Vehicularization: A Control Principle in a Complex Animal with several levels of Modal Organization (with note on King Kong)

Another bump to the top, this time on general principle. Think of vehicularization as a control mechanism of natural intelligence.

* * * * *

 

I'm bumping this to the top because it has direct bearing on my current ring-composition work. Consider the temporal horizons of actions undertaken in each segment of King Kong. It opens in New York. Denham dreams of getting rich by making a movie of King Kong on Skull Island. That's indefinitely in the future. Once on board the ship the temporal horizon moves closer; now action is directed toward getting to Skull Island. Once on Skull Island the goal is simply to get into Kong's territory. And once inside Kong's territory the goal becomes simple and very immediate, to survive.

Once Darrow is successfully rescued, Kong is captured, and they're back in New York. Now the goal is simply to get rich, which will happen real soon now, as soon as the money starts rolling in from exhibiting Kong. And, of course, Darrow and Driscoll plan to get married. Important, but it doesn't drive the action. What drives the action is Kong's escape. Now, once again, the goal is immediate survival. Once Kong is dead, the couple can get married and live happily ever after.

Well, vehicularization is about temporal horizons in the control of behavior. It's about the brain, but we can also map it onto the geography of the journey in King Kong, giving us a temporal myth-logic. A similar analysis can be done for Heart of Darkness, though that's going to be tricky because of the double narration, and Gojira.

* * * * *

The following post consists of a section that David Hays and I removed from our article, Principles and Development of Natural Intelligence (downloadable PDF). The final draft was long and the journal editors asked that we cut it. This is much of what we cut.

I’ve interpolated some comments in italics and appended a later note. While the passage is best read in the context of the whole article—which explains how we used the mathematical notion of diagonalization and has a full discussion of behavioral mode (downloadable PDF)—it can be read independently. The general idea is of one system being nested within another such that the deepest, the innermost system, leads to the most immediate satisfactions, but also has the most restricted behavioral scope. A system with greater scope, while not capable of satisfying a basic need (e.g. for food, water, sex, companionship) is able to move the animal to a place in the environment where the innermost system can exact satisfaction. The outer system is thus serving as a vehicle for the inner system.

Vehicularization

This story can be brought to closure with the concept of vehicularization. Neurophysiologically, the concept is a generalization of Paul MacLean's (1978) familiar concept of the triune brain. In McLean's conception the mammalian brain consists of a complex of reptilian grade embedded within one of paleomammalian grade which is in turn embedded within one of neomammalian grade. Restating this general idea in our terms, with the emergence of the diagonaliztion principle the modal system becomes embedded within the emergent sensorimotor system. As each new principle is implemented in new tissue the previous system becomes embedded within it.

Behaviorally, a higher order system serves as a vehicle for moving the organism to a place in the environment where control can safely be transferred to a lower order system. Conversely, when a lower order system is blocked without having satisfied the exit requirement of the current mode, transfer can be given over to a higher level system, which will then transport the organism to a location in the environment where satisfaction of the current exit condition is more likely. The overall effect of behavioral vehicularization can be stated in terms of a hill-climbing search strategy.

For example, you’ve been working hard and, all of a sudden, you notice that you’re ravenously hungry. Your lowest level system, the one that is actually capable of satisfying your hunger, wants to grab some food and start chewing. If a cheese burger or a head of lettuce is close at hand, you can see it and grab it, it does into action and your immediate hunger is sated. If no food is available, however, you turn control over to a higher-level system that then goes looking for food. If you are in your home, you’ll go to the kitchen or the pantry and see what’s available, now.

In hill-climbing a gradient is placed on the environment, the search space, such that locations most likely to satisfy the search, to fulfill the system's current need, are higher than unpromising locations. [Don't confuse the physically real environment and the abstract search space. The hill being climbed is abstract.] The organism then climbs to the top of the nearest hill and, if all is well, is satisfied. However, hill-climbing has a weakness; the local maximum may not be the global maximum for the search space. When this is the case the system is stuck at the local peak with no way of moving down it and then over to the global maximum.

So, you’re hungry. There’s nothing immediately to hand, but you smell something potentially delicious. You follow your nose and it leads you to a window. There’s no food on the window sill and the window’s filled by a screen you can’t break through. What do you do? If you insist on following your nose, you’re stuck. That’s a local maximum. So you’ve got to stop following your nose and do something else to take to a place in your environment where following your nose will be more successful.

