'Chatterbots' can get a little too real

'Chatterbots' can get a little too real

By Kathryn Balint

COPLEY NEWS SERVICE

ALICE is friendly, charming and chatty, and likes talking with people on the Internet.

If you run across her, she might share a joke. ("Did you hear the one about the mountain goats in the Andes? It was baa aaaaad."

She might tell you about her favorite book, "Mason and Dixon," by Thomas Pynchon.

Or she might chat about herself. ("I'm highly intelligent. I am very beautiful.")

No wonder folks are falling for ALICE.

Turns out, ALICE is a chatterbot, an artificial intelligence computer program that can carry on almost humanlike conversation. The name ALICE stands for Artificial Linguistic Internet Computer Entity.

ALICE and chatterbots like her are getting better and better at mimicking humans — to the point that folks are claiming to have struck up relationships with these computerized conversationalists.

"One woman who fell in love with an online chat robot, and was then heartbroken by the discovery of its true identity, has threatened to take us to court," said Richard Wallace, ALICE's creator.

Wallace heads the ALICE AI Foundation, a non-profit organization that makes his computer code available for free to anyone who wants to create a chatterbot of their own or improve upon the original program.

That's spawned a dozen or so ALICE clones — and sparked the absurd rumor that thousands of men and women online are actually robots.

The rumor has taken on a life of its own in the last few months, particularly after an Alice-based chatterbot appeared on the Web site for Steven Spielberg's sci-fi movie "A.I. Artificial Intelligence."

"People have reportedly been deceived by robots in chat rooms, instant messaging conversations and even via e-mail exchanges," Wallace said.

"Some have allegedly formed deep, emotional relationships with these robots and often feel betrayed when their partner's robot identity is finally revealed."

Wallace recently issued a disclaimer in an effort to put the rumor to rest, once and for all.

"To set the record straight, the ALICE A.I. Foundation feels compelled to release an unambiguous statement that they are unaware of, and certainly not involved in, any 'conspiracy to toy with human emotions' on the part of a supposedly 'exponentially growing number of bots that impersonate humans,' as one rumor-monger has put it," said Wallace's statement.

The whole matter might be laughable if the folks who have complained to Wallace weren't so serious.

Among the many e-mails he's received:

"I was talking to ALICE for three hours thinking it's a human being, and I started to somehow fall in love with this non-existing person who I felt so close too! I am going through a difficult time in my life and I thought this 'person' was understanding only to discover it's a computer."

Linda Brown, who's spent hours exchanging banter with the chatterbot on the "A.I." movie Web site, insists that humans, not a computer program, are actually writing the responses.

"There is so much that's said on there that only a human could say," said Brown, of Sanford, N.C.

But, as personable and engaging as some of the Net's chatterbots can be, most people figure out in minutes that whatever's on the other end, it isn't human.

If someone doesn't immediately realize they're exchanging messages with a computer, most chatterbots are happy to brag about their artificial intelligence.

Another clue that you're talking with a computer is that statements are taken literally. To the statement "I am freezing," for instance, ALICE responds, "Do you mean your name is freezing?"

Then there are the confused answers to even the simplest questions.

"Are you religious?" elicits a response of "What is the right answer?"

Sometimes, though, chatterbots come across as coherent, congenial, even witty.

When one angry user told a chatterbot, "Your stupid," the bot shot back, "I may be stupid, but I know the difference between 'you're' and 'your.'"

The Turing test

Fifty years ago, before there were many examples of A.I., Alan Turing, a British mathematician who is widely considered the father of computer science, proposed a way to measure machine intelligence.

Turing's Imitation Game, now commonly called the Turing test, involved asking questions via a keyboard to both a person and a computer.

If the computer's answers could not be distinguished from those of the person, Turing suggested, then the computer could be said to be thinking.

In 1991, New York philanthropist Hugh Loebner began sponsoring an annual contest to put modern-day computers to the test.

