in particular his battles with IBM's Deep Blue, occupies much of the space in this book. Even so, he does a capable job of tying that history--deeply personal for him--into broader considerations about human and machine intelligence.

As Kasparov aptly notes, chess is not a great measure for intellectual ability in general. Notwithstanding the game's demands for memory and concentration, correlations between chess skill and general intelligence are weak. For all its dazzling variety of possibilities, chess is not deep enough to have required the sort of machine learning that recently enabled a program to beat the top human player of the game Go. Rather, chess programs rose to Grandmaster level through simpler programming--using brute-force techniques of searching through numerous positions and assigning values (based on material strength of pieces and other readily measurable factors) rather than developing strategic concepts, let alone any understanding of the psychology of an opponent.

Consider the ascending level of difficulty an AI would face in accomplishing the following tasks (this hierarchy is my notion, not Kasparov's):

1. Playing chess really, really well. (Done.)

2. Playing a randomly selected chess variant really well, with different rules, pieces, shape and size of board, etc. (I considered this possibility a decade and a half ago in a perhaps slightly weird article.)

3. Inventing a new chess variant or other board game, with an emphasis on one that has many novel, unexplored possibilities. (Bonus: marketing the game and persuading people to play it.)

4. Subjectively experiencing any of the above, including feeling emotions about its activities.

5. Deciding unprompted to move on to a new career, such as writing, running for office, etc.

A computer that does all that does not seem likely to arrive anytime soon, even as AI logs genuine accomplishments in a wide range of fields. Kasparov, who has about as much reason as anyone to be chagrined about computers' capacity to outperform humans and take away their jobs, takes a persuasively optimistic view, seeing technology's potential to expand rather than replace human abilities. He notes the high performance that human-computer combos have had in the chess world, and that in addition to human and computer ability a key role can be played by "process"--being the best at meshing the two to maximize their strengths. He makes many other interesting observations, such as about the danger of excessive reliance on optimization--fine-tuning existing abilities--as a substitute for innovation that opens entire new avenues and measures of what can be done.

Fears of computers taking away jobs--or taking over the world--may soon give way to fears of humans doing bad things with the vast computer resources now at hand. I suspect the latter danger will get growing attention as more is learned about how the Russian government and its allies (or pawns) have deployed hacking and data mining in American and other elections around the world. As it happens, Kasparov has been a prescient observer of the geopolitical danger emanating from Russia.

The rapid advances in computing and other technologies require, well, deep thinking, and this book by that name deserves a wide audience, including but not limited to chess aficionados.

Note: I'd originally intended and been commissioned to review the book for the magazine Scientific American Mind, which unfortunately is no longer appearing in print form, hence my choice to write a review here at my own blog instead.