by Leon Rosenshein


And I don’t mean Eliza Doolittle. No, in this case I’m talking about one of, if not the first, natural language processing artificial intelligence agents. I’m talking about the Eliza that came out of MIT’s AI lab in the mid 60s.

Eliza and I are about the same age, and I first ran into it in my early teens. And like any typical teen, I spent a bunch of time getting Eliza wrapped up in knots and saying things that made no sense at all. But in fact, if you took it seriously, you could have what appeared to be a real conversation. Could it pass the turing test? No, but in a limited domain it could feel real.

There are a bunch of things to be learned from Eliza. Not the least of which is that much conversation is shallow, and doesn’t require much, if any, subject matter knowledge. The second is around systems thinking (more on that at a later date) and emergent behavior. Eliza used a very simple script, with some pattern matching and lookback, to generate its responses. Remember, this was almost 60 years ago, and there wasn’t any semantic understanding of the words. But still, some people thought they were talking to a real therapist and say it helped.

It turns out that recently someone found the original ELIZA code and the DOCTOR script it used to act like a therapist. That website has links to the code and lots of interesting info about how Eliza’s original author felt about how Eliza was received.

And if you want to try it out yourself and don’t have an IBM 7094 running MAD-SLIP, check out this javascript version. I’d be interested to hear how your conversations went.