Web site logical path: [www.psy.gla.ac.uk] [~steve] [best] [this page]

Gibson, Affordance

There have been discussions on what J.J.Gibson means for visual perception, for the notion of affordance in HCI, and for the notion of affordance in e-learning. We all have enormous trouble integrating his ideas with our habitual ones, and the more we are trained (indoctrinated) in computer programming and reductionist "science" the more trouble we have. A number of authors have attempted to define and redefine terms like "affordance". Normally, I am strongly drawn to this kind of approach. But the trouble is that when we attempt to define a term, we fall back almost entirely unconsciously on existing patterns of thought, and do not even consider giving up anything we already believe. Gibson however constitutes a challenge to some of our patterns of thought.

I think the simplest short thing we can do is to list some basic lessons Gibson taught: whenever we write something inconsistent with these, then we are showing our prejudices in ways that are incompatible with the real world (as well as with Gibson's ideas, which aren't necessarily right in all ways). Here's my list of such Gibson inspired points, or facts.

• Optic array: Almost all work on human perception, past and present, analyses it by sense organ or modality. But our actual perception works by integration not only of multiple sense organ inputs but also of our motor system with our perception. If you press your eyeball, the world seems to move (because if you don't send the signal to move the eye yourself, then the integration goes wrong). At a larger scale, we walk round objects (but retain an integrated percept of a single object), or rotate small objects in our hand. We move our eyes through the optic array in order to sample different parts of it and so pick up more information, like a dog zig-zagging across a scent trail to determine from which direction it is coming from: perception and active sampling motion are a single system in real organisms (though not in most labs). It is unjustified philosophical prejudice, not science, that makes most of us analyse perception and action separately. In real life they are fundamentally integrated at the lowest levels.

• Optic flow. One of the things our perception does is recognise objects, but it does other things which do NOT depend on recognition. We control our locomotion (whether by foot, or in driving a car) based on the flow of visual texture fields in our field of vision, and our brains compute information on our motion without having any knowledge of the nature of the flecks of texture (e.g. dirt on the window, patterns on the ground, ...). The optic flow field gives us the value of the variable "time to collision" directly without our knowing (even unconsciously) the distance to the object we are approaching. This is why anyone can drive a car, and can do so with very bad depth perception. Our prejudices and education predispose us to make up a story about how we recognise the car in front as a car, compute the distance to it using "depth perception" (parallax, stereopsis, etc.), estimate our speeds, etc. None of this turns out to be a correct description either of the function actually performed by our perception or of the mechanism that delivers it.

• Intermediate states do not have a meaning in the way that final percepts do. Optic flow and time to collision are one case where the intermediate computational states are not those you might expect, and in particular do not correspond to things in the world: they are internal states with less meaning than a carry digit isolated in the middle of a piece of arithmetic. But this general point comes up much more pervasively in computation, even though it goes against our strong intellectual preferences to understand calculations as putting together parts each of which are meaningful in the same way as the whole is. For instance, the kind of story we like is that we recognise a duck because we recognise a duck-like head next to a duck-like body and wings, and recognise the whole by virtue of first recognising the parts. But even in a world of information, this is not generally true. Information, technically, is a reduction of the number of alternative possibilities in a situation. For example, consider that you are recognising capital letters, and then that the first bit of information you get is that the right hand side is a vertical line. This reduces the possible identities of the letter from 26 down to one of [H, J, M, N, U]. We do not have a (conscious) concept that represents that subset of letters, but the recognition system must have such an internal state and many others of that kind, each representing a different increase in partial information; but these have no correspondence in the world of letters or words.

• Direct perception. Gibson seemed to deny that there were any intermediate states or computation in perception. Of course there are if we want to analyse things that way (and I do): but these states have no meaning correlates themselves in the outside world. Furthermore, the real point here is that the phenomenon of perception could and should be described as a function from properties of the world to percepts with no reference to the internal computational intermediate states. This shows a clarity of thought, a realisation of what is primary in understanding perception. This is exactly parallel to the way that Darwin defined evolution with no concept or knowledge of genes, let alone of DNA. Genes are one implementation or mechanism for evolution. Evolution is the primary theory and does not require genes. Gibson formulated perception as a function from world to percept: not muddling it up with possible mechanisms. One of the worst, but most pervasive, intellectual vices is to assume that mechanism is primary. An early example of this fallacy were those who dismissed Newton's theory of gravitation for being "action at a distance" that didn't satisfy their irrational desire for a mechanism transmitting the force through intermediate entities.