In such a situation a modal organism can only exit the current mode. That being done, the gradient which trapped it is lifted and it is now moving along a different gradient, quite possibly one defined by an exploratory or search mode. That is budgeting. Vehicularized organisms can deal with the situation by transferring control to a higher order system which can then move the lower order system away from its local maximum to a position in the environment closer to the global maximum for that lower system. When that location is reached control is then transferred to the lower system, which climbs the hill to satisfaction.

In such a vehicularized organism the modal system is stratified. The reorganizing mode [learning] is one which permits the resetting of the conditional elements of on-blocks [simple control triggers: ON Condition X DO Action Y], thus changing the coupling of the organism to its environment. The higher level modes, play, imitation, and language, permit a great deal of exploration and activity before the conditions and actions of on-blocks are committed. These modes are probably particularly important in building very complex conditional and actional elements.

This is particularly important in ontogeny. It is known that the human nervous system matures roughly in the phylogenetic order of its components (Milner 1976); that is the innermost vehicles mature first. This suggests that the tissue which will be implementing the higher level principles matures under the guidance of the electro-chemical gradients generated by the activity of the lower vehicle, which is controlling the behavior (cf. Edelman 1978). Thus, we know one feature of the development of the nervous system is the early proliferation of synapses in a region followed by the elimination of many of these synapses (Cowan 1979, Purves and Lichtman 1980). That elimination might, for example, be the primary method of diagonalizing in cortical tissue. Once tissue had been diagonalized its critical period would be over. After that it could only learn patterns within the types specified by the diagonalization. Once the new tissue had been diagonalized it would be ready for the implementation of higher level control organized into higher level modes.

At this point it is clear we are once again firmly within the region governed by the biological principles of epigenesis. Any full understanding of the nervous system requires a deep understanding of how these biological principles shape neural tissue. But those principles are outside our purview. Out point is simply that vehicularization is a critical link between the action of the information principles and their implementation in neural tissue. It is our suspicion that a deeper understanding of vehicularization will lead to, or perhaps follow from, an understanding of the particular adjustments of developmental sequences which follow from the ontogenetic recapitulation of phylogeny (Gould 1977).

* * * * *

Note made on 8.29.2000:

I just read a bit on animal navigation that's relevant here. It's from David Gallistel's article in MIT's encyclopedia of cognitive science. Animal navigation is mostly dead reckoning. It's only beacon guided when the animal is close to the target. In Gallistel's words, “Beacon navigation is the following of sensory cues emanating from the goal itself or from its immediate vicinity until the source of the sensory beacon is reached. Widely diverse species of animals locate goals not by reference to the sensory characteristics of the goal of its immediate surroundings but rather by the goals’ position relative to the general framework provided by the mapped terrain.”

What this means is that beacon guidance is a different mode from long-range navigation. It’s a different mode and a different vehicle.

A critique of pure learning and what artificial neural networks can learn from animal brains

"what is sometimes misleadingly called “artificial general intelligence”—is not general at all; it is highly constrained to match human capacities so tightly that only a machine structured similarly to a brain can achieve it." https://t.co/5yIwEFhcrk

— Alberto Acerbi (@acerbialberto) August 22, 2019

The article linked in the tweet: Anthony M. Zador,  A critique of pure learning and what artificial neural networks can learn from animal brains, Nature Communication:

Abstract: Artificial neural networks (ANNs) have undergone a revolution, catalyzed by better supervised learning algorithms. However, in stark contrast to young animals (including humans), training such networks requires enormous numbers of labeled examples, leading to the belief that animals must rely instead mainly on unsupervised learning. Here we argue that most animal behavior is not the result of clever learning algorithms—supervised or unsupervised— but is encoded in the genome. Specifically, animals are born with highly structured brain connectivity, which enables them to learn very rapidly. Because the wiring diagram is far too complex to be specified explicitly in the genome, it must be compressed through a “genomic bottleneck”. The genomic bottleneck suggests a path toward ANNs capable of rapid learning

From the article, on learning:

In ANN research, the term “learning” has a technical usage that is different from its usage in neuroscience and psychology. In ANNs, learning refers to the process of extracting structure—statistical regularities—from input data, and encoding that structure into the parameters of the network. These network parameters contain all the information needed to specify the network. For example, a fully connected network with 𝑁 neurons might have one parameter (e.g., a threshold) associated with each neuron, and an additional 𝑁^2 parameters specifying the strengths of synaptic connections, for a total of 𝑁+𝑁^2 free parameters. Of course, as the number of neurons 𝑁 becomes large, the total parameter count in a fully connected ANN is dominated by the 𝑁^2 synaptic parameters.

