Friday, 16 December 2016

Affordances are Not Relations, Part 1: Chemero (2009)

Affordances are on my mind right now as I develop the throwing research programme, and a major commitment of that work is that affordances are (dispositional) properties of the environment picked out by organisms in the context of tasks. This commitment has become important enough that it's time to get into developing specific arguments against the various 'affordances are relations' papers that are out there. I am working towards a paper summarising my objections to the relations account that also strongly advocates for the properties account on the grounds it enables a lot more science. This will be an occasional series of posts as I read and draft my arguments; as always, feedback welcome.

In this first post, I want to draft a response to 'Affordances 2.0', from Chemero's (2009) book Radical Embodied Cognitive Science. I previously blogged this chapter in two parts here and hereThere are two basic features of Affordances 2.0. First, affordances are relations between features (not properties) of the world and features of the organism. Second, these relations are dynamic; they change over developmental (and presumably evolutionary) time. 

Critique 1: Affordances 2.0 Do Not Exist Without an Organism
Chemero claims that affordances are the kind of relations that do not require anyone actually observing the relevant features in order to exist. They simply require that there be something out there that could do the observing (this analysis aligns very closely with the Rietveld and Kiverstein 'field of affordances' framing too; here and here). This is the idea that affordances are 'lovely'. 
A feature of some situation might exist just as it is even if there are no animals. There will be affordances in which that feature takes part as long as some animal exists with the appropriate ability. This is the case even if that animal is nowhere in the vicinity of the situation that affords something to it.
Affordances do not disappear when there is no local animal to perceive and take advantage of them. They are perfectly real entities that can be objectively studied and are in no way figments of the imagination of the animal that perceives them. So radical embodied cognitive science is not a form of idealism. But affordances do depend on the existence of some animal that could perceive them, if the right conditions were met.
pg 149-150
I do not think that Chemero gets to claim this. Relations, while real, do not persist when the relation is not implemented. While you can say the female hippo is lovely so long as there is at least one male hippo in the world, this is actually shorthand for 'if the male hippo was here, it would find the female hippo lovely'. No male hippo, no loveliness is manifested. In fact, relations that are simply possible connections between features sounds an awful lot like dispositions and their complementary properties.

So my critique here is that Chemero claims relational affordances persist but does so on the basis of a description of dispositions and not relations. 

Critique 2: Affordances 2.0 Are Not Sufficient
The organism part of the affordance relation is the current relevant abilities of the organism. If I perceive affordances in terms of such features, then I can only perceive what I am currently able to do. How can I change my behaviour? Certainly not via the perception of affordances, because right now these are only relations that are currently available to me. So in order to learn to be able to enter into a new relation with the environment (learn a new skill) I need that learning to be supported by something other than affordances. On this account, I also cannot perceive the affordances for other people. So if I cannot reach a high shelf, what enables me to go find a taller person? 

Affordances are 'for good or for ill'; Chemero doesn't get into it but perhaps an affordance relation you can perceive is 'that object cannot be reached by me'. Personally I'd worry that this is a much more open ended relation than 'it can be reached by me'; I'll leave working this out to someone who thinks it would work.

The key critique here is that if affordances are relations, they cannot support things such as the learning of affordances, and this means the ecological ontology (our theory of what the world is like) needs more as-yet undescribed elements. Chemero now owes us a story about those. 

Critique 3: Relations Do not Create Information, Properties Do
Chemero argues (using the literature on moveability perceived via dynamic touch and in terms of the inertia tensor) that relational affordances can be perceived. I'm happy that inertia tensors and moveability are things; the question I want to address here is, is the affordance 'moveability' a relation, and if so, can that relation create information? The answer to both is 'no'.

Any object resists being moved in each of the three spatial dimensions, and the inertia tensor is the mathematical description of this resistance. As Chemero notes, this (dynamical) property is the source of the information used in perceiving all kinds of things. It's the thing that interacts with haptic sensors and produces information. Work has shown that it underpins perception of the affordance moveability. 

