vTAO
Geertz, nor Lakoff, et. al., do not make any attempt to construct a model of hermeneutic cognition.
Neural networks, an underlying technology within the field of AI and, in particular, a metaphor describing the operation of neural nets might provide a basis for constructing a hermeneutic model.
A quick overview of essential aspects of neural nets.
A neural net consists of thousands of individual nodes connected to a multiplicity of other nodes. The function of each node is limited to “firing” (discharging an electrical impulse) or not firing. Firing is determined by inputs received from connected nodes – the sum of those inputs needing to exceed a threshold value. The threshold value (node weight) is adjustable.
The first ‘layer’ of nodes receive input from sources external to the net. These are channeled to multiple ‘layers’ of internal nodes and, eventually, to an output layer. Feedback is used to adjust the weights until a given set of inputs results in a defined set of outputs – the net has “learned” the expected behavior.
Hopfield provides an interesting metaphor to explain the workings of neural nets: a topographic surface, a landscape.
“Consider a countryside laden with hills and valleys. In the valleys lie lakes. If you pour a bucket of water on a hill, it flows down the hill into one of the lakes. No matter where you pour the water, it will eventually come to a place to rest; the system of mountains, lakes, and flowing water will eventually reach a stable state. … neural nets have contours like the hills and valleys in a countryside; they also have stable states.” [Hopfield 1986]
The landscape’ arises from the “training” of a network: inputs are channeled to outputs; feedback is provided (did the inputs end up at the expected outputs); and node weights are adjusted. High value weights create the peaks, low value, the valleys. Once formed the topography (Hopfield calls it an e-surface) consistently channels inputs to the expected outputs.
Figure 1. The Hopfield metaphor illustrated. The connected symbols represent the physical network, the ‘landscape above is the virtual e-surface.
Early neural nets were trained for specific tasks like recognizing a cat in a photograph. (Or tanks in a military application.) Sophistication and elaboration of learning algorithms created more interesting applications like ChatGPT. But the operation of such nets remains consistent with Hopfield’s metaphor.
Network topography the e-surface is virtual – not physical. Changes alter the e-surface; they do not restructuring/rewire the network itself.
If we want to use neural nets and Hopfield’s metaphor to model cognition in an organism, e.g., a human being, some elaboration is required.
First, emphasize that inputs come from an environment external to the organism we are modeling; and, that at least some of the outputs have the effect of altering that environment.
Second, instead of a simple threshold, we have a constraint window (many computer-based neural network models do this already) with both a threshold value and a ceiling value.
Third, and most important: is the introduction of a new concept, “Constancy,” that alters inputs to the net with a qualitative attribute.
In a standard neural net, inputs are simply signals and either exist or not. Input signals are summed and, if they exceed an established threshold, they are passed on to the next “layer” of the net. Weights are adjusted by feedback.
‘Constancy’ assigns an attribute to the signal: a value resulting from a function that synthesizes how often the signal is received per unit of time (frequency); the strength of the signal being the same each time received (consistency); and, the signals are received, mostly, via the same set of input nodes (regularity).
The higher the Constancy, the lower the threshold determining if the signal is passed on. Higher Constancy yields “valleys” or reduces “mountains” in the topography.
Why is Constancy important?
It tends to stabilize the net, the topography while providing a different criterion for “learning.” Instead of ‘right’ or ‘wrong’, a kind of “complementarity”—outputs as complements of Constant inputs.
Far more importantly is provides a mechanism by which multiple organisms can co-evolve. This is very similar to the “functional coupling” offered by Maturana and Varela.
It also provides a mechanism, metaphorically, for how culture might emerge. Output from one organism increases the constancy of an input from the environment, thereby affecting the Constancy of that input for itself, AND, any other organisms in the immediate vicinity. Multiple nets begin to conform to a common topographic feature.
The result: a model, vTAO: virtual (the e-surface); Topographic (the metaphoric landscape); Adaptive (changing itself both within, adjusting threshold windows, and by altering the environment to make it more constant); Organism (a biological entity, e.g., a human being).