• Affordance. Gibson's basic notion of affordance links action and perception the other way round: suggesting that (because beings in general and people in particular are concerned with acting in the world) we perceive as a direct attribute the actions we might perform with or on each object. This is a surprise only to those who presuppose (without evidence) that perception is independent of action, and is prior to it. It may also be a surprise to those who think perception is about measuring the absolute attributes of objects, because affordance is obviously not an object attribute, but a relationship between object and perceiver. But in fact (contrary to our habitual but faulty thought patterns), perception, like measuring, primarily and more easily detects relationships. Even when we are measuring an object carefully with a ruler, we are using our perception of the object's relationship with the ruler to derive an "absolute" property.

• The perception of novel objects and situations. There has been frequent discussion of whether Gibson denied or took insufficient account of the role of learning in perception. It is perhaps more important to ask whether most research on perception puts too great a weight on it. If we consider object recognition, then by definition a perceiver cannot do it until they have encountered the object, perhaps frequently. Yet an infant animal would presumably die immediately if learning had to precede successful perception and action. In the real world, beings encounter unknown objects all the time and a large part (in fact most?) of perception concerns what and how we perceive unknown objects. In designing artifacts and user interfaces, this is a chief concern: can a user operate a novel device first time, without training? This is the context in which most discussions of affordance in design are situated.

• Success without understanding. In design, the mindset which designers must escape is the assumption that successful action (such as operating a new device) depends on knowledge which is obtained from prior training, and that it is patronising to think that some users cannot master this. Contrary to this, a user centered approach sees learning as a cost and a burden that good design should avoid. Indeed, published experiments illustrate how in situations where information is rich, users still haven't learned what to do after many trials: they simply rediscover it on the spot each time. The aim of design in most cases should be to avoid the need for learning and indeed for thinking, and to aim for users to operate in "flow" from the start. While understanding is a pleasure, it is not the pleasure being sought when users try to open a door or operate a tool to do something else. The use of the concept of affordance in design is to allow users to perceive what to do without learning. When a door has no handle but a push plate, this is achieved. When an on/off switch has no label and a delay before any effect is visible, it is not, since the user cannot tell whether to press once, twice, or to hold down the switch.

• Human analogical machinery. Whenever someone sees anything, the thing is always to some extent novel: the day is different, the time, the phase of the moon, the lighting; if it is a person then their clothes are different, many of their cells have been replaced, etc. When it is a novel object, the differences are greater. But fundamental to perception in fact is deciding what this new stimulus is most like. When faced with a novel artifact or user interface, and (as is almost always the case today) the assumption that they will try it out without reading any instructions, then the big question is: what is this thing most similar to, which analogy will the user implicitly and unconsciously draw? Current computation has no impressive mechanisms for this, but human minds are fantastically good at it, yet it is little studied. Why will one situation (design) trigger an association, and with it, user behaviour? A few unpublished examples however suggest that these associations are not always and reliably predicted by our conscious categories. That is why almost the only thing we know about HCI is that we have to observe actual users on particular designs: we cannot adequately predict their response.

  One such example from a computer system long ago in a country far, far away is as follows. The command for listing the emails in the inbox was "ha" (Headers All); and the command for listing the files in the current directory was "ls". Not particularly frequent, yet persistently repeated, was the user error of sometimes typing "ls" for "ha" or vice versa. How could these two be confounded in the user's mind? The command names, the documentation, the nature of the objects (file names versus emails) were different in anyone's ontology, especially that of the technically experienced users exhibiting this error. This kind of error seems to reveal hidden, tacit categories (e.g. approximately "show me my bag of things") that have the power to determine user behaviour.

  ??

