Lecture 12: The Space Enigmas IV: On Learning Space Perception (Turvey, 2019, Lectures on Perception)

This Lecture is in roughly two parts. The first brief section walks through a Helmhotzian method for perceiving depth via unconscious inference. This inference process is learned (Helmholtz wanted to be an empiricist) but as usual entail loans of intelligence in the form of some givens not acquired via experience. This then raises a question: what exactly is experience, and what about it is used to be the basis for future inference? It turns out what counts as the relevant parts of experience can be very non-obvious, raising many problems that need more modern, less Cartesian solutions. 

For Helmholtz, the perception of depth is based on unconscious inferences based on prior binocular visual experiences with objects in depth. The ways in which sensations vary with the details of these experiences are learned via association, and what is learned is implemented as an unconscious inference to solve the problem at any given moment. Specifically, this reasoning is a form of abduction: reasoning from observations via a hypothesis to an underlying fact. In the case of depth perception, you would reason from sensations, using your learned unconscious inference about depth, to infer the cause of those sensations. (See page 179 for examples of the different kinds of reasoning). This can be expressed in Bayesian terms (hence the now-common use of Bayes theory to model sensation-based perception). 

The first problem is that this process needs a leg up from a loan of intelligence; the inference requires a hypothesis, and at least some parts of that hypothesis must be given (e.g. the assumption that sensations are caused by something). So even if a lot of the process is learnable, not all of it can be, and the required loan cannot be repaid (as per usual; Lecture 6). 

The second problem is figuring out what experience is required for the hypothesis to be learned in the first place. The usual approach is to figure out what it necessary and sufficient for a given perception to occur; what are the plausible and sensible building blocks. You then get these either from the sensations, or (more commonly) from the process that integrates and combines sensations and enriches them into perception. The problem is that what counts as necessary or sufficient can be far from obvious!

Turvey then describes many examples of non-obvious necessary and sufficient parts of experience required for something to be learned (a key name here is Gilbert Gottlieb). 

• chicks need to see their feet within 2 days of hatching before they will handle mealworms correctly
• Squirrel monkeys only show fear of snakes if they eat live insects
• honey bee waggle dance behaviour depends on the temperature during their incubation

Another important aspect of these behaviours is that they were often considered innate (not learned) until it was shown that you could interfere with their acquisition.

Turvey reviews many other examples. The point is Gottlieb's conclusion (called probabilistic epigenesis): development unfolds in the context of  dense web of interconnected parts at multiple scales (genetic, neural, behavioural, environmental). The connections are loopy, and describe a complex nonlinear dynamical system that unfolds over time, and has it's own internal structure (attractors and the like). Development is not a program to be followed, but a dynamic who's form is execution driven as the dynamic plays out in the specific local context. The dynamical system exhibits stability and resilience to some but not all perturbations, and how it responds to perturbations can be extremely 'non-obvious' because of the nonlinearity. But the obviousness or lack thereoff is not a feature of the system - it's a limitation of the scientist. 

In summary: Helmholtz provided a sophisticated attempt to use experience to build hypotheses to power unconscious inference on sensations in order to derive perception, but a) even he entails problematic loans of intelligence and b) the role of 'experience' is actually distributed across an organism-environment system (c.f. Lecture 1) and cannot be placed in a program to be executed.

Turvey ends by accusing Fodor of occasionally thinking like an ecological realist :)