Site Presented By
Wednesday, Nov 25, 2009
So what does it mean to be human, anyway? What does it mean to have a mind?
Minsky has written books on what it means to have a mind. So has Dennett. Likewise Hofstadter, Searle, Newell, even Turing. But Wallace doesn't care about such things, and neither, for that matter, does Loebner.
So I asked Tracy Quan about it. She's a writer and former sex worker who has also been a Loebner competition "confederate" (that is, a human respondent to judges' questions), and subsequently a Loebner competition judge.
Tracy and I chatted for quite a while about bots. Bots were amusing, she said. They were stupid, and yet she liked chatting with them because they were good for her vanity. They seemed so interested in everything she said, and were always willing to talk to her.
The overall sense I got from her was that she thought chatterbots were about as interesting as goldfish. Which are, you know, pretty interesting if you're in the right mood for watching goldfish, but really not the kind of thing about which one would write whole books of philosophy. And then she said, "I'm a relationship person. I don't care how the chatterbots work, I just care about my relationship with them. There was this one bot, I think his name was Fred. He was always so complimentary! He was like a flattering boyfriend. We had a very nice relationship."
And then I asked her if she found the idea of artificial intelligence philosophically troubling, in the sense that someday one of these A.I.'s might become more intelligent, wise, funny ... whatever, somehow more human, than any of us.
"What?" she said. "No, you're joking." I said, no, I wasn't. There was a pause. And then she laughed and laughed and laughed.
She was right, of course. Spend a few minutes chatting with even the best of the bots, and you will cease to be threatened by their imminent eclipsing of humanity. Their performance is, in Loebner's own word, gruesome. So I felt pretty silly about all the deep anguish to which I had subjected myself on that score.
And yet, a few weeks later I happened to read the first page of Tracy's novel, "Diary of a Manhattan Call Girl," in which the protagonist, "Nancy," confesses to her diary an embarrassing incident in which she had been found out faking arousal in order to stimulate her client. She was reciting a canned speech used countless times before, but evidently her client, "Howard," knew and didn't mind.
And that was the point. Nancy wasn't so much having a real conversation with Howard as she was engaging in stateless verbal behavior with him, just as an ALICEbot might. "Many people are interested in what happens when you talk to someone. Your behavior when you are speaking is called verbal behavior, and the behavior of the person or persons listening to you (if they respond in some way to what you've said) is called verbally governed behavior." In Howard's case the verbally governed behavior was orgasm.
When you look at it this way you can see that although Wallace's theories of our non-consciousness may be hard to credit, there are bound to be enormous economic benefits to his approach to the Turing test -- just as soon as Nancy's repertoire of canned sexual responses have been typed into the ALICE brain, which they may well be by now.
That's why porn is one of the beckoning frontiers of stimulus-response-style A.I., along with video gaming. Even without much A.I. technology, gaming is a bigger business than Hollywood. Imagine what could come next: A.I.'s that act more or less like goldfish-humans in video games or on porn sites will engage lots of people's interest and earn scads and scads of money.
Even before they pass the Turing test, in other words, chatterbots will become economically significant by evoking the desired "verbally governed behavior." And it seems likely to me that some program based on the plodding, tortoise-like strategy of the bots will pass the Turing test before any sophisticated hare of a self-aware program based on "a truly competent language-comprehension system" from "the world's best A.I. labs."
If Wallace is right, the first "intelligent" machine according to Turing's criterion will indeed be as dumb as a bag of hammers. It will win the prize without ever learning to parse pronouns or deal creatively with enthymemes.