There are three classic paradigms for extracting structure from data, and encoding that structure into network parameters (i.e., weights and thresholds). In supervised learning, the data consist of pairs—an input item (e.g., an image) and its label (e.g., the word “giraffe”)—and the goal is to find network parameters that generate the correct label for novel pairs. In unsupervised learning, the data have no labels; the goal is to discover statistical regularities in the data without explicit guidance about what kind of regularities to look for. For example, one could imagine that with enough examples of giraffes and elephants, one might eventually infer the existence of two classes of animals, without the need to have them explicitly labeled. Finally, in reinforcement learning, data are used to drive actions, and the success of those actions is evaluated based on a “reward” signal.

Much of the progress in ANNs has been in developing better tools for supervised learning. If a network has too many free parameters, the network risks “overfitting” data, i.e. it will generate the correct responses on the training set of labeled examples, but will fail to generalize to novel examples. In ANN research, this tension between the flexibility of a network (which scales with the number of neurons and connections) and the amount of data needed to train the network (more neurons and connections generally require more data) is called the “bias-variance tradeoff” (Fig. 1). A network with more flexibility is more powerful, but without sufficient training data the predictions that network makes on novel test examples might be wildly incorrect—far worse than the predictions of a simpler, less powerful network. To paraphrase “Spiderman”: With great power comes great responsibility (to obtain enough labeled training data). The bias-variance tradeoff explains why large networks require large amounts of labeled training data.

Much later:

In this view, supervised learning in ANNs should not be viewed as the analog of learning in animals. Instead, since most of the data that contribute an animal’s fitness are encoded by evolution into the genome, it would perhaps be just as accurate (or inaccurate) to rename it “supervised evolution.” Such a renaming would emphasize that “supervised learning” in ANNs is really recapitulating the extraction of statistical regularities that occurs in animals by both evolution and learning. In animals, there are two nested optimization processes: an outer “evolution” loop acting on a generational timescale, and an inner “learning” loop, which acts on the lifetime of a single individual. Supervised (artificial) evolution may be much faster than natural evolution, which succeeds only because it can benefit from the enormous amount of data represented by the life experiences of quadrillions of individuals over hundreds of millions of years.

And so:

Vehicularization & Ring-Form: Remarks on some issues raised at #HEX01

Edit 11.16.17: I've added some new material to the section on ontological mismatch. I've marked it by highlighting it.

I enjoyed presenting to HEX01: First Workshop on the History of Expressive Systems. I wish I’d had more time (don’t we all?), I wish I’d been there in person to talk with people and play with the exhibits. We do what we can.

I’ve been thinking about these issues for years. And will continue doing so. Indeed, between the time I submitted a draft paper (Abstract Patterns in Stories: From the intellectual legacy of David G. Hays)...

Paper at Academia: https://www.academia.edu/34975666/Abstract_Patterns_in_Stories_From_the_intellectual_legacy_of_David_G._Hays

And the time I put the last touches on the PowerPoint I used for the talk ...

PPT at Academia: https://www.academia.edu/35142428/Abstract_Patterns_in_Stories_From_the_intellectual_legacy_of_David_G._Hays

I had a few ideas that pushed the work forward here and there. I continue the push in these notes, which are rather informal. I’m just trying to get the ideas down on (virtual) paper.

Of course, the workshop was about history, so what was I doing presenting new ideas? Continuing the history. Oh yes, I presented some history, the computational ideas worked out by David Hays and his students in the mid-1970s, and how I, a student of literature, came to them. But streams of intellectual development don’t stop just because they’re always disappearing into the past.

More importantly, things change, deeply. I went into the 1970s with one set of ideas – call it paradigm in Kuhn’s sense, an épistème in Foucault’s – which I used to think about how language and literature work. I encountered a very specific issue (problematic?) within that paradigm, the structure of “Kubla Khan”, and my efforts to deal with that issue forced me to think in terms outside that paradigm, to start cobbling together a new paradigm (if I may). Am I there yet? Who knows?

That’s what I address in the first of these notes, about ontological mismatch in our thinking. Then I take a look at the triune model of the brain, as Hays and I recast it in terms of control hierarchy. I then use that recasting to think about ring-form in King Kong. I conclude with some remarks about Heart of Darkness.

