Thursday, October 9, 2014

A non-binding resolution to the binding problem?

Contributed jointly by Dragan Rangelov and myself

The binding problem is a specific example of a more general problem in brain studies, namely that of integration, that is to say of how the many, specialized, areas of the brain interact to provide the integration that is evident in our perceptions, thoughts and actions.

Binding has come to refer more to this problem within the confines of the visual brain. Here the binding problem becomes the problem of how the several, parallel, processing systems in the brain interact to give us our unitary perception of the visual world, in which different attributes such as form, colour and motion are seen in precise spatial and temporal registration.

The initial mistake is to suppose that we do see these attributes at precisely the same time. In fact, psychophysicalexperiments show that this is not true and that we see and become aware of some visual attributes such as colour before we see and become aware of others, such as motion.

This raises a question which has so far remained un-addressed, namely of whether there is some central station in the brain that “waits” for all the processing systems to complete their tasks before “binding” the results of their operations. There clearly is no such system because, over very brief time windows, we bind the colour that we see at time x to the motion that we had perceived 80 ms before. We therefore mis-bind in terms of the objective reality.

We discussed this issue some two years ago while at a meeting and thought that we should conduct some more experiments on this problem. Our approach was as follows: we presented subjects with lines of different orientation that could be in a number of colours. If colour is bound to orientation at perceptual or pre-perceptual stages, then the accuracy of reporting one attribute, say colour, should co-vary with the accuracy of reporting the other attribute (orientation), when the two are presented to subjects over very brief time windows.

If, however, the two attributes are not bound at the pre-perceptual or perceptual stage, then the accuracy of reporting one attribute (colour) should vary independently from the accuracy of reporting the other (orientation).

Our results, just published, showed that the accuracy of reporting the two attributes is independent, with the accuracy of reporting colour being always greater than the accuracy of reporting the orientation, probably reflecting the fact that colour is perceived before the orientation of lines by about 40 ms.

This suggests that these two attributes, at least, are not bound at either pre-perceptual or perceptual stages.

This result leads us to conclude that binding does not occur by physiological interaction between cells in the visual areas, but rather occurs at pos-perceptual stages, perhaps through the intervention of memory. We only experience attributes as being bound even though they are not bound physiologically, and only because they occur within the same, very brief memory time window.

Our results may provide, we think, an interesting resolution to the  binding problem, namely that there is no such problem to resolve at the perceptual level.

If binding occurs post-perceptually, then the search for how binding occurs shifts to a different arena.

Time will tell whether we are correct in our interpretation. 

We may of course be wrong, but we hope that our new view provides the ground for interesting new experiments and debates.


Richard Crowder said...

As a hobbyist researcher looking into hierarchical temporal memory, this shared information is valuable for consolidating my own thoughts in this area. Thank you for sharing, Richard.

S.Z. said...

Thanks for the comment. SZ