Visit Your Local PBS Station PBS Home PBS Home Programs A-Z TV Schedules Watch Video Donate Shop PBS Search PBS
SAF Archives  search ask the scientists in the classroom cool science
scientists from previous shows
cool careers in science
ask the scientists
PREVIOUS SCIENTISTS

Photo of Rodney Brooks Rodney Brooks as seen on Natural Born Robots: Body Builders

Click on Rodney's photo to read a brief bio.



q Do you believe in the years ahead, like when I'd still be in school, robots will be able to achieve normal, everyday tasks, such as teaching a class, or cleaning houses? (Question sent by Caitlin, 6th grade student, Gainesville, FL)

A I don't think you will see robots teaching a class anytime soon. A number of companies are working on robots for cleaning houses. It is technically possible right now; handling the fact that furniture may be moved from day to day so the robot needs to be clever about using landmarks like "next to the couch" in order to navigate, since the landmarks may change! The robots can also make guesses about what needs to be thrown out (e.g., scraps of paper or a blown in leaf on the floor), versus what needs to be moved but not discarded (e.g., a pair of underwear), but these guesses are not yet perfect. The big challenge is that these robots are still too large to be practical, and too expensive to be attractive to consumers. Both these situations will slowly improve until some company's product "catches on", and then, as with VCRs, there will be rapid improvements in performance and price. Your question is when...I think you'll still be cleaning your own apartment in college, but your children won't when they get to college.


q What does it take to be a robot builder? What skills and studies do you need to be successful? (Question sent by many viewers)

A Robot building is a very broad activity crossing many fields, so it takes a mixture of skills and background to do well. First, you need to study math, as that comes in all the time in designing the mechanics, electronics, and software. Then you need some basics in each of those three fields: mechanics, electronics, and software. Even if you are the electrical engineer, say, on a robot building team, you will need to be able to work with the mechanical engineers and the software engineers and speak their language, and they yours, so that you can all get the problems solved and the interfaces made.

Now if you want to build robots that interact with humans you will also need to study cognitive psychology, and a little neuroscience. And if you want to make small walking robots you will need to study insects, and perhaps their neuroscience. Whichever sorts of robots you choose to work on, you'll need to be a generalist and study the appropriate models from nature to get insight how to build better robots.


q My question stems from articles and reports about the "other end" of robot/AI engineering -- the so-called "top-down" approach that is being attempted in Texas where the engineers are trying to codify "rules of common sense" in the hopes that the machine will reach a critical mass and begin to learn on its own. I was wondering if you give much credence to this approach or do you believe the biological model that you are trying with Cog to be more beneficial? Do you see any way to combine the two philosophical approaches? Would a Cog body with a "top-down" mind (forgive me but I can't remember the project name) be possible and, if so, could it lead in places that neither approach alone could get? (Question sent by Jason, Waco, Texas)

A I believe you are referring to Doug Lenat's CYC project. My personal view is that CYC will not ultimately work out in the way Doug wants it to. It is an impressive project and has built a very useful data structure, which is a super-hyper-linked encyclopedia. I say "super-hyper-linked" because the links are not just pointers but are in terms of deeper meanings. However I think that it will at most be exactly an encyclopedia for systems like Cog, in the same way that people use encyclopedias---it will be a reference work which can give valuable knowledge in an off-line sort of way, but just as memorizing an encyclopedia won't make you into a skilled car mechanic, nor an artist, nor enable you fly an airplane, having the CYC data base available to a robot won't actually enable it to have any new skills-it will still need to learn them. Why do I think this? Well I think a good analogy is to think of CYC as a Korean dictionary---I chose Korean because it is a language that I know none of, nor do I recognize any of its letters. Urdu, or Arabic, or any other language where you know nothing of the alphabet would do just as well. Now suppose you build a robot that knows absolutely nothing about the world, but has this Korean dictionary. Does the dictionary help it in any way? No, as there is no grounding of any of the symbols that relate the dictionary entries to the sensations the robot feels, or the commands it can give its motors. So although all the words are defined in terms of each other that is not enough. You need grounding and you need to understand the language. But even if you add grounding, it will still be just a book, and not a set of skills.

That is my intellectual bet. Maybe I am wrong.


q It seems that legs for Cog can't be very far off. When/if this happens, do you see a steepening of the learning curve for Cog? Do you think independent ambulation will cause the machine to speed its learning or will a pair of legs be just like adding more arms; interesting but not significant? (Questions sent by several viewers)

A We don't have any plans for legs for Cog itself, but we are working on two robots which will not be humanoids, but will have some of Cog's capabilities and will have legs. One of them is a small gorilla-like robot, and the other is a horse-like robot. However, I don't think this will make learning significantly faster. In principle it might, but the fact of the matter is that our algorithms for learning are so primitive that our robots are not limited by lack of stimulation, but are hobbled by their current algorithms---only when we have better algorithms will our robots be limited by their lack of the ability to walk around.




 

Scientific American Frontiers
Fall 1990 to Spring 2000
Sponsored by GTE Corporation,
now a part of Verizon Communications Inc.