The late, great Alan Turing -- generally considered a giant of computing theory and artificial intelligence -- neatly sidestepped this theoretical gridlock by proposing his own functional definition: A computer that responds with enough subtlety and complexity to convince users that it's human is said to pass the Turing Test, and thus meets the standard for true machine intelligence.

Turing Tests are commonly staged as chat-style conversations, since natural language -- with all the subtleties of syntax, semantics and wordplay -- is widely intuited as an unmistakable earmark of real intelligence. There have been many attempts by artificial intelligence researchers to create language generators that pass the Turing Test. One example is the Eliza program -- a popular, decades-old conversation module written to mimic a psychotherapist. A recent exchange between me and Eliza went like this:

Me: Hi, Eliza.
Eliza: How do you do? What brings you to see me?
Me: Not much. How are you today?
Eliza: Does it bother you that are me today?
Me: Uh, excuse me?
Eliza: Why do you say uh excuse you?
Me: You're not really trying very hard, are you?
Eliza: Earlier you said are me today?
Me: Are you trying to tell me I'm crazy or something?
Eliza: Maybe your plans have something to do with this.
Me: Is this visit covered by my HMO?

Eliza quickly betrays herself with the wooden and brittle conversational style of the machine. Clearly she was just parsing my words (in this case, badly) and reflexively shoveling them back, lightly repackaged as questions -- a cheap ruse on the part of the designer to make her appear interested or contemplative. When my statements exceeded a threshold of complexity, Eliza changed the subject -- bad couch-side manner, to be sure, and enough to make me consider therapeutic alternatives like Prozac or electroconvulsive shock. At any rate, Eliza roundly flunks the Turing Test; she is transparently digital and less than scintillating company to boot.

On the other hand, a recent exchange in an AOL chat room was sufficient to convince me that my interlocutor was all too human:

Me: How are you today?
BiteMe100Times: What's it to you? Unless you're my mother or my shrink, you can fuck off.
Me: Just trying to be friendly.
BiteMe100Times: Yeah, but I want a commitment.
Me: Actually, I'm writing a magazine piece on the Turing Test. I'm trying to figure out if you're human. You could be a machine, you know.
BiteMe100Times: Oh, sure I'm human. Two plus two is four. Four times four is 16. Four to the 16th power is [core dumped]
Me: Very clever.
BiteMe100Times: What the hell do you expect? I'm running Windows NT. Now go away.

My chat room partner, unlike Eliza, strongly displayed many of the key features of carbon-based consciousness: sarcasm, irony, misdirected hostility, frustration with Microsoft. Definitely a real person. I took comfort in the proposition that my humanity was secure; differentiating a person from a machine was trivial, even in an Internet chat room.

The Free Software Project
Read Andrew Leonard's book-in-progress on Linux and open source -- and post your comments.