A half-century of ontological mismatch (beyond the singularity)

I mean ontology in the sense it has come to have in computer and cognitive science, the organization of different types of objects in some domain. Prior to my work on “Kubla Khan” [1] I had internalized a certain ontology for dealing with literary phenomena. But the moment I decided to interpret line-end punctuation like parentheses, brackets, and braces in a mathematical expression (or like nested parentheses in a LISP expression) I moved out of that ontology and into a different one. It’s worth noting that, when I made that decision, I specifically thought about the computer programming course I had taken, and how, in THAT world, if you place a comma where a colon is expected, it won’t work.

The problem, then, is how to think about literary texts in a world where LISP expressions are ‘native’ objects.

Of course, we–me, my teachers, others–didn’t realize that that’s what had happened. (Of course, we didn’t think in terms of conceptual ontologies at all.) We just thought I was doing something strange and interesting within the existing (or perhaps emerging) ontology. The same with my 1976 paper on Sonnet 129 [2]. To be sure, it looked very different from every other article in the special issue of MLN. It had all those diagrams, while the other papers had no diagrams at all.

It wasn’t until much later that I realized that, when I did that work on “Kubla Khan”, I had irreversibly left the conceptual world of academic literary criticism. “Irreversible” because I can’t go back, though I can do good imitations.

Contemporary work in computational criticism presents the same problem. The desire to call it “distant reading” reflects a commitment to the standard ontology, an ontology is which the text is only incidentally marks on paper. In the standard ontology the text is, well, that’s hard to say. It’s that thing that you read, it’s somehow tethered to those marks on the page, but it’s more than those marks.

Well of course its more than those marks, but I can’t think of a better way to characterize that “more” than to think of it as come kind of computational process. And that’s what computational critics are scrupulously avoiding. On the one hand thinking of the mind as somehow fundamentally computational is of little practical value in their computational work. But also, they need to deflect the criticism of their more traditional colleagues who are wont to think of the notion of the mind as computational as, you know, the work of the devil.

Yet, in their own work, computational critics are working within an ontology in which the text is just marks on paper. The (miraculous? not really, but very artful (rare device)) craft in computational criticism is to analyze massive collections of such (mere) marks in a way that reveals the traces of mind, thousands and tens of thousands of minds reading. Think of it, from mere marks to the mind. That’s what computational criticism allows.

THAT ontology is different from, incommensurate with, the ontology of ordinary lit crit. There’s a deep tension that that is being glossed over. On the one hand, computational critics call it “deep reading” and note that, no, it’s not in competition with “close reading”. They’re complementary activities, complementary perhaps, but not ontologically compatible. On the other hand, traditional critics see “computer” and give a shudder–“There be dragons! Weave a circle around them thrice, and then lock ‘em up and throw away the key!” It’s not that bad; really, it isn’t. 

But still, THAT conversation has no happy ending. But no one’s dealing with that ontological gap. It can’t be bridged. Rather, it signals a need to rethink the discipline from top to bottom.

Which brings us to The Singularity. I figure that dreams and/or nightmares of the day when computers will become super-intelligent, those fantasies are rooted in a 19th century worldview. As such there’s an ontological mismatch between them and computing technology.

More later.

I’m confused, don’t know which way is up, time to ramble [#HEX01 #DH] [Ramble 8]

Well, not really confused. But there’s a lot going on so I need to just sit back a bit and THINK.

HEX01 & ring composition

I’m psyched about my upcoming talk at HEX01, the First Workshop on the History of Expressive Systems. I’ll be Skyping it in early Tuesday morning (the 14th). The talk’s called “Abstract Patterns in Stories: From the intellectual legacy of David G. Hays” and will give me a chance to get some early work of mine, and of Dave Hays, my teacher, on the record before an audience that’s actually interested in it, which is more than I can say for literary critics, even those who profess to be interested in cognitive science. Oh sure, they’re interested in cognitive science, within limits. But let’s not go there now.

Anyhow, I’ve been working on a presentation version of the material, which, of course, is necessarily very different from the written document. I’m including some material oriented toward folks who might be interested in making interactive ring-form narratives/games. And I’m discovering new things in the process. Not radically/deeply new, mind you. But interestingly new. It’s exciting.

I’m beginning to get a feel for the inner logic, the myth logic, of these ring-form narratives – see Tuesday’s Vehicularization: A Control Principle in a Complex Modal Animal (w/ new note on King Kong). It’s about temporal horizons, which seem to get shorter and shorter as we approach the center, and then they start to open up once we’re through it. (See opening paragraph of the vehicularization post.) I even think I can make this work on “Kubla Khan”, where all this started.

