Friday, 14 May 2010

Affordances, Part 3: Dispositions or relations - which is it?

Affordances are difficult entities to wrap your head around. We talk about them imprecisely, they seem like odd, ghostly entities that couldn't possibly exist, and even when we get precise about them, we end up with two different accounts of the kind of thing they are.

Gibson did not have a logical framework with which to express what he mean by affordances, although his verbal formulation is quite clear:
An affordance is neither an objective property nor a subjective property; or it is both if you like. An affordance cuts across the dichotomy of subjective-objective and helps us to understand its inadequacy. It is equally a fact of the environment and a fact of behaviour. It is both physical and psychical, yet neither. An affordance points both ways, to the environment and the observer.
(Gibson, 1979, p. 129)
There have been two attempts at formalising this definition, with two incompatible results: for Turvey, an affordance is a dispositional property of the environment, like solubility, which is complemented by an effectivity of an organism. For Stoffregan and Chemero, affordances are relations between the organism and environment. These are not the same and the difference matters: so which is it?

For a long time I was in the relations camp. Affordances 'point both ways', in Gibson's phrase; they are only defined with reference to an organism, and hence must be relational. This doesn't affect the essential realist stance of the ecological approach, because we are 'radical' - relations are just as real as properties, and can hence serve as objects of perception. Defining affordances as relations also allows you to solve a troubling aspect of Turvey's dispositions, namely the compulsory actualising or manifesting of the disposition in the presence of its complement. Clearly we do not attempt to realise all the behaviours afforded to us; behaviour is selective.

I was all about this when I read Chemero's recent book, but got schooled by my PhD advisor - no, he said, affordances are dispositional properties, not relations. Geoff is generally right about things like this, so while I couldn't see how it could work I kept mulling it over. Reading Heft (2001), Chapter 3, I had an intuition about how this could work, and I've now officially re-converted - affordances must be dispositional properties of the environment, so that they can persist when we aren't around and so we can perceive how to effect a currently impossible behaviour.

Affordances must persist when there is no-one around to effect them
Defining affordances as dispositions solves at least one important problem raised by defining them as relations: dispositions exist even when the complementary property is not present, and hence affordances, if they are dispositions, exist as potential in the environment before any organism is present to effect it. Stoffregan and Chemero both dodge this bullet in a deeply unsatisfying way - they simply declare that affordance relations persist, without any stated reason.

Why does this matter? Surely it makes no sense to say a surface affords locomotion if there is nothing around that needs to locomote. However, no one wants to say this, not even Chemero or Stoffregan; this is why they need the additional claim that the affordance relation persists. The reason for this is simple: the goal of ecological psychology is to create a naturalised theory of human perception and action. To naturalise a theory is to make it consistent with facts of biology and evolution, and this is a vital part of any theory about a biological organism. Modern cognitive psychology runs circles round itself whenever it tries to think about this problem, because the information processing metaphor is simply not up to generating a naturalised account of anything (as evidenced, I think, by the fact that biologists never ever invoke the metaphor). Ecological psychology aims to explain perception and action in a manner that is fully consistent with theories of biology; accounts of our behaviour therefore require an account of our niche, and hence affordances, which are what we perceive in order to control our behaviour, must be persistent features of our experience so this niche can be sensibly defined. Relational accounts include the requisite stable properties of the environment; they just don't call those bits the affordance. Chemero in particular thinks that only this relational account achieves the required naturalisation.

Affordances for behaviour we cannot currently effect
But we still have a problem - if we directly perceive affordances in order to control our behaviour, and affordances are the relation between our abilities and properties of the world, then we can only directly perceive opportunities for behaviour that we can do, right now. This has one glaring problem: we cannot directly perceive an opportunity for behaviour that we cannot currently effect, nor what is required to effect that behaviour. However, we clearly do this all the time: even a young child knows that if they cannot reach the cookie jar on the bench, their options include getting a stool they can climb that will bring the jar into reach, or finding an adult who can reach the jar.

