Brooks: I'm Rodney Brooks. I'm the director of the MIT artificial Intelligence lab.

Stork: What was it like the first time you saw 2001: A Space Odyssey?
Brooks: I had read the book. The book was very exciting to me. And then when I saw the movie it was a complete revelation because I grew up in a place without a lot of technology and I was really interested in AI and then to see that movie, it told me that there were other people in the world with the same sort of weird ideas that I had.

Stork: Did it look completely unrealistic?
Brooks: Of course 2001 was the first movie I saw with computer graphics and those sorts of things so it just looked incredibly futuristic to me. And the model of the space ship which are now commonplace was completely new.

Stork: Tell us about the state of the field of AI in 1968 when the film first came out.
Brooks: Back in 1968 AI was dealing with teletype input where you could type a very simple phrase or equation and the system could echo back with some analysis of it. There was no real time vision. There was hardly any vision at all - maybe of a single photograph taking many hours to compute. There was no speech understanding. So to have a computer which could interact with people on the same basis that people used was something that was clearly far into the future.

Stork: How did you get into the field. Did the film inspire you?
Brooks: From when I was as young as I can remember I was always interested in building intelligent computers. I knew almost nothing about what was going on in the world I just had a few simple books. But the film really inspired me and pushed me to push my whole life towards Artificial Intelligence.

Stork: So, would you say there is a particular philosophy behind the MIT AI Lab? What are you're Lab's general research goals?
Brooks: At MIT these days we are very interested in computer vision and robotics. We are also interested in knowledge access and natural language and some of the more traditional AI systems. But a very strong emphasis on robots and vision.

Stork: What about any particular approach or worldview?
Brooks: The MIT AI lab is a very large lab. 230 people so there are different views that different people have. My own work has been very much motivated from a bottom up point of view. Looking at how systems evolved overtime, and starting with very simple systems - I worked on insect robots for a long time. And slowly building that up towards more human level interaction. But not coming at it from the knowledge end. Coming at it from the basic underlying mechanisms on top of which knowledge can be laid.

Stork: Give us some examples with your little insect robots.
Brooks: When I started building walking robots, I built six legged robots and rather than have them compute ahead of time a stable way of walking I made the legs very sensitive to things that they touched in the environment and made it so that it was safe for the robot to fall down and had it learn to scramble over rough terrain, feeling its way as it went instead of sitting back looking at the whole terrain in front and computing the optimal path through, my robots got in down dirty with the environment and interacted with the environment at every step.

Stork: Do you think that's how real biological systems do it?
Brooks: I think that evolution has had to find just the very next little incremental capability and as a result it is always just on the edge of being able to do something and it does it rather poorly and over time evolution gets it to do it better and better but it is based on very simple solutions initially which then get elaborated. That's a little different from I think the world view of many people in AI who look at intelligent behavior, very cerebral behavior and start it with that and try to break that down into logical engineering pieces that need to be put together to generate them.

Stork: Now you talked about evolution. What about the role of learning?
Brooks: When we look at biological systems, even the simplest systems have learning capabilities in them. There is a worm that people have studied - Sea Elegens - it has about 900 cells, 300 of them are neurons and that exhibits every form of learning that we see in the higher animals. So clearly evolution has found it very useful to put learning into all the systems it builds, all the animals that it builds.

Stork: So what have been some of the successes of this general approach?
Brooks: By doing this bottom up system we have been able to build systems which rely on very little computation. So now we are starting to see toys getting into the market which are cheap because they use simple processes which use this approach. The Sojourner rover that went to Mars was based on a project that we started at JPL back in 1989 which was called Tooth, and then it became Rocky Two, Rocky Three, and sojourner was Rocky Six. And when Sojourner was on Mars, ultimately after the primary mission and secondary mission were over, it was allowed to operate autonomously using this approach.

Stork: So what about Humanoid robots?
Brooks: Inside every AI researcher is someone who wants to build the ultimate intelligence. And I think all of us are ultimately inspired by HAL. We want to build that intelligence at the human level. From my work I came to believe that it was important to have a body and to interact in the world and to interact with people in a human like way in order for the system to be able to learn how to be human. I think in hindsight that HAL was a technological mistake in thinking that it could be as intelligent as it was without a body. So I've been pushing at looking at robots with a human form so that people know how to interact in the natural sort of way. And it has the same sorts of interactions as the developing baby has as the baby grows up.

Stork: What kind of interaction?
Brooks: When 2 people are talking to each other they make eye contact, they break eye contact, that's how we know when someone's paying attention. They nod, they say yes, they give this very simple feedback which is cross-cultural and lets us know how the conversation is going. When we try to build a system without those feedback cues it can become very disorientating knowing where we are going. And so that's why those speech systems that we talk to over the telephone seem so infantile. Because they have to spell out every step of the way or we'll get lost knowing what they understand. By exploring, building a robot with human form we can just rely on the natural cues that we understand.