Individuation: Result or Cause?
ASADA--What I'm working on now is, for example, using PC's to gauge changes in the others' size, position and direction to predict their movements, and judge how to move accordingly. Developing descriptions of the other, in a sense. Quantifying one's actions in relation to one's observations of the other's actions. The problem is that there isn't much psychological play that's going on. I mentioned eye contact earlier, but we've not at all come that far. When we have two robots in play, in order for the other robot to recognize this robot's actions, we need them to understand each others' perspectives. At present we have no way to secure this recognition. Of course, it is possible to recognize the other's eye or neck movements when they get the ball, or the ball's movements and infer a causal relationship between the line of play and their apparent perceptual facilities, but we cannot claim that this is the line of their reasoning. Just a posteriori inference, and only partial a posteriori knowledge at that, especially if we begin dealing with sentient beings. So the question is, "what kinds of core knowledges need these robots be embedded with?"
SAKURA--With living beings there are so many different possible cases and exceptions. The simplest form of feinting signal are those which are genetically formed, but abused for duplex ends. You have, for example, a species of firefly that eats another species of firefly. The female of the predator species knows to send the mating signal of the prey species. The male comes, ready to mate, and is promptly eaten instead. The original meaning of this call is buried deep in the female's genetic inheritance, but its application is quite clear. There are also birds that live in the Amazon basin, who flock with other species--as a survival measure--which works when there is enough food, but when there isn't enough to go around these birds are capable of imitating their flocking partners' alarm calls. Once their partners have fled, they divide the food source among themselves. They are able to employ imitation to instill confusion to their advantage. You can see phenomenon like this in many other species of animal, but trying to translate this into a workable feint in soccer is a whole other issue. Trying to second guess the other players' strategy is really high level technology.
ASADA--The question is really why and how so many such templates were created within the history of evolution.
SAKURA--Isn't it a matter of so many genetic transformations, and various remnants of adaptive measures, part conclusions of trial-and-error, or natural selection?
ASADA--I recently heard the biologist DAN Marina from Osaka Prefecture University say that there is an unambiguous reason for these, that it is latent survival traits. Maybe it is correct to interpret that at some genetic level or cellular level some reason has made itself clear. Yet how can we be sure that a given template was an evolutionary by-product? I find interesting the question whether it is something that has naturally come out of the individual, is genetic, or whether it is some kind of coding, or . . . ? To put it another way, whether you're speaking of bugs and birds, there are relatively few template changes within one life span. When speaking of primates, however, and the developed cerebral cortex, where the plasticity of what you're dealing with increases, these kinds of changes become possible within one generation. And if the mechanism for skill accumulation exists in one experiential cycle, one lifetime, then perhaps it is something that we can apply to robots.
SAKURA--Yet even in birds there is a great deal of potential for learning. Even within one fowl generation considerable change is possible. And signals may well change according to territorial issues as well. Even fireflies may have some potential for generational development. In this sense I don't think that you can describe the phenomenon as a potential based on genetic templates versus experience, but rather that of each having quantitative differences. Recently it is popular to say that learning-based behavioral changes lead genetic adaptations, that actions function to promote biological evolution. With behavioral changes the habitat itself changes, so after a few generations it is natural that the genetics would naturally follow in turn.
When you mentioned templates versus plasticity it reminded me of "Disparity Theory of Evolution," a fascinating hypothesis about genetic evolution by Daiichi Pharmaceutical Company's researcher FURUSAWA Mitsuru. In it, FURUSAWA deals with the fact that of the two strands of DNA, only one strand is used. The other appeared dormant, like a redundancy, or back-up of sorts. But this never quite set well with him. His new "Disparity Theory of Evolution," however, puts this strand into the spotlight, stating that it is the basis of evolutionary leaps.
