Monday, November 20, 2006

What is a robot?

Is your washing machine a robot? Or could we say that you are a robot? Is a computer a robot? How about a spider? What exactly is a robot and what are the criteria for defining it? According to one of the Webster definition, "a mechanism guided by automatic controls", even a trap or a thermostat is a robot. Another definition, "a device that automatically performs complicated often repetitive tasks" sounds a little more accurate but it's also awfully imprecise. How complicated should something be to say it's a robot and how repetitive should its tasks be?

What counts in these definitions is, on one hand, the idea of "automatic control", rather than being remote-controlled by somebody, and on the other hand, the fact that the robot, in spite of being autonomous, isn't a very creative device capable of improvising and exhibiting novel kinds of behaviors.

But it seems rather awkward to claim that a trap or even a washing machine is a robot. They are both autonomous and have a fixed, uncreative, behavior pattern. Nonetheless, they are not robots. They are not even robots. This in spite of the fact that a washing machine is quite a complicated device; so, at closer scrutiny, the criterion of complexity doesn't seem very relevant.

What makes these devices not robots, but mere automatons, is the fact that their behavior is hardwired. A robot in contrast has a certain program which is distinct from its brain. If you want to change the program you can do it without employing extensive physical alterations to the hardware. In case of a washing machine if you want to change its programs you need to actually change a piece of hardware.

The distinction between software and hardware is not as sharp as one might think. This was more obvious in the first computers where the programs were stored in the form of pierced cards. Changing the cards meant changing the program the computer was running. But in a certain sense the cards themselves were also pieces of hardware. In today's computers the pierced cards are replaced by things like magnetic patterns, but even so, the software still remains a physical entity and changing the program involves a certain change of hardware.

Nonetheless, the distinction between software and hardware is a very useful distinction, although not a hundred percent sharp distinction, and provides a basis for telling apart a robot from a "mere" automaton.

But this isn't enough. After all, a computer is also like that and computers are not considered robots. A computer can however be the "brain" of a robot. It becomes so if we attach to it more active peripherals such as wheels or hands. A usual computer has rather passive kinds of peripherals (like a monitor, a printer or speakers) with which it can only communicate without doing very much. So, besides having a programmable brain, the robot also has to be an active device capable of doing things.

Are animals more like automatons or more like robots?

What about something like a spider? Is it more like a washing machine or more like an industrial robot? Are there animals that are more on the robot side and others that are more on the automaton side? Or are they even more than "mere" robots?

The classic view on animals, promoted by Descartes in the 17th century, is that they are automatons. The modern view is that they are robots. The biological software, the equivalent of the pierced cards, is the DNA while the animal body (the "vehicle") is the hardware. The animal's behavior, the way it reacts to environmental stimuli, is programmed by its genes.

The DNA program consists of "coding genes", of "regulatory genes" and of meaningless junk. The coding genes are the equivalents of program routines, while the regulatory genes are the things that call certain routines or inhibit their expression. The coding genes are like a toolbox and the regulatory ones are the programs that use the toolbox.

For example, when an animal eats something, some of the ingested molecules reach the DNA and block the inhibition of certain coding genes; due to the fact that these genes are no longer inhibited, the production of certain enzymes (which are coded by the genes) starts; these enzymes are catalysts that mediate the decomposition of the ingested molecules; when the molecules have been "taking cared of", the inhibitory mechanism is once again in place and the production of that kind of enzymes is stopped.

The difference between various animals, say, a mouse and a bat, is especially due to relatively small differences in the regulatory genes - each living being is just another program working on the same "operating system" of the coding genes. For example, the bat becomes a bat because of its genes that regulate the growth of its fingers; unlike the case of a mouse, this growth doesn't stop. (Read more about the evolution of bats.) Marc Kirschner and John Gerhart have determined that in the 4 billion years evolution of life on Earth there were only five major upgrades of the genetic operating system (the first one being the very "invention" of the DNA operating system). Richard Dawkins called the occurence of these upgrades the "evolution of evolvability", because with each upgrade the possible variability of life (the number of possible different viable organisms) increased enormously.

In other words, the difference between various animals is, in a large part, due to a difference of software and not of hardware – the same kinds of processes (the ones in the "toolbox") are used in different ways. This is why today's biologists consider animals more like robots and less like automatons. The programmer behind these robots is natural selection which, in the given environment, favored certain programs and disfavored others – some kinds of programs have spread while others have not.

