Monday, November 7, 2011

Please Go On

How angry can one get at one's own words? Quite angry indeed. Often, we get angry only over time, once we've had time to reflect on things we've said in the past. But we can get emotional with ourselves much more quickly than that.

We just need to have someone conversing our words back to us. Even if that someone is a computer.

In 1966, Joseph Weizenbaum of MIT created a chatterbot, one of the world's first, called ELIZA. A chatterbot programmed earlier, DOCTOR, used a tiny knowledge base and Rogerian therapy to arrive at diagnoses. (Rogerian therapy, named for a founding father of psychotherapy, Carl Rogers, uses passive techniques, neither agreeing or disagreeing with any statement, and aims to let the patient arrive at their own solution.) Weizenbaum, mocking how little actual knowledge DOCTOR had, decided to create a parody in a field that required even less actual knowledge. Thus, he created ELIZA, which did little more than make DOCTOR from a doctor into a therapist. A therapist can be utterly oblivious to a patient's state and still remain able to maintain conversation.

If you'd like the full blow-by-blow of the program's innards, Weizenbaum goes into that here.

What happened? Well, I just explained to you what ELIZA is about- using various little tricks like keyword searching and canned responses to communicate your own words back to you. Give ELIZA a spin and see for yourself. Take as long as you like.


My, but you're worked up. That's the funny thing here. I told you how it worked. I told you it was just a computer, a very dumb chatterbot, and yet you're still emotional. So were the subjects when Weizenbaum decided to test ELIZA out. Some got so attached that they didn't want to show Weizenbaum the chat logs because they thought it an invasion of privacy.

Weizenbaum actually became frightened of what ELIZA could invoke, so much so that in 1972, he put out a book called Computer Power and Human Reason, the purpose of which was to attack the concept of artificial intelligence, and ELIZA specifically. Much of the book is spent educating the reader as to what's going on in the computer (though as it was published in 1972, it's nowadays very primitive). The result is a book about the morality of computing that remains very well-received today.

Though it's probably fair to guess that, by the time of his death in 2008, Weizenbaum had become fairly dismayed about how extensively AI had been developed.

No comments: