By Jeff W. Dahms
This short essay proposes that a simple principle of functional design structures the cognitive operation of all animal brains. Based on this functional template it suggests a unique evolutionary path for the development of the human mind/brain and outlines some biological and psychological consequences.
Models in science can be thought of as ranging from the tightly empirical, frequently mathematical to the distinctly metaphoric (e.g. Ranging from say the standard model in physics Q.M./Relativity theory at one extreme to perhaps Freudian psychodynamic theory at the other). Issues of the quality of various types of models and the kind of distinctions assumed by such a scale are perennial topics in the philosophy of science. For the moment it is useful in notionally locating this model - say mid range on the above scale. The account is meant as an heuristic metaphor to some degree, but it has a substantial degree of biological content and literal intent. Its purpose is to provide an overarching interpretative perspective on a vast area covered by many disciplines. We need the vision to guide the enterprise of determining the nature and function of the parts which will then help in redefining the original vision. Historically, Darwin's model, for example, has many of these features, being put together before we knew about genes or mechanisms of variation and natural selection. It served as a conceptual guide to be expanded, added to and subtracted from and as a broad framework to focus specific research domains in biology.
The challenge is to propose a model of mind/brain that is general across animal species, makes the least number of robust biological assumptions and yet is rich enough to account for the range and character of the human psyche.
The coupled reference "mind/brain" is used to indicate a mix of biological and psychological language in the account. The deep, difficult, philosophic issues (the nature of consciousness, etc.) are being left aside in addressing the apparently more straightforward question of cognitive processing. Elucidating just these cognitive operations seems an intractable problem to many.
So what are the kinds of features that a model of the human mind/brain will have to handle? Experience of our own mind/brain is of a world of shadowy, ill defined phenomena. Our cognitive processes range from trivial, crude, stumbling meanderings to the computationally impressive to the occasional creative and inspirational insights. One of the mind's very curious properties is the way its ordinary mechanism of operation is largely hidden, even from our best attempts at introspection. Consider the nature of the bulk of our ordinary mental operations. They are very fast, subtle and complex. e.g. Meeting someone and from a two second impression strongly liking or disliking them. If we were asked about this we would usually tell a "just so" story - a lengthy description of the person's features, our relevant past experience and how we put all that together etc. If questioned about what actually took place in those two seconds, it is apparent that we know only about the outcome. We infer, quite reasonably, that the above mentioned factors must be central in some way to what did take place but we did not actually witness the processing. We can only conceive of the process that actually takes place by substituting for it a very lengthy simulacrum - the presumed equivalent that might take place if our relatively slow verbal and serial concept processing had to do the job. All of our experience is either directly of relatively simple elements - words, concepts, feelings, sensations, etc., or indirectly of the outcomes of many high-speed processes functioning out of the range of our awareness. So the task is to provide an account sufficiently rich in structure to make sense of features like the above along with the many other characteristics of the human mind.
We know a lot about the general structure of human and animal brains, less about their broad mode of operation and almost nothing about their deep operating principles. Our chief perspective is of brains as bio computers. Quite reasonably we have started to address the question of their mechanism of operation using our knowledge of neuroanatomy. Brains are very big complexes of neural networks so we try to model cognitive operations using computer generated neural net models. Cognitive scientists are making very worthwhile attempts at simple neurocognitive models using bottom up approaches, but there is clearly a long way to go on even the most optimistic accounts.
The scientific usefulness of the proposed model lies in how well it functions as a device for structuring cognitive and psychological issues, for making sense of the range of complex functions of the human psyche and for shaping neurocognitive and psychological research. In addition it attempts something more personal - it potentially provides a deeply satisfying way of making sense of self and of human possibility.
The first part of the thesis is: All animal brains, including human, function as if they operated using hierarchal multi-level machine languages. This contrasts with the single level machine language of most computers - the processing of the on and off bits at the bottom level of the computer's layered information structure. To make the metaphor concrete, visualise a pyramid shaped assembly of neural layers with computation/processing occurring horizontally across any or all levels and information outcomes moving up and down the levels. Many parts of animal/human brains are in fact structured in an anatomical hierarchy. e.g., The visual system. The suggestion here is that the cognitive processes operate as if, literally, the neural componentry was arranged in anatomical hierarchal levels - and perhaps some of it is.
The way it operates is that levels are filled with data and organised as the organism learns; directly coding simple information at lower levels and then re-coding at higher levels for groupings of lower level information. Grouped data at one level may be represented and/or accessed by a much smaller operation at a higher level and so on up the hierarchy with the lower levels representing the simpler elements. Cognitive processing occurs horizontally (in the metaphor of the layered pyramid) at every level where sufficient encoding for groupings of lower level data has occurred.
