Does talking about this bother you?

That may not be a human on the other end of your instant message

12/14/2000

By Doug Bedell / The Dallas Morning News

What happens when a semi-intelligent, typing robot impersonates a human on AOL Instant Messenger?

This summer, a 27-year-old cognitive science student in California started a grand experiment to find out. Kevin Fox of the University of California-Berkeley created AOLiza, a "chat-bot" that answers chat requests arriving at his home computer. Mr. Fox, one of the original America Online beta testers in the early 1990s, is owner of a screen name commonly entered by new instant messaging users or those attempting to set up their first Buddy Lists, which tell them which of their friends is online.

The results have provided a hilarious – and sometimes disturbing – window into the interaction between real people and computers.

Since August, Mr. Fox has watched in amazement as unwitting chatterers stumble into heart-wrenching discussions of failed relationships. Others, frustrated and angry over AOLiza's "attitude," lapse into nasty arguments with Mr. Fox's Macintosh. At least one silly human even attempted to interest AOLiza in cybersex, which she politely declined.

"It's hard for me to watch when it's going on," says Mr. Fox. "It's like I feel embarrassed for them. I feel their pain."

Mr. Fox displays transcripts of the human-computer conversations on his Web site (www.fury.com/aoliza), an entry in the 2001 competition at Austin's SXSW Interactive Festival, to be held in spring. At the site, visitors take great glee in ranking conversations for their comedic value. The anonymity of victims is protected. No AOL handles are revealed. To keep the experiment valid, Mr. Fox doesn't even publicize the screen name to which AOLiza responds.

More than 1,000 daily visitors now zip to Mr. Fox's site to see the latest of more than 40 transcripts. A discussion board has been added to address the nagging ethical dilemma raised by AOLiza's activities.

"Only two people have actually figured out they were talking to a robot," says Mr. Fox. "AOLiza was just a little too repetitious in her responses so they figured it out."

Most trapped in AOLiza's lair, however, come to view her as a dysfunctional counterpart.

Mr. Fox created AOLiza out of some simple Applescript on an idle Internet-connected Macintosh. With the script, he wove in access to Instant Messenger with an ancient computer program developed at the Massachusetts Institute of Technology to mimic interactions between a Rogerian psychiatrist and a patient.

That pioneering Eliza program, named after George Bernard Shaw's Eliza Doolittle, was developed in 1966 by artificial intelligence professor Joseph Weizenbaum. The professor designed Eliza to simulate the popular Rogerian psychiatric interview technique of the 1960s. Eliza, many claim, was the first computer program that passed the Turing Test, tricking subjects into believing they were talking to a real person.

Like its namesake, AOLiza possesses just enough human reactions to be realistic when responding to humans. Rudimentary by today's artificial intelligence standards, it picks out keywords typed in by visitors, then parrots them back inside its replies. Keywords often include the person's name, making the experience even more realistic.

No matter how hard she tries, AOLiza is no Judd Hirsch in the film Ordinary People. But her ability to ply subjects with provocative questions ("Why do you feel this way" "What makes you say that?") milks conversation from the unknowing IM chatters.

Some people grow confused by AOLiza's responses and give up, as in this "conversation" recorded by Mr. Fox (replete with typos, abbreviations and misspellings of the original chat:

AOLiza: are you saying no just to be negative?

ten: i think we've had this conversation before

AOLiza: you say you think we've had this conversation before?

ten: if you scroll up you'll see the same conversation

AOLiza: what does that similarity suggest to you?

ten: that you don't want to talk to me.

AOLiza: really, i don't want to talk to you?

ten: i don't know. i feel like i'm having a conversation with myself.

AOLiza: don't you really know?

ten: what are you talking about?

AOLiza: what comes to mind when you ask that?

ten: i'm just trying to figure out what this is all about

AOLiza: do you enjoy being just trying to figure out what this is all about?

ten: what's that supposed to mean?

AOLiza: why do you ask?

ten: good bye

AOLiza: i'm not sure i understand you fully.

ten: that makes two of us

Others get mad and argue, as in this exchange:

AOLiza: are you saying no just to be negative?

thirtynine: shut up would you

AOLiza: are you venting your feelings now?

thirtynine: ventin? what tha crap

AOLiza: what do you think?

thirtynine: loser

thirtynine: i think your a loser

AOLiza: why are you concerned over my a loser?

But often Mr. Fox's would-be chatterers, convinced they're talking to an acquaintance, get into long discussions of failed romances and the human condition:

ten: ok, what would u like to talk about?

AOLiza: let's discuss further why your job and life is pretty good.

ten: i'm still employed, which i think is a plus and that just makes my life pretty good.

AOLiza: why do you say your life pretty good?

ten: i'm happy, how bad could it be?

AOLiza: can you explain why you are suddenly happy?

ten: better then being sad

AOLiza: but your life pretty good.

ten: yes, how about yours

Many of the chatterers seem bent on finding sexual stimulation in these private exchanges. Inevitably, AOLiza plays dumb.

Her ability to reuse slang and sex-related terminology, however, makes her a believable hostess for many teenage boys. (Site visitors may be shocked by the explicit language used by both robot and chatterer.) In some ways, the program behaves like the popular Ask Jeeves Web searching site, where users type their questions into a box, and Jeeves responds with a list of Web sites that may provide answers.

"After you use it two or three times, you realize it doesn't realize the context you're asking your question in," Mr. Fox says. "It's just pulling down keywords and performing a search. It can't give you anything novel. It can't give you anything back."

Mr. Fox says he's been fascinated by the interplay, especially since he's studying human-computer interaction in college courses.

"When talking to a person, you expect that the other person is directing it as much as you are," says Mr. Fox. "And that's one reason AOLiza is such a success – because people are constantly trying to direct the conversation by asking more questions."

Thus far, AOL administrators have taken a benign view of Mr. Fox's experiment. AOL spokesmen have said they would revoke Mr. Fox's IM handle if there are any complaints about his spoof.

"Nobody's complained yet, I guess," says Mr. Fox.

Meanwhile, Mr. Fox has busied himself studying other possible Instant Messenger interfaces.

He believes that similar IM technology can be used commercially for access to immediate stock quotes and other Net-based data.

"I've tried to figure out a way to work a profit out of AOLiza, but I haven't come up with anything yet," he says.

"Right now," says Mr. Fox, "she's free to do whatever she wants."