Tuesday, December 16, 2008

Movement: Not as Simple as it Seems

I remember reading a short story many years ago by Issac Asimov (I think it was in his book I, Robot). In the story there is a robot named Robbie that plays with a child. While the robot can play ball and many other things the one thing that it cannot do is talk. Apparently Asimov thought that talking was such a complex thing that when we did get around to building robots, like Robbie, that we would have an immensely difficult time getting the robots to talk, whereas all the other normal human functions such as walking would be easy and natural to reproduce.

Thus for the most part, actions such as walking, picking things up and the like are almost always presented as simple for a robot and usually the first thing they "learn" how to do, whereas talking and language in general is presented as being the hard thing to deal with. This attitude towards the complexity of language and the simplicity of walking and moving was not particular to Asimov's stories. There are many more examples of this way of thinking, such as Data from Star Trek, Gort from The Day The Earth Stood Still, the Terminator and others. With these robots they have an easy time moving about but they have an immensely difficult time learning how to talk (or at least talking normally) and communicate.

There are a few notable exceptions to this, but this usually happens in movies where they actually try to make a robot as opposed to having a person wearing makeup (for example, Jonny 5 from Short Circuit). In these cases the actual logistics of making a robot become apparent and the "simple" things such as walking and picking things up are no longer so simple, while the other thing, language, which authors and philosophers considered so difficult can easily be reproduced.

Lately there have been real attempts to make functioning robots, like they have in movies, and while scientists have been able to make computers talk and respond to speech for years, they are having an immensely difficult time with something so simple as walking. Even with all the combined brain power of a corporation like Honda the best they can do is to get a robot to walk up an incline (with the incline already programmed into the robot). Other groups at universities that are developing robots are running into similar problems, they can make them talk, but when it comes to moving and picking things up then it becomes much more difficult.

So it is ironic that authors and philosophers would consider language an extremely complex thing and difficult, if not impossible, for robots to master ends up being one of the easier things. This mistake is understandable because as we learn and grow we first learn to move and to walk and then we learn how to talk. While we learn how to walk fairly quickly, it takes us years to learn how to use language. Granted computers cannot use language to our level, but they can reproduce language in a fairly believable format, while they have difficulty with the simple things like walking.

I guess the reason why I am pointing this out is that I am amused by the fact that many people egotistically think that their words are so hard and complex to understand that surely a robot would have immense difficulty reproducing them, and for robots that is what they must struggle with the most, because that is what we struggle with the most. But when it comes down to it, our language is comparatively simple and it is everything else that is complex. Perhaps the complexity and problems of language are self-created and our inability to learn it does not come from our lack of intelligence, but our lack of intelligibility.


James said...

Having studied language most of my life, I must disagree with your characterization. Although robots and computers can generate speech, they cannot talk. IBM demonstrated speech recognition software at the Worlds Fair in 1963 and unfortunately there hasn't been much real progress since IBMs early attempts. It turns out that language is so complex that no one has yet determined how to make an adequate model of even one language, much less languages in general. Just try asking the computer voice on the phone how the weather is in New York and see how far that gets you.

I do agree that reproduction of physical movements, like bipedal walking and bird flying are difficult but not impossible. It is not necessary that a robot walk like a human as long as it can access the same types of terrain that a human can.

One thing that is apparent, is that robots are not and will not be anything like the movie versions anytime in the near or even predictable future.

We could go on and on about this subject, but that is enough for now.

Euripides said...

I agree with James. Mimicking speech is an easy process. Processing speech with understanding and responding with even the most basic grammar, syntax, and comprehension is extraordinarily difficult.

I do see your point with trying to get robots to walk, however. Making a robot that walks like a human has proved to be a real problem. Robots with four legs prove to be a bit easier to handle unusual walking situations. But the difficulties with using legs at all is the reason mobile robots usually have wheels.