"A better analogy for a gene than either a word or a sentence is a toolbox routines in a computer," wrote Richard Dawkins in The Ancestor's Tale. "The reason all Mac programs have the same 'look and feel' (that very similarity famously became the subject of litigation) is precisely that all Mac programs, whether written by Apple, or by Microsoft, or by anybody else, call the same toolbox routines. ... The genome, sitting in the nucleus of every cell, is the toolbox of DNA routines available for performing standard biological functions. The nucleus of a cell is like the ROM of a Mac. Different cells, for example liver cells, bone cells and muscle cells, string 'calls' of these routines together in different orders and combinations when performing particular cell functions including growing, dividing, or secreting hormones. Mouse bone cells are more similar to human bone cells than they are to mouse liver cells – they perform very similar operations and need to call the same repertoire of toolbox routines to do so. This is the kind of reason why all mammal genoms are approximately the same size as each other – they all need the same toolbox." (page 185)
How about humans?

The main difference between humans and the rest of the animals is that we are exquisitely good at imitation. Moreover, we are not just capable of copying the goals of others, but also of copying their methods of doing things. This is the main difference between us and the rest of the apes – read more.

What this means is that our ability of copying methods provides us with yet another toolbox. We are not just imitating the methods of doing something while we are pursuing the goals we have observed – we are also capable of employing the same methods for achieving new, personal, goals. So, a new kind of toolbox and of programming is available to us, a kind of programming which is (largely) not available to other animals. Thus, our behavior is not just a function of our genes and environment, it is also, in addition to the genes and to the environment, a function of our memes (the cultural replicators which had appeared and which spread because of our ability to imitate). As a side effect, our brains are no longer just computing machines serving to spread of genes, like the brains of virtually all other animals; our brains are also involved in using, acquiring and changing the memes - an endeavor we call creativity.

So, we have returned to the issue of creativity, which robots are not supposed to have. But what exactly they don't have?

There are two kinds of views on creativity. According to the first view something that produces new things is creative. In this sense an industrial robot that produces new cars is creative. But most of us have a much higher standard. We're talking about creativity when the produced thing is so different from what already exists that it stands in a class of its own. What exactly does this mean?

I can give you the other extreme of what "being creative" means. Suppose you have a given toolbox. Then it is possible, in principle, to list all the possible combinations, all the possible uses of the routines in the toolbox. Some of these combinations already exist in practice, most of them don't. But if you're just producing something of this virtual set (a new program put together of these routines) then you're not creative. You're just playing an already established game. According to this somewhat radical view, to be creative means that you improve the toolbox itself and then use the new items to create something. You're not just playing an already established game – you're doing it your own way. The "history of life according to the gene" described by Kirschner and Gerhart only involved five such bursts of creativity which ultimately were random happenings. But how about the cultural creativity? Is that also random?

This is what metaphors are all about – one is using the language in such an unusual way that old words are turned into new tools. The cultural toolbox is changed via metaphors, rather than by random mutations. The metaphors can eventually become so popular that they lose their status of metaphors. For example "on the other hand" was probably a metaphor a long time ago, but now it's not. Metaphors don't seem to be just random happenings, but what they are exactly remains elusive.

Now think about the lego game. According to this radical definition of creativity nobody can be creative here because one cannot create its own bricks. Nonetheless, some constructions can be more creative and "original" while others are common. In other words, there isn't such a drastic distinction between creative and not creative. We understand the concept of creativity in a more tolerant fashion and as a consequence there is a gradual transition from the totally unoriginal industrial robot to the best poet.

From what has been said it should be clear that, given the current style of programming, humans are not robots. The kinds of programming systems employed today are not capable of exhibiting creativity in the strong poetic sense. For one thing, we have no clear understanding of what metaphors are. Moreover, while it's easy to program something to follow certain ascribed goals, it is difficult to program it to have its own goals. However, I think that in principle it could be possible to invent another style of programming that would allow such tasks. (I have speculated about this here.) Then the very definition of robot might change. But until such novel programming languages are created, we're not robots and our metaphors retain their mystery.

No comments:

Post a Comment