| Backbone TV |
NEW Geoweb video
|
|
|
| Ultimate computers |
March 17, 2008 |
 Within a decade, artificial intelligence will drive our cars, monitor our health and make communities safer. Right now, though, the best examples are busy playing chess
By Ian Harvey
Would you trust a computer program with your money? How about your life?
Did you pause before answering? The truth is we trust our lives to software and hardware every day, often without realizing it.
Aircraft, for example, operate through autopilot programs and literally hundreds of thousands of lives depend on the artificial intelligence (AI) underlying that software. And people with money in the bank depend on AI to trigger stock trades in managed Registered Retirement Savings Plans.
Do we stay up at night worried we’ll be penniless at retirement, or that planes will fall from the sky? Hardly, and in the next decade we’ll rely more on AI to drive our cars, monitor our health, and make tasks less onerous and communities safer.
And we’ll probably not even notice it’s all being controlled by software. “People don’t appreciate AI and AI research because it moves from interesting technology to invisible technology, like airlines and autopilot,” said University of Alberta professor Jonathan Schaeffer, who chairs the university’s AI program and studies game strategies to further develop AI concepts. “Elevators in large buildings, for example, use AI software to schedule elevators to cut down waiting times. They learn when people push the button at different times so when they’re not being used they move to floors most likely to need an elevator next. Those robotic vacuum cleaners (Roombas) have AI. People just don’t notice it.”
The most pervasive images of AI remain those from Hollywood: evil HAL killing off humans in 2001: A Space Odyssey; the malevolent Terminator sent back in time by Skynet in The Terminator series; or an exploration of whether machines are capable of love and affection in Steven Spielberg’s AI.
But far from Hollywood magic, AI is arguably the highest level of machine functionality and the Holy Grail of the technology industry.
The game’s the thing
On Feb. 10, 1996, some 40 years after the term “artificial intelligence” was coined, the first big example of AI caught the public’s attention. Deep Blue, an IBM computer program, won a single game against the world’s reigning chess champion Garry Kasparov. In 1997 a revamped Deep Blue took an entire match from Kasparov, who later claimed a human was behind the machine, like the Wizard of Oz, directing moves. He demanded a rematch but IBM’s response was to retire the machine and its program, spawning conspiracy theories that further fed Hollywood mythology.
A decade later AI again grabbed headlines when Canadian professor Jonathan Schaeffer’s Chinook program mastered checkers, the conclusion of 18 years of work. After knocking off two world champions, the team withdrew from competition to run a theoretical model on a machine. It was no simple calculation: the board has 5x1020 variable spaces and it took dozens of computers running continuously for 14 years to arrive at an answer: in a perfect game in which neither opponent makes a mistake, the outcome is a draw.
The benefits of this learning, Schaeffer said, are not necessarily immediately obvious. Games, he said, are critical to AI development because they involve variables and predictive abilities, something we humans take for granted. While chess and checkers are two-dimensional games with rigid rules, there are enough variables that predictions are difficult even with the speed of modern computers.
Even more difficult, he said, is his current research around poker, which involves a significant amount of unknown data and one in which successful human players rely as much on instinct and their opponent’s “tells” as their own abilities to calculate the risks. A machine may have the ultimate poker face, but it does not yet have human intuition based on a millennium of body language development.
Games hold a fascination for programmers and Ottawa software engineer Robin Burgener is no exception, but he has taken a reverse approach: using a game to fund his AI development.
The game, 20Q, is licensed and sold worldwide by Mattel as a sub-$20 electronic version of the classic What am I? game. Players think of a subject and the game responds with the first question: animal, vegetable or mineral? Using an algorithm and an on-board database, it then uses up to 20 questions to whittle down the various possibilities to one answer. What’s scary is that it’s tough to beat — even if you cheat. (Try it at www.20Q.net.)
“It learns from every game it plays,” Burgener said, adding that capacity separates it from a simple if-then set of default program commands. “If more people think a dolphin is a fish, it learns that, even though it’s not true.”
But it was in moving the game from its original computer and Web-based home to a handheld gadget that produced a major revelation and insight into the future of AI, he said. “Despite the limitations in taking what it knew and putting that into a small handheld game, we discovered because it is a neural network you could remove entire parts of it and it wouldn’t make any difference,” Burgener said. “It’s similar to a medical case in which a man lost part of his brain while tamping dynamite. But only specific bits were lost and his brain adapted and he was able to function. It’s the same with the game. It’s like a piece of fabric. No matter how big or small it is, it still has the properties of that fabric.”
The root, he said, is a simple algorithm applied on a massive scale. “It’s not hard math, there’s just lots and lots of it,” he said. “The human brain synapses fire every few seconds but there are trillions working together.”
The challenge for AI, he said, is that while computers can process operations faster than a single synapse, they don’t have the capability yet of working across large neural networks simultaneously. 20Q is, however, starting to take on some “human” forms. It even has its own blog, though there may be more of Burgener in Q than he’d admit.
Still, the learning path is critical to his work in AI, not just in paying the bills but in developing other applications. Last year he traveled to Florida to address NASA programmers about using his neural network to guide astronauts through intricate repair and maintenance procedures in the event they get cut off from mission control or are so far into space that real-time communication is impossible.
Helping humans
There are many other areas the AI neuron net could be applied: in the case of disasters like Hurricane Katrina, a handheld device could be useful in managing victim triage. Similarly, a stand-alone device could help identify toxic chemicals in the event of a spill or fire by stepping through a series of questions.
In the case of the former, Burgener is already in talks about commercializing the product, and foresees a more robust version in the future which could help primary-care physicians at the front line. Instead of spending their time reading magazines while waiting to see a doctor, patients might give their history and articulate their symptoms to a device. The answers would be reviewed by a doctor who would explore the most likely prognosis based on his/her experience. The point, Burgener said, is not to replace the physician but to streamline the process and make better, more effective use of physicians’ time, especially in third-world countries where there aren’t enough medical professionals.
For all the advances in AI, however, there still remains a chasm between machines and man. Computers may be able to think faster in solving some challenges but they are not sentient beings. The question then becomes not whether a computer can think, but whether it can feel and, if so, can it cry or otherwise sympathize or empathize? It’s not a trivial distinction. Some researchers believe the critical component of intelligence — which separates the human mind from the animal and mechanical — is that ability to feel. Only when machines can replicate emotions in combination with intelligence, they argue, will true artificial intelligence have arrived.
HotTech Archive
|
|
|
| Green Innovation |
|
|
| Top 300 Issue |
 
|
| Gadget of the Week (Canadian) |
|
Pick the best 3G for you
RIM Blackberry Bold
Choosing the right smartphone is an important decision, and here’s the good news: while both the new iPhone and the Bold are excellent, the feel is entirely different, making it easy to choose.
more>>
|
| Gadget of the Week (Japanese) |
Sounds of Japan
Why record just the visual when you can capture the sounds as well.
more>> |
| Backblog RSS feed |
Click to subscribe  |
|
sponsored by 
