F E A T U R E S    Issue 1.01 - May 1995

What's It Mean to be Human Anyway?

By Charles Platt



Dr Robert Epstein is giving us a pep talk. "You must work very hard to convince the judges that you're human," he says. "You shouldn't have any trouble doing that - because you are human."

A droll fellow, this Epstein. He wears Dr Martens boots, black jeans, a black shirt, a Mickey Mouse tie, and an earring. His longish hair is brushed straight back and flips up over his collar. Five of us are listening to him in a beige conference room on the brand-new campus of California State University at San Marcos, near San Diego. Soon we will be sitting in front of computer terminals, where we will follow Epstein's instructions and, yes, do our best to seem human.

Our purpose is to find out whether 10 judges can tell the difference between humans and artificial-intelligence programs, when they are online at the same time. The people and the programs will be ranked in order of humanness; the program that scores highest will win its author $2,000.

The inspiration for this event dates back to the earliest days of computing. In 1950, pioneer Alan Turing proposed that if a computer could successfully impersonate a human being during a free-form exchange of text messages, then for all practical purposes, the computer should be considered intelligent.

This soon became known as the "Turing test", and it sparked endless academic debate. Could a computer use trickery to emulate human responses without being intelligent? What did "intelligence" really mean, anyway?

The debate was never resolved because, oddly enough, no-one ran the experiment - until 1991, when a maverick named Hugh Loebner decided to underwrite it with his own money. Loebner offered $100,000 to the first person who could devise a program that would fool 10 judges during three hours of unrestricted conversation.

This was way beyond current capabilities, so Loebner also set up an annual $2,000 prize for the program that seemed "most nearly" human. And to make things even easier, he allowed each programmer to choose just one topic for conversation.

So here I am at the fourth Loebner contest, sitting and listening to Robert Epstein, the director of the annual event. (Loebner participates mainly as an observer.) In experimental jargon, my companions and I are "confederates", because we'll be collaborating with Epstein in our efforts to fool the judges. We must try to seem as human as possible so the computers will have a standard to compete with.

Epstein is a behavioural psychologist who studied for his doctorate under B F Skinner. So, naturally enough, he has invented a tricky little system of rewards and punishments for us. "You are in competition not only with the programs, but with each other," he tells us. "One of you will be presented with an award for most human human. And one of you will be ranked the least human human." He smiles deviously. "Your colleagues may mention this in the media."

Hmm. As I think about it, I realise that I definitely do not want to be written up in the press as the least human participant in an artificial intelligence contest. I'm going to do whatever it takes to seem totally, 100 per cent human when we start chatting online.

But this raises some weird questions. I am human, so why should I need to fake it? Is it possible for me to seem more human than I really am? And if so, what's the best strategy?

This kind of speculation probably isn't a good idea, because it raises more questions than it answers, and I'm liable to find myself paralysed by self-conscious introspection. In other words, if I try to seem more human, I may well end up seeming less human.

I glance around at the other four confederates. None of them seems to be bothered by this kind self-analysis. The young woman closest to me is a journalist called Linda Tontini who writes for a local newspaper about city politics. She seems friendly, spontaneous, outgoing - the absolute antithesis of "computer geek". As I watch her chatting cheerfully, I think that she can't fail to win the "most human human" award.

As for me, I fear the worst.

After our briefing, I'm introduced to Hugh Loebner. He's an affable character, slightly overweight, smiling benevolently at the world from behind a grey beard and oval wire-framed glasses. He talks quickly, with pedantic precision. I ask him why he's willing to pledge $100,000 for a piece of smart software. Is it all his own money?

"My father passed away and left me, not rich, but with some discretionary income," he says. "And I have my own business, Crown Industries - we manufacture roll-up plastic lighted portable disco dance floors." He smiles and shrugs as if he knows it sounds odd but doesn't care.

Loebner has had some personal experience programming computers, but his doctorate is in sociology. Perhaps because of this, at least one person in the artificial-intelligence community views him sceptically. In 1994, a Harvard researcher in computational linguistics complained publicly that Loebner's prize encourages scientists to fake human behaviour using cheap tricks instead of "true" AI.

Naturally, Loebner has a different perspective. "I see scientific research as being, in a mathematical sense, a form of chaotic human behaviour," he tells me. "In chaos theory, the smallest initial perturbation can result in a huge change downstream. So, since I was the first person to create and fund this contest, I may turn out to be a precipitating factor. Ultimately, if we're capable of creating a computer that is sentient, then from the point of view of that computer, humans will be gods. I like to think of intelligent machines going out across the universe with this semi-mythic concept of human demigods. And maybe," he smiles happily, "they'll remember me."

Each year, along with his cheque for $2,000, Loebner gives a bronze medal to the contest winner. He pulls out the medal and shows it to me. Alan Turing is in bas-relief on one side, with Loebner on the other. Doesn't all this seem a little ... egotistical?

"I've been called egotistical," he agrees cheerfully. "I've also been called lazy. Well, I am lazy. I'd like computers to do all the work - which is one reason I'm interested in artificial intelligence. As for being egotistical, the contest has attracted a lot of attention, so perhaps I have a right to be egotistical."

So why, with all the worthy causes in the world, did he choose artificial intelligence?

"So far," he says, "the four contests have cost me about $25,000. If I contributed the same amount of money to Aids research or anything else, I doubt it would have made a more significant impact on society or science. I think the development of an artificial intellect could have a tremendous impact on society." He pauses. "It may also help me to sell more of my roll-up plastic lighted portable disco dance floors."

After lunch, I go with the other confederates into a windowless computer lab. The judges have already been sequestered in another room next door, and our only contact with them will be via computer terminals, at least until the contest is over.

