Sizing up the coming robotics revolution
Rodney Brooks, director of MIT's CSAIL and CTO of iRobot, discusses AI, robots and the coming bicentennial man. Photos: Humanoid robots come to life at MIT
Rodney Brooks is the Panasonic professor of robotics at MIT and the director of CSAIL. He is also the co-founder and chief technology officer of iRobot and one of the principal architects of iRobot's Roomba vacuum.
On Tuesday, RoboBusiness 2007, an international conference showcasing consumer, commercial and military robots, will convene in Boston. To gain insight on what's in the pipeline, CNET News.com sat down with Brooks, one of the leading experts on robots and artificial intelligence.
From his office at CSAIL, Brooks shared his thoughts on the best AI readily available today and the four things it will take for the magicians of science to match science fiction fantasies.
Q: As a kid who watched Star Wars and reruns of The Jetsons, I was convinced that when I grew up, I'd have a robot. Now I have a home of my own, and the closest thing to Rosie or R2-D2 is your Roomba. What do you say about this dichotomy between the high expectations that have been raised by fiction and the reality of consumer robotics?
Brooks: Well, at least we got part of the way. If it wasn't for the Roomba, we wouldn't be there at all and we'd be really disappointed. You may notice you don't have a flying car either.
I think everyone misjudged how some things work, and I normally talk about it in terms of the founders of artificial intelligence, who just had their 50th anniversary last year for 1956. In 1966, they set a summer project to solve the vision problem and they put an undergraduate in charge of it...a young undergraduate put in charge of object recognition!
And that's what Rosie needs. Not only object recognition, but categorization, form, function, all sorts of things which we're way, way away from being able to do in artificial intelligence.
MIT's Domo and iRobot's Roomba are vastly different, yet both are considered robots. What makes a robot a robot?
Brooks: To me what makes a robot a robot, and as with every definition you can poke it enough until it breaks, but for me it's something that senses the world in some way, does some sort of computation, deciding what to do, and then acts on the world outside itself as a result.
What other technology needs to be perfected before a Domo can become a relatively affordable, artificially intelligent majordomo for the house?
Brooks: That is not where the name came from by the way.
I know. Domo arigato.
Brooks: Yeah, (laughs) which is really hard to explain to Japanese visitors. Anyway, I can say there are four research topics that, as we make progress on each one, will enable our robots to do a lot more, and so I have set these goals: the object recognition capabilities of a 2-year-old child, the language understanding of a 4-year-old.
What do you mean by that?
Brooks: They can do conditionals, recognize different accents, words in noisy environments. As an adult talking to a 4-year-old, we may dumb down the vocabulary, but we don't dumb down the syntax. We have clauses and stuff like that, so I think it's pretty reasonable that we could talk to our robots like that.
The third target is the manual dexterity of a 6-year-old. Roughly speaking, a 6-year-old is capable of every manual task a worker in a Chinese factory building goods for Wal-Mart can do and, maybe not at the same strength level, most operations an agriculture worker does.
And fourth?
Brooks: The social understanding of an 8-year-old child--knowing the difference between what you say and your actual intent, all those things that make us human.
How close are you to each of these four objectives? How many years away do you think?
Brooks: Ah. You must be a reporter. I'll never answer that, because, you know, in 1966 they thought it was going to be three months for the object recognition.
Are robots going to be a big-ticket item like the family car where you invest in one that suits your needs, it comes with a warranty, and when it breaks you take it to a mechanic?
Brooks: First off, our cars are going to get more robotic, and we are starting to see that already: The high-end Lexus self-parking, automatic lane changing, staying at a fixed distance from another car. That's going to continue, because these are safety issues, and the Japanese car manufacturers in particular and the Germans want safety.
Brooks: I think that willingness to give up control is going to be slow. The car companies aren't saying, 'let's build an autonomous car right now.' They're saying, 'let's build aids.' I think gradually over time people would become more accustomed to this and we'll see gradual shifts.
You are the director here of the Computer Science and Artificial Intelligence Laboratory at MIT. You are also the CTO of iRobot. How do you balance your responsibility to deliver marketable products and make (iRobot) a profit with your responsibilities here (at MIT)?
Brooks: We started the company in 1990. Pretty soon after that, I stopped all work on mobile robots at MIT and started working on humanoid robots, so that my students here working on their theses wouldn't think that I was sort of subconsciously directing them toward stuff that was going to help iRobot. So, I tried to separate those two in that way.
Do you favor wheeled robots or bipedal robots?
Brooks: We have built our environment for dynamically balancing taller, skinnier things with arms that can reach up and down. So in that sense, and Honda and Toyota have argued this, it may make sense to have bipedal robots because they are sort of compatible with what we are in our environment. On the other hand...I don't want one of (those walking robots) in my house at this point. So, we are a long way from...
From Asimo doing the dishes?
Brooks: Yes. You know a lot of what you see in various corporate videos of their robots are really nice videos, but then they imply that the robot is a lot more capable of deciding for itself what to do than it may really be.
Where is the most interesting work being done?
