Выбрать главу

On the input side, we shall assume that many of your mental resources—such as K-lines, Frame-slots, or If-Do-Then rules— can alter the states of some micronemes. Then the present state of your micronemes could represent much of your current mental context—and the states of those fibers are changed, your far-reaching bundle of micronemes will broadcast that information to many other mental resources—so that this will change some of your attitudes, outlooks, and states of mind. In other words, this system could switch you into other, different ways to think. I think that this concept of micronemes could help to replace many old and vague ideas about ‘association of ideas.’ In §§Brain-Waves, we suggest more details about how such a system could play central roles in how our mental processes are organized.

A Hierarchy of Representations

The sections above have briefly described several kinds of structures that we could use to represent various types of knowledge. However, each of these representation types has its own virtues and deficiencies—so each of them may need some other connections through which they can exploit some of the other types of representations. This suggests that our brains need some larger-scale organization for interconnecting our multiple ways to represent knowledge. Perhaps the simplest such arrangement would be a hierarchical one like this:

This diagram suggests how a brain might organize its ways to represent knowledge. However, we should not expect to find that actual brains are arranged in such an orderly way. Instead, we should not be surprised if anatomists find that different regions of the brain evolved somewhat different such organizations to support mental functions in different realms—such as for maintaining our bodily functions, manipulating physical objects, developing social relationships, and for reflective and linguistic processes.

This hierarchy of representation appear to roughly correspond to the various levels of thinking that were proposed in our previous chapters—in the sense that increasingly higher levels will tend to more depend on using story- and script-like representations. However, each of those levels itself may use several types of representations. In any case, even if this diagram turns out to be a good description of the relations between those representations, it is unlikely to closely match the gross anatomy of the brain—because the structures shown in this diagram need be spatially close to each other. Indeed, a substantial volume of a human brain consists of bundles of nerves that interconnect regions that are quite far apart.[177]

How do we learn new Representations?

From where do we obtain our ways to represent knowledge, and why do we find it so easy to arrange our knowledge into panalogies? Are these abilities installed genetically into our infant of memory systems, or do we learn them individually from our experiences? These questions suggest a more basic one: how do we manage to learn at all? As Immanuel Kant pointed out long ago, learning to learn is one of the things that we cannot learn from experience!

Immanuel Kant:“That all our knowledge begins with experience there can be no doubt. For how is it possible that the faculty of cognition should be awakened into exercise otherwise than by means of objects which affect our senses, and partly of themselves produce representations, partly rouse our powers of understanding into activity, to compare, to connect, or to separate these—and so to convert the raw material of our sensations into a knowledge of objects?”

“But, though all our knowledge begins with experience, it by no means follows that all arises out of experience. For, on the contrary, it is quite possible that our empirical knowledge is a combination of that which we receive through impressions, and [additional knowledge] altogether independent of experience … which the faculty of cognition supplies from itself, sensory impressions giving merely the occasion.[30]

So, although sensations give us occasions to learn, this cannot be what makes us able to learn, because we first must have the ‘additional knowledge’ that our brains would need, as Kant has said, to “produce representations” and then “to connect” them.[178] Such additional knowledge would also include inborn ways to recognize correlations and other relationships among sensations. I suspect that, in the case of physical objects, our brains are already innately endowed with machinery to help us “to compare, to connect, or to separate” objects so that we can represent them as existing in space.

All this leads me to suspect that we must be born with primitive forms of structures like K-lines, Frames, and Semantic Networks—so that that no infant needs to wholly invent the kinds of representations that we depicted above. However, I doubt that we’re born with those stuctures complete, so it still requires some effort and time for us to refine those primitive representations into their more adult forms. I hope that soon there will be some research on how that development process works.

Could any person ever invent an totally new kind of representation? Such an event must be quite rare because no type of representation would be useful without some effective skills for working with it— and a new set of such skills would take time to grow. Also, no fragment of knowledge could be of much use unless it is represented in a familiar way. For reasons like these, it makes sense to conjecture that most of our adult representations come either from refining our primitive ones, or by acquiring them from our culture. However, once a person has learned to use several different representations, then that person might be more able to invent new ones. Indeed, we see such skills in the work of those exceptional writers, artists, inventors, and scientist who repeatedly discover new and useful ways to represent things.

How should a brain proceed to select which representation to use? As we have emphasized several times, each particular kind of description has virtues and deficiencies. Therefore it makes more sense to ask, “Which methods might work well for the problem I’m facing—and which representations are likely to work well with those methods?”

Most computer programs still, today, can do only one particular kind of task, whereas our brains accumulate multiple ways to deal with each type of problem we face. However, once we better understand how to organize such resources, along with knowledge to help us decide which technique to use in each situation. To do this we need to develop a wide range of ways to represent those all those capabilities—so that, whenever the method we’re using fails, we can switch to another alternative.[179]

∞∞∞∞∞∞∞∞∞∞∞∞∞∞∞∞∞∞∞

Part IX. The Self

вернуться

177

Also, several such functions could be superimposed in the very same spatial regions, by using by genetically distinct lines of cells that interact mainly among themselves.

вернуться

178

Later Kant claims that our minds must start with some rules like “Every change must have a cause.” Today, one might interpret this as suggesting that we’re born with frames that are equipped with slots that we tend to link to the causes of changes. In the simplest case, of course, that need could be satisfied by a link to whatever preceded the change that occurred; in later years we could learn to refine those links.

вернуться

179

There is more discussion of this in web.media.mit.edu/~minsky/papers/SymbolicVs.Connectionist.html.