All products featured on WIRED are independently selected by our editors. However, we may receive compensation from retailers and/or from purchases of products through these links.
SEATTLE -- In the 1983 movie War Games, Matthew Broderick's character teaches a wayward Pentagon computer about no-win situations by training it to play Tic-Tac-Toe.
Now an Israeli software-development company called Artificial Intelligence N.V. (Ai) is planning to use games to learn some fundamental lessons in how to teach a computer to think.
The company took the occasion of the 17th International Joint Conference on Artificial Intelligence this week to announce a competition to test computer learning algorithms by having them play a series of simple games without knowing any of the rules.
The programs will be given a list of the allowable moves, each represented by a meaningless symbol, said the company's chief scientist, Jason Hutchins. By choosing their moves and getting scores back, the programs are supposed to "learn" the most effective strategies to win the game. The contest, which will take place over a month's time, will be a round-robin competition of between six and 12 games such as "Rock/Paper/Scissors."
"Each of the games are being created to train specific abilities," Hutchins said. "What's exciting is that we will have the Computer Science Department at MIT competing against a 12-year-old hobbyist from Australia."
The first notice of the competition was made by putting cards into the registration packets that attendees received when they came to the conference center. Hutchins said the response was so far above what they expected that Ai's network managers thought they were seeing a denial-of-service attack. By Tuesday afternoon, after less than 48 hours, 500 contest packets had been downloaded from the company's website.
Ai is best known for creating a natural-language learning program -- nicknamed Hal after the computer nemesis from 2001, A Space Odyssey -- that is being taught to speak English simply by being talked and read to.
During a break in one of the conference sessions, Hal's trainer, neurolinguist Anat Triester-Goren, said Hal has made some of the same errors in learning as real children do.
Hutchins said that people who have read transcripts of Hal's conversations were unable to tell them from transcriptions of real toddlers talking. Having a language program that people cannot tell from the real thing -- known as the "Turing test" -- is the Holy Grail of one branch of artificial intelligence research, he said. He adds that most people forget that the test's inventor, computer pioneer Alan Turing, also said the best way to accomplish the goal would probably be to build a "baby machine" and train it.
Smart bots: Rodney Brooks, director of the MIT AI Laboratory and chairman of iRobot Corp., told a packed room Thursday about his company's experiences trying to bring intelligent robots to the mass market over the past several years.
One device iRobot is presently working on releasing is what he referred to as a "physical avatar," a remote-sensing bot that can be controlled from any Web browser with a high-speed connection.
Because of the inevitable delays, the device needs to have enough on-board intelligence to maneuver itself around in response to commands like "go into the next room."
The current version, built with off-the-shelf parts, has remote vision and two-way voice capabilities. One use he suggested was for remote attendance of business conferences. When a member of his staff did that recently, he said, the 'bot was able to follow participants around the room as well as talk and listen.
"It's hard to predict what's going to happen" when you put a new technology like this into people's hands, he said.
In one particularly galling case, iRobot built a robotic floor cleaner for stores and office buildings that could sweep, mop and polish floors, as well as navigate itself around obstacles and overlap the cleaning areas. The cleaning companies were thrilled to have a machine that could do all three things at once, Brooks said, but having it be self-controlled was too much for them.
To sell it, the company had to take out the robotic components and sell it as a machine for human janitors to operate while following behind it.
We're gear, get used to it: Manuela Veloso, a computer science and AI professor at Carnegie-Mellon University, said robots are already making themselves useful in a number of settings, including hospitals and nursing homes -- where they are taking over some of the drudgery usually associated with practical nursing; and in the operating room -- where remote-controlled robots are performing delicate operations in spaces too small for surgeons to reach directly.
"There is much more use of AI incorporated into the real world than people realize," said Sony Corporation's Hiroaki Kitano, whose division developed the Aibo robotic pet. "When it comes to it's maturity, it becomes invisible." At the same time, he questioned whether humanoid robots that could mimic a range of human activities was even a worthwhile goal.
"Do we really need robots who can do many things well?" Kitano asked. "Computers can do number-crunching far better than mere humans. If computers were capable of many things, but could do calculations only twice as well as humans, would anyone want them?"