Although organisms do make changes to gross physical aspects of the environment—building a house, damming a river—most changes will be to micro aspects—painting on a cave wall, a song—or conceptual or behavioral.
It should also be noted that the environment changes independently of any changes made by organisms.
The vTAO is a model of a single system. There is but one topography combining both organism and environment. The topography is dynamic and constantly changing. Some changes originate ‘within’ the organism, some from the environment. Because some outputs from the organism alter the environment so that Constancy of specific inputs increases, establishes circularity. The topography can be seen in terms of co-evolution or, more correctly, the dynamism of a single system.
Extending the Metaphor
A physical landscape is shaped by multiple forces. Might we find analogs for our virtual landscape, and thereby extend the metaphor?
Several suggest themselves:
Cellular [planetary core] – Bergland and Conrad discuss in detail the information processing capabilities of enzymes and cellular (and sub-cellular) entities. Inputs at this level are highly constant, usually variance is seen only as a consequence of mutation or malfunction.
Organismic [tectonic plates] – Maturana and Varela, extended by Winograd and Flores, show how organismic organization affects cognitive behavior as well as the arbitrary classification of behavior into “intelligent, cognitive, and aware” on one hand and “instinctive, stimulus-response, and non-intelligent” on the other. Their general argument, widely accepted, is that so-called higher functions like cognition are structured by the organism and its ontogeny and phylogeny.
Sensual [planetary crust] – the level of the five known senses. The effect of constancy of inputs at this level is readily apparent in, at least, two ways: outputs that increase the constancy of “pleasurable” inputs, decreasing “unpleasant;” and, somewhat ironically, the “attenuation” that ensures the more constant an input, the less likely we will be consciously aware of it. Constant background sounds provide an excellent example.
Cultural [geography] – this level embodies the heart of the hermeneutic argument; alterations of constancy giving rise to the social construction of meaning, the cultural parameters of cognition, and the behavior-symbol-context-cognition associations that are empirically evident.
Habitual [landscape] – the primary source of individual variation as well as constancy. We are all creatures of habit.
Analytic [architected surface variation] – A thin veneer layer commonly associated with “thinking,” the kind of activity at the focus of efforts in AI and cognitive psychology. Constancy is least evident, but clearly present, at this level.
All of these ‘layers’ operate simultaneously, they are not hierarchic, nor are they reductionistic. Just as an orchestra combines and synthesizes multiple threads of sound to produce a single performance.
On other presupposition. The most useful definition of a system comes from von Bertalanffy: “a set of elements and the relations among them.” Every element is connected (related) to every other albeit, in some cases via one or more intermediary elements.
Language forces the appearance that some aspects of the net are localized, particularly in one of the vTAO layers, but this is not accurate.
Consciousness
Might the vTAO model inform, at least metaphorically our explanation of consciousness? Perhaps; but first a working definition:
“Consciousness,” as defined in Wikipedia: “subjective awareness of yourself and the world; thoughts, feelings, sensations, and existence; the ability to process internal and external events, forming a unified worldview.”
Parsing this definition, specific aspects of consciousness are discernable:
Differentiation: recognition of fundamental differences, e.g., [food | not food | anti-food], [here | there].
Association: patterns in the inputs that imply this set comes from “That,” this other set comes from “There.”
Categorization: as inputs are differentiated and associated, along with variable Constancy, the topography becomes patterned; “features” emerge analogous to mountains, valleys, watersheds, lakes, ranges, …
A finer grained categorization of inputs also occurs. Color perception would be an example, ala the work of Berlin and Kay on color names in languages.
An efflorescence of categorization occurs at the sensory layer of the vTAO—establishing a foundation for the emergence of language.
A noteworthy instance of categorization arises from the fact that certain inputs are categorized as originating “from within” or “without” as are output ‘desitnations.” A foundation for the perception of “Self—Other.”
Significance: is a kind of quantifiable overlay on the categorized inputs to and ‘features’ of the topography,
As inputs and “features” are categorized, “things” with “attributes” emerge, as described by Brian Cantwell-Smith in his book, Origin of Objects.
