Lecture 19: Reflections on the Physical Symbol System Hypothesis (Turvey, 2019, Lectures on Perception)

The final lecture about the computational-representational perspective is focused on the general claim that intelligent action can arise from a computer, specifically a physical symbol system. This hypothesis underpins the computational-representational perspective; it proposes that an organism can come to 'know about' things via computations over symbols. The last few lectures have, of course, revealed flaws with this approach, but there are a few issues left to address.

First, complex systems are never solely symbolic. They can involve a symbolic mode, but this is always coupled to a dynamic mode. A simple example is protein synthesis: the protein that is made depends on the actions of symbols (the DNA code) and the dynamic process of how that assembled line of amino acids folds. In effect, the symbols serve as a constraint on the self-organising dynamics, which of course matters - self-organisation is only possible when there are constraints, or boundaries. So symbolic modes can be important, but they are, at best, half the story. 

Second, computation over symbol systems is simply not the only way to do things, despite being a very powerful way to do some things. Turvey discusses Rodney Brooks' subsumption architecture as a way of building robots that act adaptively in their environments without resorting to symbols or representation at all. Importantly, as we set up the ecological perspective, there are options to be found in dynamics. 

Third, even when symbols are present and serving as constraint, they are not typically immune to the effects of the dynamics. The standard dogma of the selfish gene has been replaced in modern biology with DNA a part of a dynamic process that can be 'tinkered with' by mutation but also by organism-environment interactions, etc. Turvey discusses the example of 'Pask's ear', a simple self-organising system that can readily learn to discriminate between two sounds. 

Fourth, Turvey discusses connectionism and how it is common to note that these networks do not trade in symbols (this was the core of the arguments back and forth in the 80s). 

Finally, Turvey discusses population thinking, and how things like species and immune system do not work at the order of anything giving them instructions, but instead by a process of selection from a population of variable individuals to respond adaptively to the environment. He draws parallels between this sort of selection process and the kinds of things we see in perception: the individuals can sense and tell the difference between themselves and other things, they can respond to an indefinitely large potential set of other things, and encountering specific other things leads to tuning and learning. 

He ends (pg. 296-7) with the 'top down fallacy', the unsolvable Catch-22 at the heart of every Cartesian programme:

First, if top–down knowledge is absent at birth, then the only means by which it can be acquired is perceptual. Second, any top–down knowledge acquired by means of perceiving can be no more meaningful and accurate than the perceiving by which it was acquired. If it were the case that perception is ineffectual in the absence of top–down knowledge, then top–down knowledge acquired perceptually would likewise be ineffectual. To cut to the chase, acquiring useful top–down knowledge requires acquiring true knowledge by means of perceiving. Only perceiving directly can ensure that.