• Learning isn't fundamental to perception: dealing with novelty is.
  A more fundamental mechanism is analogy: recognising partial similarity
  Learning will be most important toi perception where it is about learning to do this more effectively
  The most important things for an organism (to learn) to perceive are not objects but actions the agent can achieve: i.e. affordances.

Comments, HCI, Education

• Ask why/how "affordance" seems relevant to education
• Suggest which points from above or elsewhere might actually illuminate education, and avoid conceptual errors relevant to our understanding of education.

Here goes.

• Good examples in HCI of affordance are things like the way Apple and others carefully designed floppy discs so that they could not be inserted the wrong way in disk drives (even though there are 8 ways a perfectly square disk could be inserted). On some doors in our public buildings today, there are no handles, only plates: so without thinking (and regardless of whether you use vision, touch, or just trial and error) everyone knows you can only push it open.

• A good example in education of affordance is something like the way a child or a student will seize the opportunity, when they have a mentor in a one to one situation, of asking endless questions: that is what access to an expert affords us. Students will also, if and when they have the opportunity with friends they are comfortable with on the same course, compare the marks and feedback they get. (Only by talking to students did I finally grasp that this gives them information they can't otherwise get.) Similarly they are deeply interested, when it is available, in seeing their ranking compared to their peers; how other students have answered an assessed task, and so on. Again, most teachers (and published theories of learning and teaching) are oblivious to why they do this. By observing affordances and their uptake by learners, things otherwise invisible to us as teachers begin to manifest. They can tell us about what the needs of our students are (we often are only aware of the goals we tell them to have); and about how they can satisfy them (when mostly we only think about what we provide, not what may be there all around them). This is not only a remedy for the teacher-centeredness of most educational thinking, but it is also (just like Gibson) like what faces biologists: believing in Darwin does not in itself give any clues to what needs an organism has nor to how they may often be supplied by its niche (often quite invisibly to us).

• A bad example of affordance in education is to say a lecture affords listening and/or learning. This is both trivial and wrong. It is trivial because it is what teachers think, not an insight into actual learning. It is wrong because actually a lecture only affords that to those learners who speak the same language, who can hear clearly, and who don't already know everything that is said. Affordance, as I (following Gibson) said above, is NOT a property of an object but a relationship between a person, an object, and a goal/task/function the person has.

• Interestingly this type of misconception is manifest in several other places in education. One is due to Carroll and Bloom and associated with mastery learning. Carroll pointed out (and supported by experiments) that even though most of us still today interpret the spread of marks in a class as telling us about an attribute of the learner, in fact there is again a 3-way relation between the learner, the material, and time: almost all learners can learn all the material but take different amounts of time to do so. This mis-interpretation by teachers leads to endless needless under-performance. Again, deep and shallow approaches to learning are NOT traits of the learner, but a relationship between that learner, that learning context, and a strategy or approach; learners select their approach depending on the context, they are not stuck with one approach for all things.

• Assuming that what is actually a 3-way relationship is merely a 2-way, object-attribute property is pervasive. Psychologists call a common form of this the "fundamental attribution error", where a person explains a problem or failure in terms of the properties of another person. The e-learning field is particularly prone to this, where talk of early adopters vs. resisters is exactly this (mis-)attribution: and a failure to grasp that the "contexts" of different people are different, so that the value or feasibility of adoption of the same equipment just is different for different people, and is not a constant property of the technology independent of people. Reversing this by being user-centered would attribute the failure to the designer not the end-user thus combatting the self-serving basis for most such attributions; but it would still not tackle the common reality that some things work for some people but not for others: i.e. it is not about a single artifact with absolute properties regardless of the user.

