MOC: Do you think its possible that we're ever going to going to have human level intelligence?
Brooks: My own belief is that we are biochemical entities, we are machines made up of little tiny machines which are biomolecules. That said, in principle we should be able to build machines with other technologies, maybe build our own machines out of biomolecules which are as intelligent as people in principal. The question is whether we are smart enough to figure out how to do that and how to put those together. But I have no hesitation in thinking that a machine can be just as intelligent, just as real as a person in principal.

MOC: Does the future scare you?
Brooks : I think our science fiction movies and 2001 was one of them, have got us thinking that as we build these intelligent machines they'll want to hurt us, they'll want to take over from us. First I don't see why we would necessarily think that's the case. But in any case I suspect that there isn't going to be an us for them to take over from because I think that while we're building these more intelligent machines we're going to be engineering our own bodies to increase our own functionality more and more over the next few years. First by putting steel and silicon inside our bodies which is starting to happen, connecting it to our neural systems. And then by using biotechnology to alter ourselves so there isn't going to be those intelligent machines and us old humans, there's going to be those intelligent machines and us changed humans in the future.

MOC: Can you explain the idea of agent architecture. You once said that maybe all that we are as humans is a constellation of simple behaviors. What did you mean by that?
Brooks: When you look at biological systems simple biological animals, you find that they have very simple behaviors which then get suppressed by other behaviors which came along evolutionarily later which let them do something better than they could do before or something that they couldn't do before at all. So it seems as though evolution gets a creature which can just barely do something and then rather than tweak up that particular behavior, comes along and glops on top of that another piece of behavior which maybe modulates the lower level behaviors in order for it to be able to do something better or give it some new capability. So when I started building insect robots we started using that methodology and it was very successful to build robots which became very adaptable to their environments. Given that evolution uses that approach, it seems to me that that's perhaps all that we are: we're just glops and glops of more stuff put on top of simpler hacks which produce what we are today. People don't necessarily like that idea because they feel that it devalues what it is to be human. But I instead marvel at it, at the complexity of what we can get out from building up from the bottom, which is what evolution had to do to build us.

MOC: Would you summarize what your thesis showed to you and at least two disparate lessons that came from it.
Brooks: My PHD thesis was, I think, very much an old style AI thesis. I built a computer vision system, which was able to look at aerial photographs and understand what was in those images - it actually looked at airport scenes and identified airplanes. It used an incredible amount of knowledge, it had thousands of rules which all fired and tried to find out what was the only possible interpretation of what it could see. As I continued after my PHD and tried to apply those ideas to robots and robots being able to go and grasp something I got more and more frustrated because I had to build more and more elaborate models with more and more details and I looked at simple animals and they didn't have enough neural power in them to be doing all these complex computations. So I got disillusioned with that approach of having all the knowledge in the machine and that's when I started realizing that knowledge didn't have to be in the machine it could be out in the environment.

MOC: Could you make a distinction between the old AI and the new AI.
Brooks: What I call the old AI relies on trying to sense the world in some way, except the old AI never actually does the sensing, that's for someone else to deal with, sense the world and build a complete internal world model in all detail where everything can be reasoned about. When you want to reach out and grasp something you know exactly where it is and you make a plan for how to get out there and grab. And you can do it with your eyes shut at that point because you've planned exactly how to do it. The new AI relies on seeing what's out there in the world using the idea that the world is its own best model and senses what's in the world continuously in order to carry out that interaction with the world.

MOC: The idea of embodied intelligence; can you explain that.
Brooks: So the new AI relies on the body being in the world, the body being part of the world. Its not the disembodied HAL sitting there away from everything else, it is HAL having a body being out there in the world. And to me that was my big disappointment in the movie 2001. That HAL really couldn't do much stuff in the world it was just this intelligence sitting back there using the astronauts as his remote controls to carry out the mission.

MOC: Will they be human-like?
Brooks: As we build theses intelligent robots the question is, "are they going to be like humans or like something else?" And my feeling is unless they're like humans we probably may not feel much empathy for them or feel that they are intelligent. We see that with dolphins. We can't agree whether dolphins are intelligent or not because they are so alien to us. And I think that is because they develop in a totally different environment to us, they have a different set of experiences in the world, in a different world, and so they don't do the sorts of things that we can map to in the same way that we can map to our dogs. We understand they're not to smart but we can understand their emotions and what they're seeing and what they're understanding at some level. And that's one reason I wanted to build robots with human form on perhaps the misguided chance that they would understand the world in the same way we understand the world and would therefore would develop in a more human-like way.

MOC: Do you foresee some point when all these agents will recreate our self, our internal state?
Brooks: Our robots are fairly simple still, but nevertheless when they are operating people feel in some sense that they are interacting with a creature. And then if the robot goes into a calibration phase...it surprises people because they get intuitively that it is a creature when all its behaviors are running and then it gets lost when it goes into its calibration phase. So I think that over time as we make these robots more and more complex with more of these behaviors that illusion - if you want to call it that - will last longer and longer. Then the question is, 'When is the illusion the real thing?' Is the illusion the real thing, as I hypothesize, just when it's of good enough quality and not something completely different in kind?

MOC: Why do we need robots?
Brooks: There are lots and lots of uses we can and will have for robots. I think in the next 50 years our world will become more and more populated with robots in the same way that it has become populated with computers over the last 20-40 years. How many of those robots are going to be these emotional empathetic sorts of robots is another question. It may be that not many of them are. But there may be some of them. And by studying how to interact with these emotional robots I think we'll find technologies which will be more general through the more day to day robots that we have everywhere. I don't think robots are going to be very useful if every time you want to use it you have to program it in C++. You want to be able to show it stuff and be able to say, 'Go over there and clean up that pile of leaves.' Have that level of interaction if robots are going to be truly useful. If you everything you want it to do you have to go through a menu of this and that its going to be harder to use them. So finding ways for natural interaction with robots, I think is important.

MOC: Can you give us some sense of how hard these all these problems are that you're working with.
Brooks: In terms of intelligence, our robots now are pretty simple compared to where we would want them to be ultimately. And so I think we're more in the phase of hot-air balloons over Paris in the 1790s Whereas what we'd really like it to be is 747s flying across the Atlantic and Pacific. There's a long way to go yet.

Brooks, holding a robot which looks like a baby doll:

Brooks: This is a robot that was built at my company, i-robot coop and Hasbro is now selling. And it's a robot which understand how it's being played with and it develops over time. So it's got ticklish feet, it understands how it's getting moved about, and it has an emotional model inside it. It has desires and those desires need to get satiated occasionally. But at the same time it is responsive to how the child plays with it. Let's see if it will take its bottle. So this is a robot with insect level intelligence. This is the sort of intelligence system that we were building 10-12 years ago and it has finally made it into a commercial consumer product. And there is a gap between what gets done in the lab and what's going to show up in the world of about that length. It doesn't seem very hungry, no it just want s to play. You never know what's going to happen because it does have its own internal life.

MOC: What can it sense besides touch?
Brooks: It's a trade secret. My Real Baby has about 15 sensors in its body, some of which are used to sense when its thumb is in its mouth or its bottle is in its mouth. Others are used to sense how it's being held and how it's being moved. Its got touch sensors, its got internal motion sensors and then it has light sensors under its clothes so then it can tell whether its being held like this or whether its like this. And now putting the thumb in its mouth got it the idea that it is hungry and it wants more and I've got to give it the bottle. Now it's started to feed.