I am tooling myself up on the ecological neuroscience literature, in part for fun but also because I am developing a project looking for the EEG signatures of informational couplings in coordinated rhythmic movements. More on that later.
I have been working through Luis H Favela's book, The Ecological Brain as part of this. The early chapters are set up (what is the ecological approach, what is neuroscience, why are they always fighting). Chapter 4 lays out the varieties of existing ecological neuroscience. He covers Reed using Edelman's Neural Darwinism (Reed, 1996), Anderson's neural reuse (Anderson, 2014) and Bayesian approaches such as the FEP (Bruineberg & Rietveld, 2019). He also touches on the concept of resonance (Raja, 2018). I want to briefly discuss this, because it impacts my current approach and Favela essentially rejects it out of hand as a viable option. I think he's wrong, but for interesting reasons.
Favela has four objections to resonance. He worries that the term is imprecise (too vague to be much more than a metaphor), unnecessary (resonance might just be an example of some other dynamical process such as synchronisation), and trivial (it might just be boringly true of coupled systems, rather than a particular distinct process). I'm not too worried about these, they all seem like they'll have answers.
The fourth objection is deeper; it might be wrong. Specifically, Favela worries that resonance is exactly the kind of thing Gibson rejected from his theory. Specifically, when one system resonates to another, there is an implication that the second system is copying something about the first. Gibson explicitly rejects this sort of similarity as the basis for perception - we don't perceive our environments because information is similar to it, but because even though it is not similar, it specifies that environment. He quotes Bickhard who also fretted that allowing in this sort of copying is to miss this lesson.
Favela goes one step further, and says that resonance sounds like it entails a structural correspondence, which then means it is acting as a structural representation (of the first system in the resonating system). Representations, of course, are not allowed, because they come with all kinds of problems. Given all this, Favela rejects resonance as the right approach for a Gibsonian neuroscience.
A Reply
First, I think equating 'structural correspondence' with 'structural representation' and then assigning the problems with the latter onto the former is an error. It struck me as a weird slip.
But more importantly, I think that resonance does entail a structural correspondence and that this is a feature, not a bug. Let me try to explain.
The issue of similarity comes up in ecological psychology because Gibson has rejected all the failed attempts to make sensations similar to the environment; it doesn't work, and even if it did, it wouldn't help. He replaces this with information, that is not similar to the environment, but can be informative about it because of specification. So Favela and Bickhard are right to be a little suspicious of any whiff of similarity doing work. However, 'rejection of similarity' per se is not actually a key tenet of ecological psychology. Gibson just happens to be able to reject similarity because of how information comes to be, and that is the important bit.
Formally, information is a kinematic projection of a dynamical property. Dynamical properties are defined with units of position, time, the various temporal derivatives of position, and mass; this description accounts for how the system changes and the forces involved. Kinematics lose the mass dimension. Going from the dynamics of the environment to the kinematics of information is a projection to a lower dimensional space, and the projection therefore cannot be identical to the thing being projected (think of a globe vs a map of the world). Perception works due to kinematic specification of dynamics (see this blog post for lots of detail and citations).
However, the similarity issue only applies to projections where the dimensionality changes. The resonance hypothesis is that information kinematics is projected into nervous system kinematics, via a process of resonance. Similarity is a viable option and, if you can manage it, a good idea - because the nervous system is how information gets into the muscles.
If you examine how a behaviour unfolds over space and time, what you see is the spatiotemporal characteristics of the information used to control the action (for example, in coordination, the HKB stability layout reflects the behaviour of the information for relative phase). This is evidence that a) information features in the control of action and b) that it didn't need to be processed into a form useful for controlling an action - it already had this form when it was detected. This is exactly what you would expect to see if the nervous system copies the spatiotemporal behaviour of the information; if information is projected into the nervous system via resonance.
What this means is that I reject Favela's rejection; I think resonance remains on the table as a perfectly sensible process involved in the neural contributions to the perception-action loop. The key is to pay attention to the specific details about how perception-action loops are formed; Gibson's key insight wasn't a rejection of similarity, it was identifying that he could make perception work without it, with specification.
A Little Addendum
Thinking this through has made me think about being more careful about how I talk about various parts of the perception-action loop. The loop works as a loop, and it remains the correct unit of analysis for behaviour. But it is made up of different parts, who are not all the same nor of equal importance. These parts do need to be assembled into something that can behave in the current task, and that assembly involves a few transitions.
- There are dynamical properties of the environment (including affordances, events and also surfaces, etc). The perceptually relevant properties are intensional, not extensional.
- These dynamical properties are projected into kinematic perceptual arrays as information, via the ecological laws governing that projection. The loss of the mass dimension (Bingham, 1988, calls this the perceptual bottleneck) means information is not identical to the dynamical property it is a projection of, but it can be specific to that property (another reason it matters that affordances are properties).
- Information is projected into the kinematics of the nervous system via resonance.
- Information in the medium (?) of the nervous system flows into the muscles and is used to control the behaviour of limbs.
- The behaviour of limbs is dynamic again, so that control is projected back into the level of dynamics, closing the loop with the environment
I have not thought this all the way through; but I think it works, and that it means we should stop talking about neural dynamics and start talking about neural kinematics. We also need to think if the nervous system counts as a medium for information (my intuition is yes but I think it is complicated).
No comments:
Post a Comment