To try and formalise this a little: we can perceive the possibility of actions we can't effect, and we can perceive the possibility of an action for another organism. Possibilities for action should be affordances, but if affordances are the relationship between our abilities and the environment, these other possibilities aren't available to us and hence aren't affordances. But these possibilities are available to us, and crucially so: if we could never know that 'we could reach that cookie jar if only I was slightly taller' we could never adapt our behaviour to achieve our goals because we wouldn't know where to begin (or where our adaptation was heading) - we would be stuck in the moment and our behaviour could not flexibly change over time.

If affordances are dispositional properties of the environment, then this problem vanishes. Recall that dispositions are formed by pairs of properties: the disposition and the complement (the affordance and the effectivity, in Turvey's terminology). Which is which depends only on attention: salt is disposed to dissolve in a suitable solvent (e.g. water), but water can equally be described as disposed to dissolve a suitable solute (e.g. an ionic salt). Equally, we can perceive the reachability of the cookie jar (if we are tall enough) and also the disposition of another, suitably tall human to reach the cookie jar. We can also perceive that if we could only increase our height slightly, that cookie jar will come into reach; that stool is disposed to exactly that. If you're unconvinced, stare at this excellent image that summarises this behaviour perfectly (stolen shamelessly from the website of ecological psychologist Michael J Richardson); watch especially as the stick figure seamlessly solves the problem of the step that is initially too high in multiple ways.


I'm still thinking about this, and trying to get the formalism right; the problem of compulsory effecting of a disposition also needs to be addressed. But the relational account simply does not stand up to scrutiny and does not allow the kinds of behaviour we must be able to account for. Our environmental niche contains affordances, some presented by nature and some by artificial alterations. Biological evolution is the ongoing process of evolving effectivities that match our niche; invention and artefact design is the ongoing cultural process of extending our range of effectivities to cope with novel affordances as we expand our range of niches to include cities, space, and all the crazy places people live on Earth. These affordances that define the niche must come prior to the organism who can then evolve into that niche. The net result is precisely as Gibson described - affordances point both ways, but only if they are dispositions. A given niche implies an organism with specific abilities as much as an organism with specific abilities implies a particular niche. No organism could ever behave functionally if its possibilities for behaviour swung in and out of existence as it moved around the world, nor could it ever evolve if those possibilities were limited to what it could currently achieve.


  1. Really nice summary. I was thinking about this issue this morning while reading about how bad people are at analogical transfer. Psychologists like to think that analogies are formed by recognising the deep structural similarity between problem domains. But, it occurred to me that analogies only make sense when both domains involves similar dispositional properities. For instance apples and baskets are analogous to cars and parking lots (at least, this is a common example from the literature). Noticing this is supposed to involve abstracting the CONTAIN relation and placing apples / cars and baskets / parking lots into correspondence by virture of their playing the same role in the CONTAIN relation. But, there's another way to think about this. Both apples and cars are objects - they afford being moved. Both baskets and parking lots have surfaces that can support objects and obvious boundaries so that things can be inside of them. These properties should be directly perceivable - they aren't abstract at all! And, noticing these properties is sufficient for solving the analogy. But, these aren't the types of properties that psychologists tend to be interested in. They only notice the differences between apples, cars, etc. When psychologists break this connection between affordances and problem domains (e.g., by using isometric word problems with completely arbitrary content) people fail to form analogies. So, structural similarity in the absence of similar dispositional properties isn't sufficient for making an analogy.

  2. In your summary, you mention "trying to get the formalism right", and I'm unsure what formalism you have in mind. In case it has to do with the multiple step actualization of an affordance and is still an open issue, here's a cut I made at formalizing that case:
    Define actualization as follows:

    An affordance A is actualizable by an effector E if and only if there is an irreducible sequence of affordances {A(i): i = 1, ... ,N} in the environment such that for i=1, ... ,n-1, A(i) + E -> A(i+1) and A(N) = A, where "+" signifies actualization and the arrow signifies an actualizable affordance A(i+1) consequent to actualization of A(i).

    Affordances as currently understood correspond to cases in which N=1, ie, in which A is immediately actualizable by E. The irreducibility is thrown in to try to avoid loops or infinite regress - which is what I'm guessing motivate the second and third of Bunge's three conditions, although I don't really understand what he's trying to accomplich with them.

    One thing I remain unclear about is the "compulsory effecting of a disposition". In the case of an inanimate salt "disposed" to disoilve in a solvent, I get it - it's required by a law of nature. But in the case of a human, there appears to be a potential problem.

