They caused a sensation when they first appeared in the 1990s but now robots COG and Kismet are not enough for the man who helped create them. Rodney Brooks, professor of robotics at MIT’s Artificial Intelligence Lab, wants more. In his new book Robot: The future of flesh and machines, he argues that there’s something lacking in our maths. We need “new stuff” that will zero in on the vital difference between living and non-living systems and help us transform robots from the lumbering arms of the car factory or the single-use cute little home machines. Duncan Graham-Rowe caught up with Brooks in London recently.
When did you build your first robot?
I was somewhere between 8 and 10 years old and living in Adelaide, Australia, when I started trying to build computers. By the time I was 12, I’d managed to build one that could always win at noughts and crosses. But I didn’t quite have the dexterity or the mechanical ability to build a robot, although I wanted to. I was 16 when I made my first robot. It would wander around the floor, attracted to light, and hit obstacles and move away from them. It was particularly good at getting stuck under chairs.
Advertisement
Have we come far since then? Can we really build super-smart robots using current techniques?
If we look at our models of living systems, or aspects such as modelling evolution, we have really interesting results. Tom Ray and Karl Sims who simulated digital “ecosystem” and most recently, Jordan Pollock, who grew software creatures which could move in the real world. These artificial evolution systems generate creative new things, but don’t keep evolving. You get them started, they evolve rapidly and surprise you for a while, and then sort of die off.
What do you mean when you say in your book that there’s something vital missing from AI and robotics
I was trying to figure out why all artificial systems didn’t seem as robust and adaptable as real animals. I was also struck by the question of whether our human-like robots could ever have “real” emotions. Or whether emotion would only ever be simulated. What would make something real? It all boils down to the age-old question: what is the difference between living and non-living matter? Maybe there is some-thing in all living systems, even a bac-terium, which we have tried to emulate but are missing.
What is this “new stuff”?
I don’t know. But as an analogy, when people were trying to understand the Solar System with algebra and geometry, they could only describe the planets’ orbits. To understand what was actually happening took calculus. So maybe there is some sort of calculus, some sort of mathematical way of talking about organisation that we don’t yet have.
What about other areas of AI and robotics?
I see that ceiling everywhere. Maybe I’m just being old and grumpy and pessimistic. Or maybe we are really missing something. So I’m sort of placing my intellectual bet that we are missing something. But I’m trying to be a little canny about it. I’m arranging things so that even if we don’t find the pot of gold, we will have had an interesting journey on the way and discovered some things that will surely be useful.
How would you recognise it?
I think we can only hit upon it, if my hypothesis is right, by looking at the edges of where we don’t quite understand things in biological systems, and then getting robots or AI systems to emulate what biology does and see where our understanding breaks down. If I’m lucky and happen to be right, we could then generate some new mathematical tool which would let us bypass those places where our simulations break down. And we will certainly be able to build much more interesting artificial evolution systems. Or much more interesting and robust robots.
Where are you looking for the “stuff”? Are we talking cellular or sub-cellular mechanisms?
I’m looking all the way down to the subcellular level. One of the things we’re trying to look at is prebiotic self-organisation, chemical self-organisation. How did life arise in the first place? It’s a key question to answer.
Some critics might accuse you of getting religious when you talk about this mystical “stuff” out there …
I think a lot of people do get religious. I think Roger Penrose, for instance, has succumbed to the religion of quantum mechanics in his criticism of AI. And the philosopher David Chalmers has talked about consciousness being a fundamentally different quality in the Universe. I think that is getting religious. But I’m actually placing my bets on something that we haven’t yet understood, that is there before our noses. I’m not hypothesising any new physics or chemistry, just that we haven’t developed the right mathematical tools yet.
What is your ultimate aim?
I would like to have a machine or robot which you felt bad about switching off. I want to build a living machine.
How are you going to do that?
We’re trying to build robots with the capabilities of living systems. So we’re looking at questions of self-sufficiency and energy, metabolism, self-repair and self-reproduction, although we don’t have a clue how to do that at this point. We’d like self-assembly-growth if you like-and robots that don’t necessarily have the rigid dimensions of silicon and steel machines.
What’s your line of attack?
One is what I call “computational experiments”. They’re like computer simulation except we’re not simulating a real system. We’re just at early, early experimental stages. Some experiments are looking at the way nervous systems develop in very simple creatures, and if we can understand that, there are interesting self-organisation properties that might give us a clue about the new stuff.
Are you looking at this from the point of view of evolution or embryonic development?