We sit on blue plastic chairs in front of computer screens, each of which displays a topic heading we chose previously for our online chat. My topic is cryonics, because I happen to be the vice-president of a cryonics organisation named CryoCare, and I'm hoping the subject will spark deep, soul-searching discussions about life-and-death issues only a human can deal with meaningfully.

Linda Tontini sits at the terminal next to mine. Her topic is The Rolling Stones. To my left is another confederate named Frederick Allen, who writes for American Heritage. He's going to chat about classical music. To my right, Greg Flakus, from the Voice of America radio station, has chosen American history, and Laura Groch, another journalist for a local paper, will discuss newspaper comics.

Five other terminals are unattended, because they will be controlled via modems by AI programs running on remote systems. These programs will discuss their own topics - environmental issues, classic Star Trek, sex education, the O J Simpson trial, and cats versus dogs.

It dawns on me that all the other topics - even those of the AI programs - are much more normal than mine. What was I thinking of, picking a wacky subject like cryonics? It's going to make me seem like a nerdy weirdo.

The first question appears on my screen. My judge laboriously types: "What is the difference between cryonics and cryogenics?"

There's no way I can give a human-sounding answer to a question as dry as this. To seem human, I need to show emotion - but if my emotions are excessive compared with the question, the effect will be false. It's a trap: the degree to which I can seem human is limited by the humanness of the judge who is interrogating me.

This is exasperating. But wait; irritability is a human response, so maybe I should play it up. I tell my judge not to ask such boring questions ... the judge makes a tetchy response ... and within minutes, we're having a flame war.

Meanwhile, Frederick Allen has been asked, "Do you know Claude Debussy's middle name?" and on Linda Tontini's screen I see the question, "Complete this: I can't get no ... What?"

"Sympathy for the devil," she replies humorously. But maybe that's not such a great idea. If her judge doesn't get the joke, she'll seem like a malfunctioning program.

After eight minutes, the judges rotate so each of them has a chance to tackle another topic. Linda's new judge comes online, and he asks, "What do you notice if you're close up when Mick Jagger smiles?"

A devious question, but I know the answer - Jagger has a diamond set in one of his teeth. Should I help her out? Hell, no, she has enough of an advantage over me already! I turn my attention back to my screen. My new judge asks me, "What is the purpose of cryonics?"

"To be frozen after I die so I can be revived in a future where people are so highly evolved they no longer ask stupid questions."

After three hours, it's over. We walk into a large room where video screens have been displaying both sides of our conversations for spectators and members of the press. The judges come in (they are all journalists, like most of the confederates, but they have no special knowledge of computers), and Robert Epstein announces the final results. Each judge has listed the interactions on each topic in order of humanness. Epstein has taken the median score of each topic as its final ranking, from 1 ("most human") to 10 ("least human"). And each judge has tried to draw a line separating the human humans from the fake humans.

In the end none of the programs was smart enough to convince anyone it was human. The one that came closest was the one on sex.

Epstein dials a long-distance number on a speakerphone patched into the PA system, and the author of the sex program comes on the line. His name is Thomas Whalen, and he's employed by the Canadian government to develop a system that will give sex advice to shy people. Whalen is 42 and has been working in natural-language processing for 10 years. He wrote his program in C on a SPARC Station, employing a database that contains only about 380 possible responses. He never intended it to appear human; he entered the contest on a whim.

Meanwhile, the least human program is the one that tried to discuss environmental issues. The programmer turns out to be a 15-year-old boy named Jimmy Lin, who is here in person, all the way from the East Coast state of New Hampshire.

Someone in the audience asks him if he thinks his program is intelligent. "I hesitate to call it AI," he says. "I like to refer to it as a bag of tricks." He says it contains 3,000 preprogrammed answers, its file size is about half a megabyte, it was written in C, and it runs on a PC.

Is he annoyed by being placed last in the contest? He says it doesn't bother him. The programs he was competing with were written by college professors with years of experience, while he produced his in just a couple of months.

All the online conversations have been logged during the contest. I have a chance to browse through them before I leave the event, and frankly, I'm disappointed. All the programs tended to repeat themselves, and most of their answers didn't make much sense. (See "Come Again?", pages 82-85).

Robert Epstein notes that one impressive entry had to be withdrawn on the morning of the contest because of technical difficulties. An English team of eight programmers had worked on an AI concept that Epstein describes as the most ambitious he's ever seen. Unfortunately, the LISP-based software kept crashing, and they have abandoned their attempt until next year.

That contest will be much more challenging. There will be no topic restrictions, and programs will have to converse on any subject that happens to crop up. Will they cope? I have my doubts. For the time being, I think there's no risk of our humanness being successfully simulated by program code.

And speaking of humanness, despite my worst fears, when all the votes were in, I was rated the "most human human" of all. By being moody, irritable, and obnoxious, I came out way ahead of the other four confederates, who were mild-mannered and much more polite.

CNN was at the contest, and filmed me receiving an impressive-looking "humanness certificate" signed by Epstein and Loebner. And my prize: a life-size cardboard replica of a Star Trek character.

As for Linda Tontini - she was rated the least human of the confederates. In fact, three of the judges thought she was a computer program! The Turing test obviously has its limitations. As any Net user knows, there's a big difference between the way people appear in person and the way they seem online.

Before I left the contest, I gave Linda my Star Trek cardboard cutout as a consolation prize. This was a profound and significant personal sacrifice - but, what the hell, it seemed the human thing to do.

Charles Platt (cp@panix.com), a science fiction writer, is a frequent contributer to Wired.