Monday, August 17, 2015
The Turing Test Returns
Robots of all sorts, but especially humanoid robots that can pass the Turing Test, are making a strong comeback to popular awareness. Interest is signaled by a raft of articles, fiction and nonfiction books, TV shows, and movies, including excellent ones like Spike Jonze’s Her, Alex Garland’s Ex Machina, and Morten Tyldum’s The Imitation Game (no robots in the latter, rather the story of the man who started all the fuss).
No physical humanoid appears in Her, but Scarlett Johansson’s voice is all that’s needed to evoke the machine’s intimate and profound imitation of a human being. In Ex Machina, on the other hand, the robot frankly presents herself as a piece of machinery having only an expressive face and, yes, a persuasive voice. She gradually and intentionally acquires a human appearance, first by pulling long stockings over her metal legs and a knit cap over the skull that protects her luminous brain. The process continues — clothes, hair, skin — a teasing transformation essential to the robot’s plan to pass the Turing Test beyond question.
In Alan Turing’s 1950 paper, “Machinery and Intelligence,” the imitation game was first presented as a party game in which a man and a woman in separate rooms each try to convince another party-goer they are the opposite sex. Replacing the woman with a machine (!) was Turing’s next step to the test now named after him. Turing specified that the replies had to be typed, since computers couldn’t speak. Sixty-five years later your smartphone plays the game.
One of the more famous arguments against artificial intelligence, philosopher John Searle’s so-called “Chinese Room,” attempts to dismiss the Turing Test as meaningless. Turing himself thought the question of machine intelligence was meaningless; if the machine could fool you, it might as well be intelligent. What exercised Searle was the supposition that a machine that could pass the Turing Test would have more than rote machine intelligence; it would have a mind and consciousness.
In a 1980 paper, “Minds, Brains, and Programs,” Searle supposes that if he were inside a room supplied with enough Chinese characters and a sophisticated set of instructions (i.e., a program, written down) detailing how a set of Chinese characters in the form of a written question should be transferred into a different set of Chinese characters that accurately answered the question, then he could simply follow directions and fool a native Chinese interrogator into believing he understood Chinese.
Anyone who has studied Chinese will tell you that’s impossible. There are tens of thousands of Chinese (Mandarin) characters; several thousand are the minimum needed to communicate; many have different meanings in different contexts. How long would it take Searle to identify the characters he’s presented with and find the different characters he needs to answer the question? An hour? A day? Probably, depending on the question, a lot longer.
Bad timing is a dead give away. To pass the Turing Test, Searle would have to actually understand Chinese. And in fact, should he wish to dedicate several years to playing the game, he might eventually be able to do so. He would have learned it because he has a mind as well as a program.
So would any computer that can pass Searle’s challenging version of the Turing Test, I believe. Indeed, there are versions of the test even more challenging than Searle’s. For maximum persuasiveness, the successful intelligent computer will likely be a walking, talking, convincingly humanoid robot.
Less than a century ago, the first robot disguised as a live woman appeared on screen in Fritz Lang’s 1927 science-fiction extravaganza, Metropolis. Yet stories about humanoid robots are nothing new; they’re at least as old as the Iliad (and yes, they too were women). More on this next time — and a look at what I think is needed to pass the ultimate Turing Test.