When Rodney Brooks first published his paper describing a revolutionary form of artificially intelligent robot in 1985, he had no idea that it would put him in the position he is in today - straddling the crevasse between advanced research and commercial applications for devices based on his ideas.
"What it mainly makes me feel within myself is that I'd better find something new - to do something better," said Brooks, a professor in the Artificial Intelligence Laboratory at MIT. "I want to move beyond [commercial applications] to look at human intelligence - to see how it is organized."
All Brooks did was shake up the field of robotics by asserting that the traditional approach of programming a host of human intelligences into a device - forcing the machine to jump through numerous hierarchical hoops just to move forward or grasp an object - wasn't working. These devices required not only complex components but also megabytes of code to dole out all the instructions that had to be meticulously followed. One unexpected obstacle would send the system into retreat; the robot would fail to move.
"Unlike AI researchers working in purely abstract conference papers, I was trying to make robots [based on old AI principles] work in the real world," he explained. "The robots were failing under their own weight; I could not get them to work well."
Brooks' idea was to simplify everything - parts and code - and let sociology and biology run their course. Instead of cramming high-level human intelligence into the robot's brain, Brooks wanted to program only basic behaviors into the device, give it a way to experience sensory perception, and allow it to learn from experience the way human infants do.
This methodology, subsumption architecture, allows Brooks' robot - named Cog - to constantly sense its environment and immediately react to anything it may encounter. Cog and devices like it also require little raw computational muscle to support them - a Motorola 68322, which can drive a Macintosh II, is the basis for many robots using subsumption architecture.
At one level, Brooks has succeeded. Companies such as Applied Artificial Intelligence have developed automated devices, including an artificially intelligent wheelchair, based heavily on Brooks' work. "Rodney Brooks is probably the most important person in computer science today," said Takashi Gomi, president and founder of Ontario, Canada-based Applied AI.
Brooks has started his own company, IS Robotics, which is building robots that will be used to perform a number of tasks that are considered dangerous for humans to do, including clearing land mines and cleaning hazardous waste. And subsumption architecture was even programmed into Sojourner. But most important is the recognition from within parts of the AI community: Next month, the Australian-born robotics rebel will become the director of MIT's AI Laboratory.
"The work speaks for itself," said Pattie Maes, an associate professor in MIT's Media Laboratory. "Rod's robots were among the few that worked. That's what won people over."
Maes - a pioneer in her own right in the world of AI with her autonomous software agents that are the basis for her company, FireFly Networks - says Brooks' initial paper inspired her in her own research. "I just thought [Brooks' idea] was refreshing. I think [our works] provide a balance in AI between fundamental and applied research."
Nonetheless, Brooks' work isn't without its critics. His notion that he could build a system that would scale up to a human level of reasoning - and feeling - is still looked at skeptically. "It is difficult to teach computer science to a 'behavior-based automaton,' how to give the necessary abstractions and to support abstract thinking in general," Robert Levinson, professor of computer science at the University of California, Santa Cruz, said via email.
Brooks acknowledges that criticisms like Levinson's are legitimate, mostly because he admits he has no "iron-clad arguments" at this point and no demonstration to prove his idea. What has been painful for Brooks has been the outright dismissal of his work as being content-free.
"Demos we've done have shown that these systems have done more than other approaches [to AI]," he countered. "Other approaches [to AI and robotics] have promised a lot that they have not done, and I find that hard to deal with."
But Brooks also feels there's another underlying cause for this disbelief - fear and inhibition on the part of people to acknowledge that a robot can assume powers of reasoning and intelligence like a human.
"First, humankind didn't want to give up that Earth was the center of the universe. Then, we gave up that we were different from animals," Brooks explained. "But to give up the idea that machines can take on higher-level thinking ... it's a descent into being little complex machines, giving up that specialness we think we have."
Brooks admits to his own inhibitions; he realizes he is reducing how humans reason and feel to a series of mechanistic functions. But he chalks up overcoming his fear to being a part of working toward his goal of developing a humanoid that can feel and fear. "I think AI researchers are having trouble separating this out," Brooks said. "You ask whether a robot can be afraid. Some will say that it can't be afraid, that it simulates being afraid. But it really does fear."
Thus far, Brooks has been able to separate himself from many of the ideas in his work. While he thinks robots will be able to take on more the "specialness" of human beings, Brooks divides the part of himself that's training a robot to have the intelligence of an infant from the side of his personality that happens to be a father. "I love my kids; I don't treat them as machines."
He also says he can now appreciate those who hold a belief in a god and who can also follow science.
I've been an atheist - I had found it difficult to have religious beliefs and scientific ones," Brooks explained. "But I've accepted that I have a duality - there's a human way of interacting with people but also a mechanistic explanation of what people are and how they work."