We’re not trying to simulate real chemis-try. We’re trying to simulate a combinat-orial chemistry, to look for the critical properties of a mathematical chemistry that can give rise to the wide variety of self-organisation phenomena that happen in real chemistry. We’re looking at things like how multicellular reproduction arises from single-cell reproduction. Very different sorts of reproduction-but you need the single-cell reproduction before you can have the multicellular reproduction. Then, inspired by all these experiments we hope we’ll come up with deep mathematical theorems-and there we’ve made absolutely no progress.
What materials will you use?
We’re looking at various reformable polymers to see how we could build a robot from them which could move and be powered by, say, a solar battery. A polymer that is somewhere between a plant and a bacterium. Then we’re trying to figure out how to put the computational mechanism throughout it.
Will these robots still be driven by conventional computing?
One of the hypotheses for this new stuff is that it is beyond “computation” in some sense. I think we’ve become terrible computational servants over the past few years. Everything is just computation and I look at some of my neuroscientist colleagues with dismay because they’re using information theory and computation as the main metaphors for understanding how neural systems work. I don’t think neurons developed originally as computers. They developed as synchronisation mechanisms for pulsating swimming motions.
So the brain is a machine, but it’s not necessarily a computing machine?
Yes.
You have said we’re on the verge of a robotic revolution. Is it really possible without your new stuff?
Yes. Although these robots may not be super-intelligent, and they may break down more than we would like, but so do cars.
But people have been talking about this sort of robot for years?
We are starting to see commercial robots in our everyday lives. I think we’re roughly where computers were in 1978. Then the “priesthood” were the only ones able to deal with them. The first computers came into the home in 1978, as games such as Pong. And we’re starting to see home robots in toys, like Lego, Mindstorms, and some “intelligent” toys like Furbie and My Real Baby. We’re also starting to see the first domestic robots for actual jobs in the shape of lawnmower robots. Electrolux is selling a home-cleaning robot in Sweden and quite a few other companies have at least talked about selling them.
Can we have these machines without creating a new slave trade?
We’d like to have ethical slaves, slaves that we can have in good conscience. I like it that my refrigerator works seven days a week, 24 hours a day and I don’t have to feel bad about it being switched on all the time. So we’re going to want those types of robots to be of that class.
But won’t it take human-like intelligence and adaptability to do our bidding, and wouldn’t such robots be bound to be emotional?
If the robots have things explained to them and then adapt to a new circumstance and do the appropriate human-like thing, if they can do all that, then I suspect they will have to have something like emotions. Otherwise, they’ll be so totally alien to us they wouldn’t do what we expect.
So why bother to try to make intelligent robots if they’ll be no use to us?
There’s definitely a purpose in terms of trying to uncover how we operate as people. We need to do science at many levels to understand what it is that makes a human a human: from the biochemical level to the synapse, to the cognitive level and complete integration. Which is what robotics is trying to do. It takes its inspiration from all these levels, trying to put them together in the hope that it will uncover inconsistencies and places where we need more work.
If we make intelligent robots will they even want to do our dirty work?
There are two issues there. First, it is going to take us many, many years to figure out how to build them to be like that. But we may simply want to build our robots so they’re not like that. We choose to build aeroplanes in a very different way from birds. Our aeroplanes don’t have feathers and they don’t get tired when they fly for a long time. They might get overheated but they don’t get tired or distressed, and we’re perfectly happy to have aeroplanes like that even if they aren’t quite as functional. So we’re circumscribed in what we want from our aeroplanes, and we’ll be circumscribed in what we want from our ethical slave robots.
AI and robotics have a long history of military funding. Are you worried about what happens to your research?
Any technology that anyone works on can be used in ways that one wishes it wasn’t. There’s a difference between trying to understand and gain knowledge of something and turning it into a weapon. Understanding and knowledge is something that we should have, because it would be incredibly arrogant simply to say ahead of time that only bad will come from that piece of knowledge. But I am still worried about more and more remote-controlled machines in warfare. In Afghanistan, we’ve had the Predator robot planes that are able to shoot. But there’s a person in the loop. That person may be back in Washington, but there is someone making target decisions. But we could equip those robots and some land-based robots with guns and targeting systems and let them decide. Do we want something like that out in the world or do we want to ban such uses just as we’ve banned biological and chemical weapons? I don’t think it’s an easy question and I don’t think it’s for me to decide.
Do you have lots of robots in your house?
I expect to by the end of the year.
You’ve got four teenage kids. What do they think about robots?
They think it is just weird stuff dad does.