There are major biological design advantages to such a principle. Neural processing can occur at any level of the brain's neural hierarchy using the coded outcomes of previous learning, rather than having to reduce all data to the lowest unit of composition for computation. This is obviously a speed and complexity advantage in biological systems where neural transmission rates are very slow in contrast to EM rates (100s of metres/sec vs 300,000 Km/sec). Considered as a general design principle, for any organism the optimum neural functional arrangement is one in which, in the shortest possible time, the organism can access a distillation both of its previous experience and of all the current incoming data. i.e. A layered hierarchal system. This rationale is very important in considering human cognitive processing later.
Human cognitive/conceptual capacity developed in response to somatic demand over the last 4-5 million years. The record is very patchy but major bio and cultural events are on the timetable - timing of upright posture, the opposing thumb, development of the larynx, growth rate of the cortex, tool making, cognitive representation and so on. Evolutionary developments such as laryngeal changes enabling speech, an opposing thumb and upright posture for tool manipulation, generated very strong selective pressure for the development of cortical computational support - abstraction and cognitive processing.
In the animal cognitive pyramid model, the lowest layers represent the simplest learned elements and their combination. The pressure of evolutionary development was largely on these lower layers of the animal neural hierarchy - the precursors of the corresponding human abstracting and concept forming capacities.
In the absence of teleological "specifications" for building a sophisticated thinking machine out of an animal brain, one route was possible without the need for very complex neural design specificity.
Firstly, it was possible to generate enormous computational power by a relatively simple genetic maneuver. It was achieved by greatly increasing the number of neurones (eventually to around 10 to the 11th power) and the number of dendritic connections per neuron (eventually to around 10 to the 4th or 5th power). As a rough computing index this provided more combinatory arrangements than there are particles in the universe.
Secondly it was possible to genetically specify very general operating principles. The rules might have been like - continuously, randomly link and associate new and stored data using features like similarity and difference. This would have provided an enormous field for cognitive selection with positive operational feedback gradually reducing the randomness of the linkages.
Such a process would have been very wasteful of neural computational power and would have produced an extremely low efficiency thinking machine. The compelling argument in favour of it, is that it is a robust evolutionary path which can be proscribed by relatively simple mechanisms. Some somatic mutations - e.g. those allowing speech and tool using - would greatly amplify tiny gains in cognitive power. Very crude and error ridden cognitve processing could have been strongly selected for. It would only have to be marginally better than that which it replaces. The greater the compound selective advantage, the tinier the cognitive margins that could be selected for and the greater the operational price that could be paid.
There are major biological and operational downsides to this route. Neural and dendritic proliferation resulted in massive cortical growth. The selective pressure for this degree of human cortical development must have been very high as it took place in spite of severe inhibitory evolutionary pressures. e.g. The cortical growth produced a quadrupling of brain size. This resulted in a foetal head size that in the wild would risk the life of a significant fraction of females and foetuses in childbirth. Some 25% of the body's energy output was needed just to have this (in most animal terms) overgrown nervous system passively available. Continuous, random linkage and association plus slow refining by cognitive selection is likely to produce a large burden of weak and useless abstractions and much inappropriate conceptual linkage - cognitive junk. Additionally, in such a hugely complex, inefficient system considerable fragmentation of function would occur. Without central coordination multiple partial centres of control would develop - rather like a large country with a widely distributed human population that breaks up into uncoordinated even warring sub groups.
The above psychological benefits and costs would produce an animal that was cognitively powerful and therefore successful (on the average for most members of the species). But it is also likely to produce individuals who intermittently were cognitively dysfunctional and perhaps dangerous to themselves and to those around them. It may also be the case that this operational dysfunction is responsible for our loss of simple animal coherence both physical and mental. The problem is created by extreme evolutionary success. Nature is willing to pay almost any price when the outcome is so advantageous.
Specifications for an "efficient" thinking machine meeting human biological requirements are probably very subtle. A.I. scientists are still unsure, even in principle, of the software rules for a human equivalent thinking machine, no matter what the computational power. It may be that nature's requirement for simple genetic specification and graded intermediate stages of development severely limits the kind of models we can propose for human brain development.
This account of massive but low efficiency neurocognitive development when combined with the multi-level universal brain model has many explanatory ramifications. The size, power and number of functional levels that are possible simply because of this combination can be orders of magnitude (perhaps many) larger than those of the earlier primate brain with only a 3-4 fold increase in brain volume. In other animals, the neural hierarchy provides simple learning processes in the lower structures and speed, complexity of response and tight integration of function in the upper - much of it hard wired . In the human animal, the lower levels provide very large capacity simple learning plus routine abstracting and conceptual functions. The upper levels of the neural hierarchy are largely potential but can provide inordinate integrative capacity depending on whether sufficient input into the lower layers is made to generate them.