A kind of mindless “survival instinct” that led, at the cellular level, to differentiate between ‘food’ and ‘anti-food’ applies to things-with-attributes. Some are “desirable,” others “dangerous,” and survival depends on quickly differentiating among (based primarily on associated attributes) and appropriate reaction to them.
Awareness: is the foundation for consciousness and arises from Constancy. The more Constant an input, or the aggregate of patterned inputs identified as a thing, the less we are aware of it. This is attenuation.
Metaphorically, attenuation results in an invisible plane separating parts of the topography above the plane from those elements below. Think of an iceberg separated into the visible part above the sea and the rest, out of sight below.
That to which we attend, are aware, and the rest. Conscious and Unconscious.
It is essential to remember; even though they may not be attended to, those inputs, those patterns discerned as things are still present, still shaping, constraining, and modifying the overall topography.
Naming: outputs from the vTAO might include sounds or marks that become part of the environment and alter the Constancy of specific inputs or, more likely the aggregated inputs associated with differentiated things. This is a very basic, very simple, “naming.”
“Names” become part of the environment affecting Constancy, of course, with two very important outcomes.
First, they become the foundation for establishing language; and second becoming “evocative triggers.”
As the virtual topography forms and evolves, more and more of its totality exists below the plane of awareness. The ‘non-Conscious’ elements (below the plane of awareness) of the topography are still present, still constraining/shaping those elements above the plane, those we might say, “populate the Conscious.”
Names are part of the Conscious realm and encountering them had the effect of recalling to Consciousness aspects of the topography that, because of attenuation, become part of the Nonconscious. This is evocation.
Enter Culture
There is abundant evidence of strict limitations on the cognitive power of human brains. One such limitation is the latency period – the time taken for a sensory signal to travel to and through the synaptic network to the point that a response is generated and expressed.
Another limitation concerns the amount of conscious processing that can occur at any one time. “Miller’s Magic Number Seven (plus or minus 2)” is an example. Conscious is the key word here. The size of our neural net is such that almost unlimited physical processing is available but the vast majority of that processing occurs non-consciously.
Naming and the establishment of evocative triggers, discussed above, provides the means to bypass these limitations. Vast amounts of cognition occur nonconsciously and “recalled to mind” only as needed. The most obvious, and interesting to this audience, is found in ”culture.”
One of the functions of culture, according to Hall (1977): “is to provide a highly selective screen between man and the outside world. … Culture shapes what we pay attention to and what we ignore. This screen function provides structure for the world and protects the nervous system from information overload.” Culture keeps the amount of conscious processing necessary to survive at a level consistent with Miller’s observed limitation on our ability for conscious information processing.
Culture predisposes behavior by altering constancy of available environmental inputs – it provides, in the words of Robert Barker, “a highly structured, improbable arrangement of objects and events which coerce behavior in accordance with their own dynamic patterning. We … could predict some aspects of children’s behavior more adequately from knowledge of the behavior characteristics of the drugstore, arithmetic classes, and basketball games they inhabit than from knowledge of the behavior tendencies of particular children.” (Cited by Hall 1977). Much more recent work by Lave, Nardi, and Seeley-Brown confirm this kind of role for culture as a shaper of cognition.
The vTAO model suggests that culture regulates cognition by assuring the presence of high-constancy inputs. High-constancy inputs shape general – but persistent – features of a net topography. High-constancy inputs are almost always processed at a non-conscious level (we are oblivious, in fact, to most of our own cultural background) providing the focus (ground-object differentiation) necessary for conscious thought while simultaneously reducing the volume of inputs to match our ability to process those inputs in a conscious manner.
All living organisms have the ability to change their environment. Humans are unique in the range of modifications they can and do make. Human modifications, collectively, predispose individual organisms with regard their outputs (behaviors) thus ensuring high levels of Constancy without preventing novelty or innovation.