You see, these two strands of DNA are not identical. One has a higher propensity for mutation (the continuous, or "leading" strand) and the other is relatively more stable (the discontinuous, or "lagging" strand). Now, DNA reproduction goes from the 3' end to the 5' end, and they face each going in alternate directions. [See diagram (a).] During DNA replication, when both strands are copied in their entirety, they are read in sequence, according to their base distribution. The continuous strand replicates itself, but at the same time, the discontinuous strand reads it backwards, and for this reason it cannot read the entire strand in sequence, so it reads a section, returns to base, reads another section, returns to base, and so on. [Diagrams (b) and (c)]
This phenomenon was discovered by biologist OKAZAKI Reiji[*10] in 1966, while he was at Nagoya University, and is called "Okazaki fragment," after him. Well, once the discontinuous strand has collected a sufficient number of these Okazaki fragments, it strings them together into one DNA strand. It is called "non-continuative replication," but because it takes such a convoluted way of going about things, it is filled with inconsistencies and variations. The result is that the discontinuous strand is filled with far greater potential for mutation. But why increase the potential for copy errors, why have such a convoluted replication process? It is the same question as why there need be two DNA strands. But since the discovery of the Okazaki strand 30 years ago there still hasn't been a satisfactory explanation.
FURUSAWA's explanation is that mutation potential is the basis for having two DNA strands. The continuous strand can conduct the copying accurately, and keep the organism functioning within its existing environment at its present level of adaptation. This much information is taken for safekeeping. On the other, discontinuous strand we have all of the copy misses, the random factor entered, so that it is relatively easy to produce different genetic information. The greater percentage of what is created here is meaningless, but occasionally an important strand will be created, and this becomes part of what will be copied in the next replication.
This is what FURUSAWA calls "genetic principle security," an interpretation which says that once the organism is assured that the status quo is met, it takes chances in an analog medium. And this would mean that DNA has an extraordinarily effective evolutionary mechanism built into itself. And this is why, he speculates, organisms retain DNA and not RNA or other media for their genetic information. Until now it was always speculated that the subjects of mutation were a mass of genetic material which was unable to adapt, and that this was the reason for their evolution. It is a long standing question. Well, with this hypothesis that problem is also addressed. I really consider it an interesting perspective.
I'm not really sure if it ties in directly with what you were talking about a minute ago or not, but the idea of an unchanging template on one hand, with audacious plasticity on the other seemed close. Without both, the potential for the organism to learn and grow would seem too limited.
ASADA--Interesting topic. Take crickets for example. The females mate with the male with the most beautiful song. But crickets whose song doesn't really get the girls will group around a male whose does, and steal their females--getting close enough together that the females can't tell which one has the winning tune. How do these male crickets know that their song doesn't quite do it for the ladies, and why are they so smart as to know whose song does, and set a successful strategy to get their females? This appears to be pretty intelligent conduct.
Because insects assume this behavior, does this mean that we can assume that they have a sense of individuation? Some might think not, but I believe that this conduct is related to the individuation that we are trying to develop in robots.
SAKURA--What you just described, the "sneaker strategy" can be found in toads, fish, and other animals as well. It's hard to say how they know that they're not singing well, but it's not just the absence of females. Also, what are the conditions which need to be satisfied for defining self awareness? I'm not sure that I understand all of what you're saying, but I suppose that it comes back to the issue of the development of the central nervous system.
ASADA--Birds maintain quite a strong social element including things like monogamy. I believe that this indicates a strong sense of individuation. Crickets, on the other hand, will mate with anything in sight. Not a very developed social sense, or indication that individuation has taken place.
SAKURA--I think that it's hard to imagine a sense of self in society among crickets. Animals which lay so many eggs have quite a severe sense of survival forced upon them from conception, and their having a developed sense of self is not what is going to save them. It's a battle of overwhelming the odds. Whoever's left standing when it's all over is ok. With fowl and mammals, who only have a few offspring at one time each individual counts, so individuation may be much more important. Of course, there may be differences from one cricket to another, but I'm not very hopeful that that is the defining difference in their actions.
ASADA--So your thoughts are that without the need to assert one's individuality, individuation does not appear. If you follow the history of evolution, however, I don't know whether the cause and effect relationship of individuation is that clear.
SAKURA--Within the evolution of living things, I think that you can draw two paths. The first is where you find individuation only in statistical averaging of a species. With crickets you have a society of anonymity established, just as with honey bees and ants, which doesn't rely on individuation, but on group behavior to function. The other is groups such as vertebrate, in whose societies each individual is considered important. Humans are the classic example of this. Living things on this earth are somehow divided between these two. In the former, the individuals are only a statistical result. In the latter they are the essence of society. Perhaps with robots you are aiming for individuation, but you're still at a point of interchangeable anonymity.