    To proceed, I need briefly to wax philosophical. Being pretty much a strict determinist, I actually tend toward the belief that we don't really have options; in a given set of circumstances with several apparent options, we'll inevitably "choose" a specific one. The reason we think we are "free to choose" is simply that it is infeasible to predict which option that will be. Nevertheless, that apparent freedom is illusory. So in this case, there is no problem with the compulsory effecting - at least in principle - since we are in essentially the same position as the salts.

    But I also agree with John Searle that whatever we really believe in that arena, we necessarily act-as-if there is free will. And part of that acting-as-if is saying that we can choose among multiple affordances. In which case, there does seem to be a problem for compulsary effecting.

    So, two questions: Is determinism implicit in the assumption of compulsory effecting for the reason I describe? And if not, why is the assumption of compulsory effecting necessary since it seems to be a serious problem?

  3. One thing I remain unclear about is the "compulsory effecting of a disposition". In the case of an inanimate salt "disposed" to dissolve in a solvent, I get it - it's required by a law of nature. But in the case of a human, there appears to be a potential problem.
    It is odd. Essentially Turvey (1992) just mentions it and then doesn't really refer to it again. Chemero calls him out on it, and it's one of the things motivating the relations account.

    The issue is this: at any given moment, there are many objects in the world affording behaviour. I am not currently trying to effect all of them simultaneously; if they were dispositions, I would be. Therefore affordances aren't dispositions.

    My first pass response is to note what Turvey actually says:

    "3. Disposition + effectivity = actuality. When the entire system is present, the disposition is always effected/actualised."

    Here's my hunch: as I sit at my desk typing, I am not currently the kind of organism who is disposed to locomote. I'm sitting down; I'm attending to my typing; I don't have anywhere I want to go; etc. So the reason I'm not trying to effect the affordance of my floor for locomotion is that the 'entire system' is not currently present.

    (How does it get present? That touches on your multiple-step formalism, which I'll comment on later).

    My hunch also contains some determinism; once I am a walking device, I will walk in the presence of and under the control of locomotion affordances. We have flex in which devices we become, although that flex is constrained.

    What do I mean by device? Like most of my intellectual approaches to problems, it's grounded in something Geoff Bingham wrote. I am going to blog this in detail soon because it's becoming relevant; but Bingham (1988) on task specific devices is a paper that blew my mind in grad school and still does. This is what Chemero is missing and why the compulsory effecting causes him trouble.

  4. Your formalisation of the multiple step affordance thing is interesting and certainly looks about right. From a perception perspective, there are interesting questions; can you, for example, come to directly perceive the higher-order affordance A which requires the effecting of intermediate affordances before it can be effected?

    Heft's reading of Holt and it's connection to Gibson suggests a way in which you might be able to characterise expertise this way, which is good, because if it's not direct then it's mediated and that's a problem for eco-psych.

  5. I wonder if the compulsory actualization is an attempt to make all scenarios consistent with the soluble salt example. If so, why? Isn't perception restricted to certain animals? And don't those "make choices" (modulo determinism). Ie, why can't we treat humans as humans and not try to make their behaviors conform to those of inanimate salts?

    Of course, there is an inconsistency in my making that appeal and also claiming determinism. But one part of your post on Heft on Holt - which I had skimmed before but on rereading find very timely - suggests a way to make those consistent: the bit about the journalist on his way to the opera. As a systems engineer, I naturally think in terms of layered levels of complexity (AKA, "nesting") and have no problem with viewing things at one level from one POV and at another from a different POV. I think it's merely a matter of logic that if we start at the bottom with atoms, molecules, etc, behaving in accordance with natural laws, that quasi-determinism ("quasi" because of quantum effects) necessarily remains as we go up the complexity ladder. Nevertheless, it may sometimes be beneficial to "act-as-if" things aren't deterministic at the highest levels - in particular, to talk among ourselves making whatever overarching assumptions work out best, even the assumption that we can "choose" among options.

    And that's how I would interpret your approach as presented above - describe human behavior in a vocabulary that "works out" in context.