The affordance 'moveability' is therefore defined by a dynamical property of an object, and not by a relation between the object and a person. Moveability is not a relational affordance.

Now, 'moveable by me' is the subset of all inertia tensors that I can actually move and this subset is defined by a relation between my abilities and the tensor. My abilities pick out what that subset is. But that relation is not what is creating information; the 3D inertial characteristics of the object as described by the tensor are, and that information specifies either an object I can move or one I cannot move. If it was only the subset defined by the relation that was creating information, then we'd need a story about why just that set, why just that set for me, why a different set for someone else, and so on. 

Relations do not create perceptual information; properties do. Then, and only then, does the relational act of perceiving and acting on an object identify the nature of that affordance in relation to abilities.

Relational Affordances 2.0 do not work. The relation is not what defines an affordance such as moveability and the relation that defines moveable by me is not what creates the relevant perceptual information. This is lucky, because if the affordance was the relation it would not be able to support learning over either developmental or evolutionary timescales; how do you learn a new ability if you cannot perceive the affordance that supports it until you can perform the ability?


  1. I'm worried that a lot of these discussions seem to become linguistic rather than factual (definitional rather than scientific?).

    My bias is towards the relational approach, for various reasons. That said, I think those who like the dispositional approach are shaking their stick at real phenomena. People have dispositions: True. The environment has dispositions: True. When complementary dispositions smack into each other, things happen: True.

    Do you really think Chemero is shaking his stick at something that doesn't exist? Or do you simply think the thing he is shaking his stick at doesn't work as a definition of "affordances"?


    As for the critiques...

    I agree with Critique 1, and I think Tony is wrong to make a strong (metaphysical?) claim that affordances persist even when part of the relation is absent. I think you are correct that it is like the "loveliness" of the female hippo. All that means is that people often talk about affordances slightly wrong. They should say something awkward like "To such-and-such animal, which doesn't happen to be here at the moment, X would be afforded, but it technically isn't afforded to said type of animal at the moment, because said animal isn't here." Instead of such an awkward locution, those in our field commonly statements like "Those stairs afford running-up for a dog", even when there isn't a dog around, and everyone knows what is meant.

    Critique 2 seems odd. It seems to derive from A) an unnecessarily strong read of the claim that all perception is of affordances, and B) an unnecessarily strong claim that one can only perceive what is afforded to one's self. However, even if we keep those unnecessarily strong claims, there are still ways to get where you want to go! For example, we could spell out a system in which: I could potentially perceive - as a social affordance - that a high object is retrievable through certain actions directed at a taller social partner. I only perceive an affordance, and it is something afforded to me.

    Critique 3 is also odd. If relations are real, and real things create patterns in ambient energy under the right circumstances, and some of those patterns are capable of specification, then relations can be specified. I agree that the inertial tensor isn't the most intuitive example for that. However, to the extent that what happens when I move a rod creates information about both the rod and about myself, there is no a priori reason to declare it impossible that through such a process I would perceive a relation between the rod and myself.


    1. Do you really think Chemero is shaking his stick at something that doesn't exist? Or do you simply think the thing he is shaking his stick at doesn't work as a definition of "affordances"?

      The latter.

    2. Critique 2 seems odd. It seems to derive from A) an unnecessarily strong read of the claim that all perception is of affordances, and B) an unnecessarily strong claim that one can only perceive what is afforded to one's self.
      I don't think we only perceive affordances; but Chemero's account is, I think, committed to the idea that relations exist between current abilities and the environment. This falls out of his move to make them dynamic. If that's the case, then when we perceive affordances we only perceive what we can currently effect and thus anything that entails perceiving something else needs an explanation separate from Affordances 2.0.

      From my view, because affordances are properties of the environment, they can serve as stable targets of perceptual learning. You can learn an affordance just by learning to discriminate the stable information created by the stable property without any other ontological pieces.

      Relations are real; but few relations of the kind cited by the 'affordances are relations' people are made of stuff that light can bounce off (for example). Just because something is real does not mean it creates information; it HAS to interact with energy so as to structure than energy into an array.