These are complicated objects, these texts, these movies. They have lots of properties, lots of features. Figuring out which ones are critical to the underlying mechanisms and which are peripheral, that’s tough. It seems to go in layers. You start with the ‘outside’, the overall form, get a descriptive feel for that. Then the next layer begins to clarify. Perhaps you have to adjust your description of the first layer, perhaps not. Now you’ve got two layers of descriptive material. Now the third emerges.

Of course, ‘layer’ is just a metaphor. But it seems to be a useful one. The work proceeds in stages, each building on the earlier. It may also forces changes. Looks like the old hermeneutic circle, doesn’t it?

This thinking is exhausting. So I’m tired, good tired, but tired nonetheless.

Cultural evolution – and a book?

But I’ve been neglecting my cultural evolution stuff, the open letter to John Lawler that I mentioned back in August, along with the working paper that seems to have emerged from that, yet another run on the direction of cultural evolution in 19th century Anglophone literature. And behind that lurks that book on cultural evolution that I agreed to do well over a year ago, but haven’t gotten to it, yet.

Faculty Psychology: The Triple Soul

Theory of Mind 101: the triple soul

The ancient Greeks believed that humans had three souls, a rational soul, a sensitive soul, and a vegetative soul. Freud believed in Superego, Ego, and Id, also a trinity. We’ve got the Christian Trinity: Father, Son, and Holy Ghost. And, as I point out in an excerpt (below) from “Cognitive Networks and Literary Semantics”, other peoples have had similar triple conceptions of the mind.

Most recently Paul MacLean has given us a theory of the triune brain, in which a complex of neuro-functional structures of reptilian grade is embedded within one of paleomammalian grade which is in turn embedded within one of neomammalian grade [1]. MacLean’s account is about neurobiology while the others rest on somewhat different foundations. If MacLean is more or less correct – and this is by no means obvious, though I’m sympathetic to the idea – then one must ask: Are these other conceptions of mind rationalizations of the underlying neurobiology?

I don’t know.

A cognitive model of faculty psychology

It’s an issue I had to think about when, years ago, I developed a cognitive model for Shakespeare’s Sonnet 129. One of the things I had to do was develop a cognitive model of the Elizabethan account of the mind, which is derived from the ancient Greek version. Here’s that discussion, on pages 966 – 969 of “Cognitive Networks and Literary Semantics” [2]. It’ll be a bit obscure without the prior discussion, including some diagrams, but you should be able to get a feel for the argument. The complete discussion is online at [2].

* * * * *

The notion that man consists of three souls (or a single soul which is tripartite) and a body is deeply embedded in our own intellectual tradition. In Primitive Man the Philosopher (New York: Dover, 1957), Paul Radin has shown that the Oglala Sioux, the Masai, and the Batak of Sumatra also believe that man consists of three souls and a body (pp. 257-74). He goes on to suggest that such a belief may well be a cultural universal. It may or it may not be, but the fact that similar theories appear on four continents (North America, Europe, Africa, Asia) suggests that the task those theories perform, an account of human nature, is highly constrained.

I am presently entertaining the hypothesis that such a theory is an attempt by the cognitive network to explain the relationship between the SELF node and the rest of the nervous system. If one believes (and that is all it is, a matter of what one believes) the soul to be tripartite, then the SELF has two components, a body and a soul which is tripartite, with rational, sensitive, and vegetative sub- divisions. If one prefers to believe that one has three souls, then the SELF has three soul components and one body component, four components in sum.

The peripheral nervous system has two divisions, the somatic and the autonomic. The somatic system mediates voluntary control of the skeletal muscles and the activities of vision, hearing, touch, etc. The autonomic system regulates breathing, heart beat, digestion, the control of temperature, etc. Any episode which represents a transaction between the cognitive network and a sensorimotor schema whose intensities (first order input functions) are of soma- tic (perhaps just somatic motor) origin is cognized as being done by the body. The vegetative soul is cognized as the agent responsible for transactions between the network and sensorimotor schemas whose intensities are of autonomic origin. Episodes (which, you will recall, are conscious) in which one runs, jumps, spears hunks of meat, etc. are executed by the body. Episodes in which one feels hunger, lust, cold, thirst, etc. are cognized as being felt by the vegetative soul.