Modifying the environment to fix a set of inputs is the creation of an “Input Complex (IC).” IC’s can be semi-permanent or transitory. They can be fixed or portable. All aspects and artifacts associated with culture are a form of IC: modes of dress, specific colors, a book, a film, even ritual behaviors. IC’s can increase or decrease the constancy of inputs. IC’s are overlapping and redundant (minimizing error).
Any part of an IC can evoke the whole – in a manner similar to the smell of cinnamon evoking complex memories of baking day at your grandmother’s house. This is one way that the Miller’s Number is mitigated – another being Miller’s own concept of “chunking.” IC’s are also recursive (probably fractal) another way to address human conscious processing limitations
Human beings are constantly creating and modifying existing IC’s but that activity is moderated by the “inertia” of the mass of existing IC’s. Culture does change but, usually, slowly.
Human environments are saturated with IC’s:
“It is a significant fact about human cultures, that for the past 50 thousand years, the total amount of information transmitted from generation to generation has been increasing rapidly.
One way to measure the size and importance of this transmitted pool of information we call “culture” is to observe the things and events surrounding oneself, and note how much of one’s environment is a product of this information pool.
Quantifying information in terms of ‘chunks’ [what we are calling IC’s], or symbolic units which can be held in short term memory, it has been estimated that about 50 thousand chunks are necessary to speak a language with reasonable proficiency. Given this estimate a figure of several hundred thousand for all of the cultural information known by typical adult is quite conservative.
An estimate of several hundred thousand to a million chunks known by an individual does not indicate the total size of the cultural information pool since of the characteristics of human society is a division of labor in who knows what. The total size of the information pool could be estimated in the tens of millions to billions of chunks.” [D’Andrade 1981]
I believe D’Andrade to be overly conservative because so much of that information pool is nonconscious.
Classical computation based on recreating – representing – all this information inside of a computer program is obviously daunting, certainly impractical, and probably impossible.
An AI, like ChatGPT, has access only to the information pool that has been made explicit—written down or otherwise made overt. Moreover, only to that which has been digitized.
vTAO however does not face this challenge. In the vTAO model the network and the environment are informational complements of each other with the effect that huge amounts of the “processing” required for cognition is performed in and by the environment – not the net.
Back to Consciousness
The motivation of this essay was the question: can AI inform our investigations of or understanding of consciousness?
The answer was no; given both the popular, current, understanding of what AI ‘is’, and the formalist approach that permeates investigations of mind by the AI, Computer Science, community.
However, if we consider the study of consciousness to be at the “fringe of science” and cognizant of Quine’s dictum, perhaps a metaphor, one derived from within the AI realm, might serve us.
The vTAO metaphorical model is offered for this purpose. Some ways it might support investigation of consciousness include:
The self-other distinction. Metaphorically identified as a kind of “continental divide” in the virtual topography.
A specific area of the human brain, (medial prefrontal cortex, mPFC) begins to show electrical activity roughly 3-4 months post-birth; activity that increases for 2-3 years. It becomes quiescent when, e.g., a Zen Monk meditates; perhaps accounting for the sensation of “being One with the Universe.”
Are there similar mechanisms in the brains of other animals, and if so, are the results at all similar to what is observed in humans? How would we know?
Embodied Mind. The AI community seems convinced that “mind” is nothing more than electrical activity in the brain. An essential assumption is one hopes to replicate mind in a computer. The vTAO model-metaphor is supportive of and might provide additional insights or mechanisms for exploring how body, environment, even universal (re: quantum consciousness) connectivity shapes the human mind.
Perhaps, research might support Schrodinger’s assertion: “The total number of minds in the universe is one. In fact, consciousness is a singularity phasing within all beings.”
Intelligence. Two possibilities arise in this area.
One, Ian McGilchrist has written, extensively, about two modalities of thinking: one focuses on connecting to and with the world, the other on manipulating the world. The latter is associated with the Analytic layer of the vTAO model. The other ‘layers’ of vTAO, collectively, support and are consistent with the first modality.
