Brains and Thoughts

Douglas R. Hofstadter's Gödel, Escher, Bach

It has been a while since my last "Strange Loops" post, Souls: What Are They, Anyway?. In it I talked some more about the human mind — namely, ideas concerning the mind that appear in Douglas R. Hofstadter's one-of-a-kind 1979 book Gödel, Escher, Bach: an Eternal Golden Braid. To Hofstadter, the mind somehow arises out of the mechanical, deterministic operations of the brain. Yet it has non-deterministic attributes that cannot be fully explained at the level of neurons and their vast interconnections.

That said, Hofstadter's Chapter XI, "Brains and Thoughts," seeks a way to establish a high-level brain organization made of "symbols" that somehow emerges from "signals" exchanged by neurons, or from whole groups of neurons called "modules" by Hofstadter. An object of thought is presumably a matter of which modular symbols get activated, among the countless symbols a brain can house.

Perhaps there are large numbers of related symbols activated to represent a concept such as "car" or "road." Who can say?

I have to say at the outset of my discussion of this particular chapter that I am less than convinced that Hofstadter's speculations are on the money. I have the uneasy feeling that the author is trying to force human brain organization into patterns that are — to use one of his favorite words — "isomorphic" with the way computer programs model intelligence artificially. For example, computer programs have access to data — Hofstadter calls it "declarative knowledge," likening it to "facts" (see p. 363) — and are themselves made up of procedures that manipulate data in some way.

These data-manipulating procedures constitute the equivalent, he says, of human "procedural knowledge." An example that springs to mind is the person who says to a stranger asking directions, "I can't tell you how to get there, but I can drive you there myself."

Hofstadter's assumption, then, is that declarative knowledge may not be encoded at the brain's higher organizational levels in the same way procedural knowledge is. Somehow the symbol activation that goes on for human declarative knowledge to manifest itself is different from the type of activation that underlies procedural knowledge ... maybe.

Yet he quickly adds that "in between the declarative and procedural extremes, there are all possible shades." For example, how is a recalled melody stored in the brain? As a succession of remembered notes? As a group of relationships based on tone-to-tone musical intervals? As both of those and more: for instance, with emotional qualities that the notes evoke stored along with the notes?

Or, is melodic memory entirely procedural rather than declarative, as the previous speculations would seem to imply? Perhaps, though, it is a mix of the declarative and the procedural.

In short, Hofstadter offers just one basic high-level construct of brain organization: the potentially activated symbol. True, lurking in the background is a nebulous idea that symbols can trigger one another in mutual activation, and/or be activated in different ways for different purposes and with different results. Yet there is no real suggestion as to how, say, melodic memory can possibly combine two ways or patterns of symbol activation in one.


Another ambiguity arises in Hofstadter's discussion of "classes" versus "instances." Intuitively it seems as if there ought to be a firm distinction between broad conceptual categories — classes — and specific individual instances of those classes. "Golfer" would seem to represent a class of mental objects; "Tiger Woods," an instance. Yet both would, to Hofstadter, necessarily be represented by symbols.

Still, a "golfer" is an "athlete," possibly a "professional." And not all "golfers" are "championship golfers," the way Tiger Woods is. There can be many levels of specificity between a broad class and a specific manifestation or instance. Furthermore, it is likely that some, if not all, classes start off as instances: think of the child who encounters his or her very first golfer.

Once a class of sufficient generality has been constructed, then, each time it is activated by, say, a golfer seen on TV, something like a "rubbing off" must take place. In it, as when making a rubbing from a brass in a church, a copy of the original comes to be. The copy of the original symbol must be able to include sufficient idiosyncrasies to keep it distinct from all other copies. We can't afford to confuse Tiger Woods with Ernie Els, after all.


Another difficulty is that the roster of symbols in human brains can be extended, as we learn and gain experience, while experiments suggest that whatever symbols may be active in the brain of, say, a wasp, are hard-wired and cannot be augmented/altered. It is not clear exactly how symbols as high-level constructs that map to the lower-level patterns of neural signals in the human brain can explain the difference. Nor is it clear how our symbols undergird the capacity of the brain to deal with hypotheticals and what-if worlds.

To Hofstadter's credit, he alludes to all these conundrums. He also admits that he has no real evidence to support his hunch that symbols are very likely localizable to specific groups neurons, no matter how intricately they may be interwoven. Nor does he take much cognizance of the idea that, as I quoted a brain expert in Authority and Metaphor as theorizing, "The brain is a metaphor-making machine. It routinely expresses the concrete in terms of the figurative."


This theory of the human brain as a metaphor machine, which I have to admit gives my brain a buzz of recognition that is lacking when I trudge through Hofstadter's discussion, depends on the fact that the vast majority of each brain is composed of something called "association cortices" or "association regions." In this theory of the human mind as propounded by Baltimore neurosurgeon Michael Salcman, relational "associations" rather than encapsulated entities or "symbols" take center stage. An analogy would be the idea that music is actually a set of tonal relationships, not a set of notes.

I have to believe the capacity to make associations — metaphors, analogies, and the like — is more crucial to human intelligence than the (putative) existence of symbols that might or might not map to individual neural clusters or modules. It is in the realm of associations that leaps of insight can arise. In is in associations that Hofstadter's "strange loops" — patterns that reassert themselves at various levels of consideration — are most at home.

In fact, it seems pretty clear that Hofstadter's own brain made a metaphorical or analogical leap between computerized artificial intelligence efforts of the 1970s, when his book was being written, and human thought. I think he would agree with me, however, that not all leaps are created equal. I believe more in Dr. Salcman's leap concerning the centrality of associations than I do in Hofstadter's concerning symbols in the brain.