12/18/2009 Charlie Johnson, ECE ILLINOIS
ECE Professor [profile:selevins] has been working for the last 12 years on developing a robot that can learn through experience--that is, a robot that doesn’t require specific programming to perform a task, but can process inputs from the outside world and carry out previously unknown functions. In essence, a robot that learns in the same way that a human child learns.
Written by Charlie Johnson, ECE ILLINOIS
There is no shortage of films depicting a not-too-distant future in which artificially intelligent machines rise up to overthrow their human creators. Usually the machines get close, even dangerously close, before a perfectly made-up Will Smith or Keanu Reeves steps in to save the day, get the girl, and remind those killer robots who’s boss. It makes for great entertainment, even if the scenario can seem frighteningly realistic at times.
But ask ECE Professor Stephen E Levinson, and he will tell you he is looking forward to the days of robot/human combat. “We certainly hope that will happen someday,” said Levinson. “If I knew how create robots like that I would absolutely do it in a heartbeat.”
He’s kidding, of course--we hope.
If one wonders why Levinson is looking forward to the robot apocalypse, it’s because he’s been working for the last 12 years on developing those robots. Levinson’s most recent research focuses on creating a robot that can learn through experience--that is, a robot that doesn’t require specific programming to perform a task, but can process inputs from the outside world and use those inputs to carry out previously unknown functions. In essence, a robot that learns in the same way that a human child learns. A sentient robot.
“We’ve been at this for a while,” said Levinson. This project has been going on for around 12 years, and the ultimate goal has always been to build an artificial, cognitively aware machine.”
Levinson’s latest project on the road to a robot dominant/human subservient world is the iCub, a fully anthropomorphic, highly advanced robot that is constructed like a small human. The iCub is built by a European consortium of 30 companies and universities supported with grant money from the European Union. An open call was put out for proposals, the best of which would receive one of the highly advanced iCubs. Levinson’s proposal was one of seven accepted, and the only American proposal to receive an iCub, which has a cost of around 300,000 Euro (approximately $450,000) just for parts and assembly.
The iCub is currently receiving its finishing touches at the University of Genoa in Genoa, Italy, and Levinson and his research team expect to have it in their laboratory at the Beckman Institute soon. Already, several of Levinson’s graduate students have visited the iCub in Europe to begin training on it. Levinson is planning to visit the University of Genoa in January to give a presentation on his research and plans for the iCub.
In order to develop the iCub into a sentient human-killing, motor-oil-swilling super destructor, Levinson and his team are pursuing several different methods of engineering the robot to be artificially intelligent. Most of the team’s research follows standard engineering techniques, but they are also looking into the innovative technique of cortical modeling, trying to model the cortex of a human brain. This extremely complex technique requires Levinson and his team to create a nonlinear, entirely dynamic system to mimic the way in which humans process memories and use them to control future actions. In a nutshell, to create a robot’s brain to learn like a human’s, except instead of neurons and gray matter, the robot would have circuits and resistors to process its thoughts.
“In one sense, this sort of method had to be the right way. But, in the other sense, there is so much knowledge missing right now that it would be easy for this conceptually good idea to go astray,” said Levinson. “These sorts of methods are highly speculative. Our preferred methods, right now, are based on very classical electrical engineering techniques like detection, estimation, pattern recognition, decision making--that sort of stuff. Those are our most promising avenues.”
No matter what method eventually winds up working the best in the iCub, the most important factor in Levinson’s work is that the robot is not being programmed “to do” anything. It’s simply being programmed to learn how to do something. As the robot learns, it will become more and more capable of accomplishing more and more future tasks. It will “grow up,” for lack of a better term.
“This is not about technology, this is about science first and foremost. But, if we could do this, way off in the future it has implications for mental health, prosthetic medicine, defense, or just any kind of task for which you would need a cognitive robot. It’s pretty amazing really.”
And for any of those worried they might see a throng of robots beating down the door to the White House, don’t fret quite yet.
“All we’re trying to do right now is to see if we can get a computational model for brain and mind and then use that to do things that we recognize as ‘difficult’ things that humans do like fine motor control or language.”
But if Professor Levinson’s research continues to be so successful…well…where’s Will Smith when you need him?