    So, along those lines I'd answer "can you ... come to directly perceive the higher-order affordance A which requires the effecting of intermediate affordances before it can be effected?" like this: if "you" is a salt, obviously not; if "you" is a normally functioning adult human, absolutely - at least sometimes. We do it all the time.

  6. I'm re-reading Chemero's book and I up to the chapter where he defends his affordance-as-relations theory. One reason motivating it is that dispositions are compulsorily effected once the conditions are met, so I'm interested in killing this off.

    I still think that once you have actually assembled yourself into the task specific device that is the complementary effectivity of the affordance, you will manifest the disposition. I also think this is why affordances have to be properties of the environment and not relations: you must be able to perceive it so as to assemble the right device so it needs to be there ahead of time. I need to work on this some more though.

    So I think that, assuming it is a technical fact of dispositions that they manifest right away, you can deal with it in humans by acknowledging that the effectivity is high-dimensional and therefore not always assembled. While I can swing a baseball bat, in a real sense I cannot just at this moment (when I only have a glass of wine to hand and no baseball game around).

  7. Forgot to finish that; so yes, I think what I just said is basically the same as saying I can act 'as if'. I'm not into taking stances, intentional or otherwise, because I think they are weak-ass ways to go about business; so I actually want to appeal to things like high dimensionality, task specificity, etc.

    I also think we have access to the higher order affordances after learning; that's the point of learning! No one's ever chased that, empirically, though, I don't think.

  8. I see possible convergence of various ideas that are floating about (waiting to be actualized!)

    Your idea of assembling yourself into a task-specific device that can actualize a perceived affordance seems another way of describing actualization of an affordance that is actualizable in steps but not immediately. Let's step through the whole process and see if we encounter any inconsistencies.

    You are sitting at your desk. In order to avoid needlessly complicating the scenario, I'll assume you are otherwise relatively static, perhaps just surveying your environment with eye movements. You are currently continuously effecting (at least) the immediately actualizable "sit-on-able" affordance offered by your chair and perceived not by sight but by touch. (I'm assuming that compulsory effecting is required only for immediately actualizable dispositions, which seems not only reasonable but necessary.)

    You now spot an empty wine glass across the room, which was - even before you perceived it - offering various affordances, none immediately actualizable. Because of your propensity to drink wine, you are irresistibly moved (determinism, you know) to action by the "pick-up-able" affordance offered by the glass. You recognize that it is actualizable not immediately but via a sequence of perceived and actualizable affordances [see note below]. The first (not really, but it will do) is the immediately actualizable affordance "stand-on-able" offered by the floor. Etc, etc., ending in actualizing the immediately actualizable affordance "grasp-able".

    In this scenario are many affordances, not all necessarily perceived (but being dispositional, they are persistent). Among those that are perceived, the perceiver is always actualizing at least one (has to be supported by something). The others are not immediately actualizable, and therefore don't demand compulsory effecting. Some may be actualizable in the multiple-steps sense.

    Seems to hang together. Anything missing?

    I just noticed that my formalism needs a tweak. A1 must be immediately actualizable and each successive A(i) must be as well once it's predecessor has been actualized.

  9. This seems about right to me. It relates nicely to this idea of nesting events and affordances within each other that's been cropping up; and the persistence of affordances as dispositional properties means you can perceive not only that you can get to the wine but all the steps you need to effect to get there.

    This is interesting, actually; I did some work on reaching to grasp and turn an object (a two step action). You can set it up so the initial grip is not maximally comfortable but making this means you end up in a comfortable posture. David Rosenbaum started thinking about this, and the example is turning a wine glass the right way up: you grasp the glass with an awkward posture to then make the final turn and place of the glass easy. Adults can do this easily, but mess up about 10% of the time with their off hand. Young kids are bad at it but get better as they get older; kids with DCD (a motor control problem) almost never do the awkward first movement, because they're too busy making sure they actually grasp the object.

    This gets talked about in terms of planning, and we were thinking about it as a bias which helped select the exact grip you made which had varying effects with skill level. But I was never quite happy with that first bit: the idea that the glass affords all these weird grasps and you need a bias to help rule them out seemed weird. Maybe a better idea is that skill enables you to effect the higher order 'reach-to-turn-and-grasp', which is a unit, rather than 'reach-to-grasp' and 'turn' as a sequence.