Radical Embodied Cognitive Science.

“Radical” is an understatement. This book isn’t just cognitive science with digital computers and no explicit symbolic thought. This is cognitive science via coupled oscillators — without even a hint of anything that could look like a predicate, a concept or even an explicit variable (Implicit variables are OK, since the modeling is based on differential equations).

Let that sink in for moment.

I have to say that I admire this approach. It pushes so hard in the non-symbolic direction that at first glance, there isn’t even the glimmer of a symbol left. I won’t say that I’m convinced that this is the approach that will finally work, but its utility is undeniable.

To quote from the book:

Radical embodied cognitive science, very roughly, is the thesis that cognition is to be described in terms of agent-environment dynamics, and not in terms of computation and representation

The primary adversary here is the a priori “Hegelain” argument that something is not possible, or some form of symbolic representation is absolutely necessary (I quibble with the term Hegelian — I place this type of argument far prior to Hegel).

The primary adversary here is the a priori “Hegelain” argument that something is not possible, or some form of symbolic representation is absolutely necessary (I quibble with the term Hegelian — I place this type of argument far prior to Hegel).

The Radical Embodied approach is designed to be very empirical. However, since it is early in its development, the experiments are simple and suggestive, rather than complete and convincing. Again it’s early, the hypothesis is fascinating, worthy of exploration and much more radical than Clark’s supersizing or Barrett’s beyond the brain.

My sympathy to Chemero’s approach stems from its being based on Gibson’s affordances and Barwise & Perry’s Situated Attitudes.

Barwise and Perry argued that we can’t understand meaning or cognition without taking into account that thinkers are spatially located (i.e., situated) and so have only incomplete, locally available information at their disposal. Every thinker and speaker is someone, who is somewhere, and who is aware of only certain things

 

Chemero’s models cognition using nonlinearly coupled systems. This allows for a rich set of interactions which can evolve over time, and (may) support the necessary functionality but which are inherently difficult to modularize.

Because the agent and environment are nonlinearly coupled, they form a unified, nondecomposable system, which is to say that they form a system whose behavior cannot be modeled, even approximately, as a set of separate parts

 

Since his focus is on Gibsonian affordances, these nonlinearly coupled systems are unsurprisingly well tuned to “being in the world”.

 The action is part of the discrimination, and the discrimination determines the final form of the action. The rotational scanning on the way to avoiding or catching the object is essential to the discrimination. Second, because the action is part of the discrimination, the discrimination is not punctate in time, but happens over the whole trial

 

I prefer this second characterization to the first, since I’m never sure what “separate parts” means in a given context as there are certainly interfaces here. The systems can only communicate via a small subset of the factors that they experience individually. This is not to say that the systems aren’t tightly coupled. I’m saying something a bit simpler and more a prioiri (but hopefully not Hegelian 🙂 ), e.g., the leaf may be green because of quantum effects experienced internally to the cells, but we’re seeing “green” modified by our eyes, location, attention etc. rather than directly experiencing the quantum effects— this is similar to some of the points made in Object Oriented Ontology.

Again, this only happens, and only makes sense, when considering a being situated in an environment trying to perform an action

Barwise and Perry argued that we can’t understand meaning or cognition without taking into account that thinkers are spatially located (i.e., situated) and so have only incomplete, locally available information at their disposal. Every thinker and speaker is someone, who is somewhere, and who is aware of only certain things

 

He proposes three principles to guide his work

• Principle 1: Perception is direct
• Principle 2: Perception is for action
• Principle 3: Perception is of affordances

Affordances change depending upon the animal doing the perceiving.

Thus, the ecological laws relating to things in the niche of mice do not necessarily hold in outer space, or even in the niches of mackerel or fruit flies. So, instead of taking laws to be universal relationships between properties as the “establishment/extensional analysis” does, Turvey et al. say that properties-in-environments specify, or uniquely correspond to, other properties-in-environments. The most important ecological laws on the Turvey-Shaw-Mace view are those relating

Although not explicitly stated, he does seem to implicitly allow that the affordances themselves might change depending on the task (goal) at hand.

Note that this example makes clear that on my view, but not TurveyShaw-Mace, constraints that connect situations are not limited to lawlike connections but can also be cultural or conventional in nature; the fact that some situation token contains information about some other token does not necessarily entail that the second situation token is factualAffordances, I argued, are relations between particular aspects of animals and particular aspects of situations

Chemero goes along with Gibson’s characterization of “information for affordances” as being different from “Shannon information”. I don’t think this is correct — I’m not an information theory person, but I always thought that Shannon information included the expectations of the receiver, and so I see no a priori reason why affordances wouldn’t just be another codebook.

He closes with a strong argument about being reductionist, but not too reductionist, since the system that we’re studying is a animal functioning in the environment.

Work on object exploration may have suffered because experimenters were not as concerned about the nature of the objects they used, despite the fact that those objects can have profound effects on exploratory behavior…. Our literature review and experiment indicate that neuroscientists who use object exploration may need to reflect on and empirically validate the process of object selection, and realize that behavior and affordances matter. “So ruthless reductionism is an inappropriate position for neuroscientists to hold because ruthless reductionism can lead to questionable experimental methodology. Ruthless reductionism, this means, is also inappropriate for philosophers of science.”

The implications are significant: the animal perceives what is useful to it (which is the very nature of affordances). If the experiment/experimenter ignores this, the results of the experiment is of questionable value.

His final words are

 I offer the following threestep program for radical embodied cognitive scientists: First, admit that brains are important; second, embrace dynamical systems modeling as the brain-friendly, but still noninternalist, means of explaining the activity of brain-body-environment; third, do not, under any circumstances, make arguments based on Twin Earth, inverted qualia, or Martian psychology

 

I don’t find anything in the book that suggests that symbolic processing is impossible (there’s a detailed discussion of structures such as “decoupled oscillators” which I think could easily support it). Chemero’s suggestion is that we should be very hesitant to consider it necessary for any particular bit of cognitive functionality.