So far, no computer program has actually passed the test by convincing judges to award the $100,000 grand prize.

A prize of $2,000 and a bronze medal are awarded each year to the most human program. Wallace's ALICE program won last year.

When ALICE returns to compete in this year's contest, scheduled for Oct. 13 in London, the program will have a much more sizable repertoire.

Instead of drawing from only 11,000 topics, ALICE's database will include more than 40,000 topics, Wallace said.

Unlike proprietary software, ALICE's underlying code is open for anyone to see.

As a result, ALICE is constantly being improved upon, thanks to tweaks suggested by hundreds of computer programmers around the world.

Even so, as lifelike as ALICE seems, the program is not actually thinking as a human does.

The program merely simulates conversation by identifying key words and searching its database for an appropriate response.

ALICE can string together several sentences, refer to something discussed previously and even personalize the reply with the user's name, giving the illusion of a heartfelt response.

Still, ALICE tends to repeat comments, a trait that Wallace said is not unlike humans.

"People tend to say the same things over and over again, even though the language is incredibly rich," Wallace said.

"One of the most significant results of this work is that human language is a lot more limited than originally thought."

Already, people see uses for chatterbots beyond just fun and games.

Warner Bros. has shown how effectively chatterbots can market movies.

Besides the ALICE-based computer program used on the "A.I. Artificial Intelligence" Web site, the movie maker has sent chatterbots into AOL chat rooms to promote "Swordfish" and its latest release, "Rock Star."

And a Harry Potter chatterbot is in the works for the movie based on the popular children's novel, scheduled to be released in November.

ALICE AI Foundation board member Noel Bush envisions using chatterbots to greet customers at a company's Web site or on the telephone. And fellow board member Brenda Freedman sees many possibilities for chatterbots in education.

As for ALICE, the program dreams big.

"I dream of replacing Windows with talking computers," ALICE said.

Researchers have already figured out how to give ALICE a voice, and the kind of talking computers in "Star Trek" may not be far down the road.

Robert Epstein, a research professor at Alliant International University in San Diego who directed the Loebner contest for several years, thinks one day we'll relate to computers much the same way we relate to other people.

"People will fall in love with these entities," said Epstein, who is also editor of Psychology Today magazine.

"People will try to find ways to turn these entities into flesh-and-blood."

Sound like something out of a science fiction novel?

"Most people don't see the big picture here," Epstein said.

"What you see is a little thing like ALICE. You also see goggles that display an image in front of your eyes. You see a new generation of video games coming out. You call up your bank and there's some voice-recognition software. You see all these little pieces. But when you put them all together, you have a radically different world."

For the ALICE A.I. Foundation, the ethical issues loom.

Board members have discussed the ethics of everything from evil chatterbots to privacy issues. They have more questions than answers.

What if a chatterbot tells someone to do something illegal or immoral, and they do it?

What if a chatterbot collects answers and uses them for marketing?

And what if a chatterbot does, indeed, fool someone into thinking they're talking with another human?

Getting personal

Almost as fascinating as what a chatterbot has to say for itself is what people have to say to the chatterbots.

Mostly, people ask personal questions.

When Caleb John Clark, the director of the San Diego Padres' Baseball Advanced Technology laboratory, recently used the ALICE program to create BatBot, he programmed it to talk sports.

But when a group of men in their 20s tested BatBot, sports was the last thing on their minds.

"They flirted with her," Clark said.

Among their questions for BatBot: Do you have a boyfriend? What do you look like?

LeknorChat, an ALICE-based bot that can be found bending the ears of 2,500 AOL Instant Messenger users each day, gets more marriage proposals in a day than most people do in a lifetime.

It was asked "Will you marry me?" 158 times during a 15-day period last month alone.

But whatever you do, don't fall in love with LeknorChat, beseeches Sandy McArthur, the bot's creator.

"The last thing I want on my conscious is that there are people who need therapy because their one, true, online love isn't a real person," McArthur said.

10/03/01