Let's talk about sex ... with robots

People often talk to machines, including computers and robots, and a growing number of AI (artificial intelligence) researchers are working to enable them to talk back. And soon, human-computer interactions may include having sex with them. That's the view of David Levy, who has just won the 2009 Loebner prize for the most human-like chatbot.

The annual competition, held last week in Brighton, is based on the "Turing test" inspired by a 1950 paper written by the British computer scientist Alan Turing: that when communicating only via a keyboard, a judge should have trouble distinguishing the software from a human being.

Turning predicted that by 2000, "an average interrogator will not have more than 70% chance of making the right identification after five minutes of questioning". Conversation programs are still a very long way from that, but Levy's Do-Much-More chatbot, developed at Intelligent Toys Ltd in London, has its charms. (An earlier version was called Do-A-Lot, after Eliza Doolittle in Shaw's play Pygmalion.) One Loebner conversation, for example, went as follows:

Judge-3: What do you do when you're not doing the Loebner prize?

Do-Much-More: Oh, I do much more than I intended to do at first. I got more interested in the job as I went along.

Judge-3: Is that difficult?

Do-Much-More: No, it isn't at all difficult. It is as peaceful as a platypus playing with a potato pudding.

Surprise win

Levy last won the competition when it was held in New York in 1997, so why did he wait so long to re-enter? It does, after all, carry a lot of prestige, and this year's cash prize was $3,000. "About 18 months ago, I was approached by an American startup, and I got involved with developing a chatbot for them. So I took some work I'd done after the last competition, and we extended it. I was quite pleased with it, and it occurred to me that the advances in chatbot quality since I first won the prize were really nothing to write home about. So, more as an experiment that anything else, I thought it would be interesting to see how I fared against the cream of the crop. I didn't enter with the idea that I was going to win. It surprised me a lot."

Levy has, of course, seen dramatic improvements in chess computers since "the Levy Challenge": in 1968, he bet £1,000 that no computer program would beat him in a chess match within 10 years. He didn't lose what had become a $5,000 challenge until 1989, and by 1997, a chess computer was capable of beating the world champion, Garry Kasparov. Chatbots started with Joseph Weizenbaum's Eliza "psychotherapist" in the 1960s: why haven't they made similar progress?

"It's a very difficult problem to solve, and to solve any of the major tasks in AI requires a huge amount of effort," says Levy. "One of the reasons computer chess progressed was that the subject was so interesting that there were hundreds of people all over the world working on chess programs, and on the hardware as well. I think that if the same effort was devoted to good conversational programs – if research institutes or governments or corporations threw enough money at it – the state of the art would advance even further."

Well, people nowadays often interact with artificial intelligences in games and on the web, so why aren't commercial needs already driving that investment?

"There are two things about the commercial world: one is to have the need, and the other is to have the confidence or the courage to invest significant resources," says Levy. Until recently there was justifiable doubt whether throwing a lot of money at the problem would produce something good enough to be used commercially. Now companies are probably beginning to realise that it might bring about the kind of advances they're looking for.

"For a program to be commercially successful in this field, it has to be interesting and entertaining over a long period. It's not enough to have someone conduct a conversation for two or three minutes and say, 'Oh, isn't that cute?' "

Of course, AI researchers have developed both chatbots and humanoid or at least pet-like robots, and it seems most likely the two will eventually converge. It's hard to imagine a good companion or carer robot that can't understand what people say, and that might also apply to sex robots. This is an area Levy got to know well through researching his 2007 book Love and Sex With Robots, which he then rewrote as a PhD thesis for Maastricht University in the Netherlands. It caused quite a stir.

"It did, yes, and I was very pleased about that," he replies. "I've done more interviews about Love and Sex With Robots than I have about computer chess!"

Almost human

But so far there hasn't been any commercial interest in adding conversation software to sex robots. "The state of the art is only a little further advanced than the Real Dolls of this world," he says. "There's a Japanese company that has a product called HoneyDoll, which has some electronic sensors. If the man strokes the nipples in the right way, the doll can make orgasmic sounds … There's also an engineer in Germany, Michael Harriman, who has developed a doll that has heating elements so most of the body is warm, apart from the feet."

There's also a lot of AI research going into artificial emotions and artificial personalities; into things such as artificial skin in the medical industries; and in Japan, into carer robots, which the Japanese government sees as the only way of caring for rapidly growing numbers of older people. All these should make it possible to produce far more sophisticated robot companions than Tamagotchi, Furby, Aibo and Robosapiens.

"I think the sex robot will happen fairly soon because the bottom is dropping out of the adult entertainment market, because there's so much sex available for nothing on the internet," says Levy. "I think the market was worth something like $12bn a year, and they aren't going to want to lose all their income, and this seems to me an obvious direction to go. The market must be vast, if you think of the number of vibrators that sell to women. I'm sure a male sex doll with a vibrating penis will sell better than sex dolls today. I'll be surprised if it's more than another three years or so before we see more advanced sex dolls with more electronics and electromechanics.

"There will be a huge amount of publicity when products like this hit the market. As soon as the media starts writing about 'My fantastic weekend with a sex doll', it will be like the iPhone all over again, but the queues will be longer.

"I am firmly convinced there will be a huge demand from people who have a void in their lives because they have no one to love, and no one who loves them. The world will be a much happier place because all those people who are now miserable will suddenly have someone. I think that will be a terrific service to mankind."