• Norman said he would replace all occurrences of "affordance" in his book by "perceived affordance". I however have another view. A design which novices and experts alike find immediately usable is one thing; designs which require learning investment from users are another. Both are important; and again, the important point for designers is that the desired feature is NOT a function (attribute) of the design/object alone, but of the 3-tuple of user, object, and goal/function. This too has an important similarity with learning designs where in general designers (i.e. teachers designing a course or a lesson) are even worse than software designers at thinking clearly about this, about the fact that a given design will only work for learners who already have certain skills and will fail for others. Lectures, for example, only afford learning for learners who not only can hear and understand the words, and who want to learn about the topic, but who have skills (e.g. note-taking) that are comparable to an experienced user's knowledge of how to get writing done with a word processor. All learning designs have pre-requisites of not only teacher skills but of learner skills: that is one big reason why learning designs seldom transfer to all other contexts.

• Thus taking into account the cost of learning the skill of using something for a goal i.e. of changing the human agent so that the object now does afford them an important function, is an issue that is important but often missed in both HCI and education and indeed other fields. Thus learning is frequently an important cofactor for affordances. Many published commentaries on affordances debate whether Gibson would agree with this. A key point here is that this "learning" is quite different in kind from the learning most of us mean when we discuss education. In education we almost always expect and test what the learner can articulate, write about, and teach to someone else. But with affordances, someone who can do it may be unaware that they have learned anything, unable to describe what it is they know, or to pass it on to someone else. There are huge differences in effectiveness between teachers, as measured by the attainments of their students. This is correlated with their experience (i.e. the amount of their tacit learning). But they generally can't pass it on to others, or training would homongenise this effect away. In HE, some untrained tutors exhibit great skill at facilitation, while others who have been "trained" are vastly less effective. These effects are large, important, and not understood; they can in one way reasonably be described as "learning" yet they share few characteristics with the "learning" that organised education purveys. Quibbling about what "learning" means may interest philosophers. For those interested in understanding education, we should recognise that some aspects of Gibson's mindset illuminate issues which are dangerously ignored or neglected in most educational literature, and so are worth paying attention to.

• To get value for understanding teaching and learning from Gibson and the notion of "affordance", we need to immerse ourselves in that way of looking at things: success without understanding, picking up information that is lying around us, using it without remembering it, being able (or not) to do things without any idea why we just succeeded or whether that success will generalise to other times and places. Both teaching and learning have a strong dependence on unidentified skills of this kind. The ideal learning designer would be able to reason about this, but faces the same difficult research programme as do biologists who have not yet discovered bat sonar or the use by some organisms of the way light is polarised to detect where north is or the electric fields picked up by fish; or planetologists working on the "Gaia" theory that there are hidden self-correction mechanisms maintaining earth. In this view, the operation of teaching and learning depends in part on mechanisms we cannot see and do not yet even suspect the existence of. Alerted by Gibson's way of thinking, I have noticed many clues over the years that seem to support this view.

Gibson's ideas thus give us 4 useful things for pushing forward our ideas about education:

• We use the one word "learning" to refer to two quite different things: what is expressed learning objectives and tested in exams, and the often unconsicous changes in behaviour and capability that experience sometimes causes. This is pernicious not just as a conceptual muddle, but because when focussed on one meaning we forget to check whether the simultaneous operation of the other is crucial to the case being studied.
• A reminder that treating 3-way relationships as simple 2-term object-attributes is pernicious but very common.
• The notion of affordance can remind us to look, not at what things are officially designed for (chairs for sitting, not rocks; sticks are for trees or perhaps poking things, but not for termite fishing), but at what people (learners) actually do with opportunities. This reveals to us something about the needs they actually feel, not the ones we intend them to have; and the methods they actually use to satisfy them, not the ones we so kindly designed for them.
• On the other hand, just as learners / people are more flexible than we expect in some ways, they are less flexible than we assume or wish in other ways. Especially in the short term, we have to allow for the costs of learning how to learn.

References

Biggs ch.

Carroll, J.B. (1989) "The Carroll model: A 25 year retrospective and prospective view" Educational Researcher vol.18 no.1 pp.26-31.

Dohn, N.B. (2009) "Affordances revisited: Articulating a Merleau-Pontian view" Computer supported learning vol.4 pp.151-170

Oliver,M. (2005) "The problem with affordance" E-learning vol.2 no.4 pp.402-413.

Web site logical path: [www.psy.gla.ac.uk] [~steve] [best] [this page]
[Top of this page]