Saturday 3 December 2016

Is the Ecological Approach Radical Enough?

Radical enactivists do not just want to get representations out of their explanations for our mental life. They also want to get rid of the notion of content. Hutto & Myin (2013) is the strong version of this claim. 

Mental states or processes have content if there are specified conditions of satisfaction and if it can be evaluated for things like truth (i.e. does the thing conveying content doing so accurately or not?). Part of the concern is that ecological psychology is committed to content and thus can’t play with the other radical theories. The evidence is that we talk about things like ‘information about affordances’; that ‘about’ implies content. 

van Dijk, Withagen & Bongers (2015) took a swing at defending a content-less ecological psychology. I admire the attempt to get the radical camps on the same page, but at the end of the day I think a) the defence is grounded on the wrong notion of affordances (as relations, instead of dispositions) which means b) I don’t think it works but that c) I don’t think I care. I am as yet unfazed by critiques of content, although I’m happy to hear more on this; frankly Hutto & Myin’s book is a real struggle to read and any clarity people can add, I’ll take. 

The Problem
Content is bad because Hutto & Myin can’t find a naturalised theory of information that would allow it. This violates the ‘Muggle Constraint’, i.e. ‘no spooky stuff’; existing theories need too many tricks to allow content. Even reliable covariation doesn’t get you content. This means any account that includes an appeal to content, to meaning, doesn’t pass their test.

Their reading of ecological information is that it implies content and therefore won’t work. Gibsonians talk about information being created by the environment and specifying that environment, regardless of whether it is ever used by an organism. This implies that the information ‘means’ the environmental property; it has content independent of use. 

The Ecological Defence
van  Dijk et al first note that Gibson’s use of information was changing as he headed towards the 1979 book, moving toward ‘information for’ rather than ‘information about. The first defence is simply that Gibson was getting there as he worked through the problems, and we’ve just gotten lazy in the years since. This is not entirely wrong about the field more generally, although I’m going to suggest this isn’t quite the problem here.

The main defence is this:
…Gibson took information to be for (the perception of) animal-environment relations, rather than environmental properties, which dismisses the possibility of grounding information in environmental correspondence only.
In other words, they state as a fact that Gibson mean affordances to be relations, not properties, and given this information isn’t about them, it’s for the perception of them.

My Analysis
Affordances are not relations. They are dispositional properties. This account holds together very well and enables all the things we want affordances to do (Scarantino, 1992). Most importantly, if they are properties of the environment, then they can be things like actual surfaces, things that can actually interact with energy media and structure that energy into an array. There is as yet no account of how relational affordances can do this – there is nothing to point to that light can bounce off of, for instance. So frankly, on these grounds, the whole defence falls apart. 

van Dijk et al are right about one thing. One implication of the dispositional analysis is that it explicitly licenses content talk. Turvey and Scarantino both do this, on purpose, because they properly can. If an affordance is a dispositional property, and it’s interaction with energy media creates a specifying variable, then that variable can sensibly be talked about as being “information about the affordance”. Because of specification, that variable can readily mean the affordance to an organism.

So, does ecological psychology run into Hutto & Myin’s unsolvable Hard Problem of Content? Yes, but we can solve it. H&M make an error when they conflate variables that specify and variables that reliably co-vary. These are different in kind, not degree. Specification is not merely ‘100% reliable’ and it can support meaning talk. Now, of course, the organism has to learn that that variable means that dynamical property. Developmentally, we start out sensitive to kinematic structure in arrays and require time to become sensitive instead to the underlying dynamics (Wicklegren & Bingham, 2001). But the goal of perceptual learning is sensitivity to the dynamics via interaction with the kinematics; to succeed, we need to know about the environment. 

Conventional Based Information Use May Be Contentless
One thing that occurs to me is that Sabrina’s distinction of law based vs convention based information use kicks in here. Law-based use typically requires specification and thus supports meaning talk. Convention based use does not require specification; mere reliable covariation is sufficient. It’s possible that this part of our framework is contentless in the way enactivists want, namely there is no intentionality until the information is used. So that could be a useful point of contact between the ecological and enactivist approaches. But law-based use of specifying information about dispositional properties of the environment happily supports content talk and I think this is a feature, and not a bug. 

‘Content’ and ‘meaning’ have become words to run away from, thanks to Hutto & Myin. It seems to be almost the received wisdom that they are right and that the rest of the radical world needs to catch up to their insight. van Dijk et al work to catch up by claiming affordances are relations and information is for the perception of relations, no aboutness allowed, and argue we really can all get along. 

I just don’t buy it right now. The relational account of affordances has huge problems (namely, how do I perceive them?) and the H&M argument that no kind of covariation supports content conflates ‘relaible’ and ‘specifying’, treats them as the same kind of thing and misses the work that specification can do (as detailed in Turvey, Shaw, Reed and Mace, 1981). So I’m not currently spooked by the Hard Problem of Content and the only ecological defence doesn’t work anyway. 

Affordances as dispositions create specifying information and this account both licenses and supports content talk. It also works better as a theory of perception and action, supporting far more solid empirical research than the relational/enactivist counterparts. I suggest it might be time that enactivists start trying to catch up to us. 


  1. I guess I take a more radical view than you do.

    I don't have a problem with meaning. That exists, but I don't appeal to it. But I do think content is a problem, particularly if it comes with truth conditionals. I don't see that truth could be involved at all in perception. We judge truth by a community standard. But an individual organism has no community until after it has developed some sort of perception.

    I think you want information to be objective, while I see perceptual information as unavoidably subjective.

  2. Hi Andrew and Sabrina,

    I came here to find your replies to comments on the "Ecological Representations" paper, but I then stumbled on this and wanted to mention something:

    I feel like the debate about content is a debate from the philosophy of language that's been imported to cognitive science, whereas the debate about affordances is a debate from perceptual psychology (and later, movement science) that's been imported to cognitive science. In Hutto and Myin, we have a couple of professional philosophers who're basically (on my view) trying to privilege methods and ideas from movement science and relational psychology over methods and ideas from the philosophy of language, by means of picking out incompatibilities between ideas grounded in the lived experience of reading and writing (like content, representation, etc.) and ideas grounded in measuring dynamics (like information, action control, etc.).

    Also, on "reliable" and "specifying" - in the way that e.g. Turvey uses it (1992 p 181, for instance), "information about an affordance" is exactly synonymous with "information specifying an affordance", i.e., "a pattern in an array that is unique and specific to a disposition". Three things, then: first, if specification supports meaning-talk, it only does so redundantly. That is, if "Y bears content about X" will always and invariably mean "Y specifies that X caused Y", it seems like a synonym with no new function. Second, IF that's true, then "meaning" and "content" in this context have lost *all* of their original uses and applications from both lay use and the philosophy of language, such as a word "meaning" a concept, experience, emotion, etc. Third, a clarifying question about something I don't understand, because maybe someone else also doesn't understand: aside from the uniqueness clause and the definition of dispositions as complementary properties with other dispositions, what makes specification different from 100% covariation?

    And finally, I want to say that this is really just such an excellent blog, and that I'm really grateful to you both for maintaining it and putting in so much work to keep the community active.