So how does this functionally layered, human cognitive pyramid work? Surprisingly, the human properties that we often think distinguish us from other animals and which we generally prize most are at the bottom of the pyramid. Human evolution was driven most powerfully by the need for the development of the primitive cognitive structure at the base of the early primate pyramid. It's interesting that our computers serve this same role with respect to us. At the base are the slowest, simplest elements - concrete reference words perhaps. Above this are simple compound notions; above this again are more complex abstractions and so on. Further up, the model suggests there are more levels of actual and then many more potential levels of processing where combinations of combinations can take place representing vast and subtle aggregations of data.
An interesting consequence of the model is that upper level processing would not have distinctions that are a feature of the lower levels. e.g. The difference between thoughts, feelings and sensation. These would be homogenised by the grouping processes, but the outcome, available to awareness after processing, may produce quite novel, complex thoughts and feelings.
Much of our routine cognitive activity requires processing at a level just above our general abstractive level where new abstractions and the stored outcomes of previous conceptual linking operations are combined at high speed - as in the example given at the beginning. i.e. Meeting someone and from a two second impression and strongly liking or disliking them. All of the processing above the complex abstractive level (including most of our everyday cognitive processing) is out of the range of our awareness although we continuously have the outcomes available to awareness. This is probably a simple feature of the temporal design of our perceptual structure. Perhaps this is the territory of the infamous human unconscious. The model also implies that it is the higher forms of human cognitive processing that parallel animal cognitive processing.
Upper level operations are the source of all our artistic/scientific/creative endeavours. The levels are acquired in the same universal way - learning fills in the lower levels and grouping at these levels results in encoding at a higher level which can then be further cross linked and so on. They all process out of our range of awareness but we get to experience some unknowable proportion of the outcomes. At very high levels of synthesis, a distilled sense of self and compounded perceptions of the whole scheme of things may take place. This is may be the origin of that generic experience - the religious sense (meaning a global, unitary, cognitive/affective/aesthetic experience). Ironically the rationale for the existence of this highest level human synthetic process is the most ancient of animal brain design principles - organise neural function so that in the shortest possible time, the organism can, in principle, access a distillation of all its previous experience and of all the current incoming data.
Human cognitive evolution - the main stream - has centred on producing abstracting, conceptualising support to amplify the advantage of our somatic mutations. This was derived from an equivalent level structure in our primate ancestors. The advantages of the main programme have been so strong that nature has been willing to carry a very large burden of necessary negative design consequences. High levels of cognitive dysfunction are the norm - even individuals who destroy themselves and everything around them.
The downside features do not significantly effect the main stream - the central selective process - until they produce some huge disadvantage such as damage to the reproductive success of the whole species. By the time our genes register that fact there is no chance of evolution significantly altering our genetic programmme. The upside features of our functioning are also a by product of our mainstream cognitive evolution. Human evolution required "bulk in the basement" so to speak - a massive conceptual apparatus at the lower levels. The universal animal neural capacity for hierarchic organisation meant that vast integrative structuring of these lower levels was possible (though probably not at all necessary for human evolution). These higher level functions probably only emerged in recent history (thousands of years), too recent for impact on the much larger time frame of genetic evolution. In addition, cultural evolution may now be our overwhelming mode.
The rationale for this model is its simplicity and explanatory power. It locates us firmly in the animal world (though in an unusual way in that our upper level capacities parallel the mechanisms of ordinary animal neurological processing). It is grounded in fairly robust assumptions about evolution and brain function. It provides a naturalistic account of the most ordinary and the most extreme of human psychological properties from science to religion to the unconscious. As a heuristic device it provides a rich structure for considering many biological, cognitive and psychological questions. The division of cognitive activity into the evolutionary domains, "main stream" and "derivative", makes sense of features otherwise very difficult to explain. The model suggests a format for the design of neural networks in simulating complex functions. It also suggests ways to map otherwise elusive global psychological properties like intelligence.
Psychologically it puts us in a very interesting position. Our brains are completely paid for. The evolutionary equations at the lower levels of cognitive processing - the main stream - are balanced. The cost/benefit determination involving energy, survival chances and reproductive success has carried the whole brain structure to this point. Yet nothing in our design necessarily prevents us from reducing, eliminating, or compensating for our design downsides necessary as they might have been just to get us to this point - though this may be a very big ask. Our potential, our future, is in the upside features which are an accidental evolutionary gift without furthur cost or specification for use - a great gift from mother nature to its inheritors if they can learn how to use it and survive the downside cost of its progenitor.
The rich precious secret of our destiny is we don't have one - we are free in the real sense.
|Generously Supported by Michigan State University and The National Endowment for the Humanities.|