Second, the vTAO provides support for a distributed or comprehensive understanding of intelligence including proprioception, hormones, and even the theory of situated learning where some aspects of intelligence and knowledge arise from the environment, not just neuron firing in a brain.
The unconscious. The topography of the vTAO, at every moment, is shaped by simultaneous forces from all six layers. How much we are “aware” of the forces at each level increases as we move from cellular to analytic. But, at no level, are we one-hundred percent aware of the inputs we are receiving and responding to.
Just because we are not aware, does not mean they have no effect. All of the layers of the vTAO are, simultaneously, receiving and “processing” those inputs and shaping the topography. And awareness can expand to include inputs ordinarily ignored. The “Cocktail Party Effect” is an excellent example of this.
Human individuals have cellular, organismic, and sensual layers in common. There are some elements of culture that are universal, cross-cultural. Most cultural knowledge is tacit; individuals within a culture are mostly unaware of it. Even habitual and analytic levels are constrained by topographic forms arising from other layers.
Might these collective constraints, these inputs of which we are unaware, constitute Jung’s collective unconscious?
A more remote possibility: Rupert Sheldrake’s “morphogenetic fields” be a way of interpreting the commonalities that humans share at the cellular, organismic, and sensual layers?
Altered states of consciousness. Two avenues of exploration suggest themselves.
Starting with Huxley’s speculation that hallucinogens remove “filters” that prevent us from directly apprehending reality. The vTAO topography can be viewed as constraints imposed on the flow of energy in the net. Constraints are, effectively, filters.
Huxley’s experience with mescalin parallels, in interesting ways, aspects of meditative and mystical philosophies. The analytic layer of the vTAO can be seen as Maya and eliminating the filters that normally prevent awareness of the whole of one’s self. the outcome of the meditation. Poetically described as escaping the “valleys’ that dominate the analytic realm to view/experience the entire topography.
A second, highly speculative, possibility; some kind of mapping of the varieties of altered states to the vTAO layers. For example:
· Establishing micro-environments that magnify the Constancy of specific sets of inputs, e.g., religious spaces or shamanic rituals, to induce an altered state of calm, serenity, and awe.
· Ayahuasca affecting brain regions associated with self-other (the aforementioned mPFC) and or empathy while removing the ‘dulling filters’ imposed on the senses by survival constraints, thus creating the familiar “communing with Gaia” experience.
· MDMA might reduce the Constancy of certain inputs (or desensitizes the perception of them) suich that inhibitions are shed, cultural norms transcended.
· LSD reshapes the topography as a whole, beginning with the sensual layer, and results in “alternative interpretations,” along with changes to the outputs associated as “within.”
· The more dramatic effects of DMT might be traced all the way down to the organismic and cellular layers.
Consciousness. Nothing said here directly addresses consciousness; at least in terms of a definition or an explanation. However, …
· Differentiation is evident at the cellular level. E.g., food // danger.
· Self—Other distinction is evident at the organismic level. E.g., activity in a specific brain locus.
· At the sensual layer, the ‘continental divide’ from the cellular/organismic layer would give rise to classification of sensory inputs “from within” and those “from without.”
· Attenuation of inputs with high constancy creates an “awareness” threshold with the vast majority of inputs occurring below that threshold while, nevertheless, shaping the topography.
· Then, add language, especially language that implicitly or explicitly relies on the verb, “to be.”
“Consciousness would be an emerging, but not “emergent” property of human beings. The last bullet, re: language, might provide a means for exploring the differences between animal and human consciousness.
Conclusion
The currently popular notion of AI is but one aspect of an interesting discipline. One that is very unlikely to yield any insights into human intelligence or human consciousness.
The discipline of AI consists of multiple threads, using multiple models, mechanisms, tools, and metaphors that might yield interesting and profound results. If they manage to survive the seemingly inevitable collapse of the current “AI Mania.”
The vTAO might even be implemented in some fashion, made concrete rather than metaphorical, and be added to the disciplines toolkit.
In the interim, it might be an interesting and useful guide to research in disciplines, like the Anthropology of Consciousness, outside the realm of Computer Science or its applications like AI.