Brooks: At iRobot, robustness on the military side. You want something that can fall 10 meters down a cliff and then still get up and keep walking. On the home robot side, how do you get performance at low cost so that people are satisfied? Here, I'm much more interested in the four problems and making progress on them.
You know, the rational expectation might be that price is roughly corresponding to the level of autonomy. The reality is exactly the opposite. The more expensive a robot, the more people want to be in the loop to make sure nothing goes wrong.
What do you think are the greatest achievements in AI right now?
Brooks: I think our whole lives are surrounded by artificial intelligence, but we don't think of it that way. Google--you know, all the techniques that Google uses.
You mean the search algorithms?
Brooks: The search, all those sorts of things are cool artificial intelligence and Google, you know, uses statistical machine learning running it. It sucks up AI researchers like crazy, and it's full of AI researchers at all levels. It's a big applied artificial intelligence set of platforms and, apart from my parents-in-law, I don't know anyone who doesn't use Google on a daily basis.
From your years of study of AI, what have you learned about humanity?
Brooks: The work in my group with socially interacting robots has certainly given me insight into humans and a new level of disrespect. Just look how easy it was to get social interaction happening with humans--Kismet interacting with naive people and people responding to it, talking to it, nodding. And by the way, we have a lot of social interaction with our dogs. So it's not just that we talk with our human method with our robots; we do it with all sorts of animals...and it makes evolutionary sense for us to have ways of understanding other beings. And so I guess it was a surprise to me that a few simple cues will trigger that.
You've said that "the coming robotics revolution will change the fundamental nature of society." Generally, in what ways will it change?
Brooks: I think communication technology has changed expectations of how connected we are. I've got four kids ranging in age from 19 to 23, and even in that span of time I can see a change in the forms of communication. The 19-year-old doesn't do e-mails but uses (a social-networking site) and SMS. The 23-year-old uses much more e-mail, a cell phone, but never SMS.
Any truth behind the rumor that you're a robot?
Brooks: I am a robot. So are you.
And that's my next question. There's another evolution taking place, the one where humans are slowly taking on machine parts, as robots take on humanistic qualities. How do you see this changing us as a society?
Brooks: You know, in my view we are machines, bio-molecules that interact according to physical laws. If you take a freshman course in biology here at MIT, and a lot of it's about molecular biology now, it's not like (MIT biology professor) Eric Lander says, "You know this molecule comes down toward this piece of the DNA and then the soul intervenes and aligns the molecules." It's about the mechanics and our science implicitly, not always explicitly but implicitly, believes that that's how things work. We are, therefore, robots made up of little molecular robots.
But then there is the other thing....Will it be the ="http: en.wikipedia.org="" wiki="" bicentennial_man"="">Bicentennial Man? Which I actually think was one of the better science fiction movies. It's schmaltzy, but underneath what's going on, there are some interesting questions. Robin Williams is a robot who wants to become the human, a common theme. He starts replacing his parts with biologically motivated parts, and at the same time those biologically motivated parts become good replacements for people who are decrepit. And so the people and he are sharing the same sorts of parts. And especially as I say, us, my generation, as baby boomers get older, we're gonna want everything--and we've always gotten what we wanted, dammit.
Right.
Brooks: And you guys will pay.
Or maybe we'll just get robots to take care of you.
Brooks: Well, that's right, yeah. I think it's really part of the solution.
So, tell me something about that. I know there's this idea that because we're going to have this large volume of elderly people we're going to need helper robots.
Brooks: So, unlike in Japan--I'm told they're having companions--I don't think that flies well in North America or Europe. But I think the 20- to 60-year-olds...you'll have to be somehow much more productive than in the past. Robots can help. I've visited in Pittsburgh a hospital that has robots just go from place to place driving carts...There's a productivity increase, and it's not about turning to a robot nurse. It's about doing the stuff that the nurses shouldn't have to do.
What's the next big market for robots? Military? Entertainment? Health care?
Brooks: Clearly military is one--the sorts of jobs that can't be outsourced. You either have to import labor, which everyone is now against all of a sudden after relying on it for so many years now.
Mining. Mining's a shitty job. Especially in China it's so horrendous, but even in North America.
Meat packing plants. You hear all these stories about repetitive injury syndromes from people cutting chicken. You could have a robot that, let's say, cuts chicken legs at 1,000 times an hour. Somehow it's got to be done, and people don't want these jobs. They're shitty jobs. So, it's not going to take any jobs away from people who want them.
Brain surgery. These surgeons are now doing surgeries they wouldn't have contemplated before because they have much better tools of knowing where everything is and being able to know what's happening.
It's like, you know, computers didn't replace office workers or accountants. They have changed the nature of the work they did, increased their productivity, which led to the...
The four-day workweek? Never happened.
Brooks: (laughs) No, but it did lead to productivity increasing.
OK. Is there anything else you want to share with people concerning what you do or what you think about?
Brooks: You know, while my reality meter says that it's much more a symbiosis, working together and the robots doing the easy cases of the easy tasks, etc., my intellectual side still wants to go about building Commander Data, a fully autonomous robot that we can all love and discuss things with.
