A.I. Needs People to Set the Objectives

(p. A2) Although Deep Mind’s Alpha Zero can beat a grand master at computer chess, it would still bomb at Attie Chess—the version of the game played by my 3-year-old grandson Atticus. In Attie Chess, you throw all of the pieces into the wastebasket, pick each one up, try to put them on the board and then throw them all in the wastebasket again. This apparently simple physical task is remarkably challenging even for the most sophisticated robots.

But . . . there’s a more profound way in which human intelligence is different from artificial intelligence, and there’s another reason why Attie Chess may be important.

. . .

The basic technique is to give the computer millions of examples of games, images or previous judgments and to provide feedback. Which moves led to a high score? Which pictures did people label as dogs?

. . .

But people also can decide to change their objectives. A great judge can argue that slavery should be outlawed or that homosexuality should no longer be illegal. A great curator can make the case for an unprecedented new kind of art, like Cubism or Abstract Expressionism, that is very different from anything in the past. We invent brand new games and play them in new ways.

. . .

Indeed, the point of each new generation is to create new objectives—new games, new categories and new judgments. And yet, somehow, in a way that we don’t understand at all, we don’t merely slide into relativism. We can decide what is worth doing in a way that AI can’t.

. . .

. . . , we are the only creatures who can decide not only what we want but whether we should want it.

For the full commentary, see:

Alison Gopnik. “MIND & MATTER; What A.I. Is Still Far From Figuring Out.” The Wall Street Journal (Saturday, March 23, 2019): A2.

(Note: ellipses added.)

(Note: the online version of the commentary has the date March 20, 2019, and has the same title as the print version.)

Alison Gopnik’s comments, that are quoted above, are related to her paper:

Gopnik, Alison. “AIs Versus Four-Year-Olds.” In Possible Minds: Twenty-Five Ways of Looking at AI, edited by John Brockman. New York: Penguin Press, 2019, pp. 219-30.

Leave a Reply

Your email address will not be published.