It is important to separate the question of binding from the problem of consciousness. Undoubtedly, there are some close connections between the two: my conscious experience is of a bound unity. But my unconscious experiences -- subliminal impressions, masked primings, etc. -- might be bound too for all I know. Hence, some of the recent commentators speak too loosely when they talk of 40 Hz oscillations solving some problem of conscious perception.
Just so that I am clear about what I am talking about when I talk about perceptual binding: Binding refers to the joining together of the individually processed features at the "psychological" level. We can see clear examples of binding in the work of Anne Treisman, where she demonstrates that our visual processing system under duress will conjoin the wrong features with each other. So, for instance, if we are flashed a blue square and an orange circle on the periphery of our visual field when we are attending to another task, we might report seeing an orange square and a blue circle. Sometimes binding refers to the joining together of the component pieces of some features -- somehow grouping the neurons that all signal bits of the square shape into a single unit. This sort of binding operates at a lower level of analysis that psychological binding, for psychology (generally speaking) takes a unified feature as the basic unit in perceptual theories.
The question is how to explain psychological binding, though sometimes the question of how to explain feature-binding is also asked using the same language.
We also need to clarify the distinctions between top-down and bottom-up approaches, a hierarchical processing stream, and the different levels of organization and analysis in the brain. All three have been bouncing around in the recent posts as though they are all more or less the same thing.
Here is how I look at it: A top-down approach starts with the largest unit of analysis (cognition, say) and then breaks that unit down into smaller pieces (langauge processing, visual perception, etc.) and then breaks those units down into still smaller pieces, and so on, until you have little tiny easily understood bits such that when you put everything back together again, you will have build a cognitive machine, or whatever. A bottom-up approach starts at the other end, at the level of neurons or maybe ion channels, and then figures out how they are connected to and influence one another so that we can get ensembles and then how the ensembles interact to get lobes, or whatever. Ideally (according to me), each investigative approach is carried out in tandem and each works to constrain the hypotheses of the other.
A hierarchical processing stream highlights feedforward aspects of information processing. Most (but certainly not all) of the hypotheses involving perception that I know of today are hierarchical. This is not to say that hierarchical stories completely ignore the feedback connections among processing areas, but they do minimize their influence. Non-hierarchical theories are often nonlinear dynamical ones which focus on stable oscillatory or chaotic patterns found across the areas.
Levels of organization and analysis refer to the different ways one can look at and theorize about the brain. Ionic transport across cell membranes would count as a lower level of organization. IPSPs, EPSPs, action potentials, and neuronal equilbrium are found at a higher level of analysis. We then find cell ensembles, areas, networks, lobes, and so on, up to the entire CNS considered as one unit. Different theories operate at different levels. The Nernst equation focuses on the fairly low level of neuronal equilibrium; Walter Freeman's chaotic patterns over rabbit olfactory bulb are focuses at a much higher level.
In my way of looking at the world, no single level (or set of levels) is priviledged in explanation. Different phenomena require different levels of explanation. The trick in cognitive science and in brain studies in particular is to get the level of explanation right, for some things that we take as a weird and peculiar phenomena are only artifacts of looking at the wrong level for an explanation.
This is where I suggested that the phenomena of psychological binding might fall. If we focus on a lower level hierarchical theory of visual processing, then binding looks completely mysterious. The 40 Hz oscillations of Singer and Gray (but not so much of Llinas; I think those are at a different level of organization that what S & G discuss) have been suggested as one way to accomplish binding at this lower level of analysis. Contrary to what some have thought, though, this hypothesis does not require some higher level homunculus to discover which groups of neurons are firing in tandem. Instead, the coupled firing just is what it is to be bound together. That is, this hypothesis resides only at a single level of analysis.
What I claimed in my JCS article was that the evidence linking those sorts of 40 Hz oscillations to psychological binding was extremely weak (despite Pat Hayes1s claims to the contrary. This is where wetware experiments have it over computer simulations. Maybe the only way we can get computers to reason is in terms of lower level oscillations. Maybe that is a failure of imagination on our parts and should not be generalized to the brain.) In fact, I believe it is nonexistent. What Gray and Singer have found might be connected to featural binding. As described in the recent posts, they did find the neurons oscillated in unison if they were responding to a single bar, as opposed to two bars. However, they also found no evidence that neurons oscillated in unison across different features of the same object. That is, if the cat were shown a blue square, they did not find any color-responsive neurons oscillating with the shape-responsive neurons, which is what you would need to find if you think their 40 Hz oscillations solve the problem of perceptual binding. (To emphasize, too, the distinction between consciousness and binding, it is important to keep in mind that these studies were done on anesthetized cats whose eyes were pinned open and directed toward a screen. Hardly the ideal preparation for studying conscious perception!)
My suggestion was that if we look at a higher level of analysis, the level of firing patterns across areas of cortex, then the problem of perceptual binding might disappear. It is now well-documented that there are oscillatory cycles above the areas of cortex responsible for perception that correspond to our "interpretations" of incoming stimuli. (That is, we see consistent patterns for stimuli that are meaningful to the subjects; no consistency or little consistency for stimuli that are novel, incidental to the task, etc.) But at this higher level, we don1t find a hierachical processing stream because there are not separate perceptual information flows that somehow need to some together. The problem then simply doesn1t exist at the higher level of analysis.
Here is an analogous way of looking at what I am trying to say. We are all familiar with the rolling pattern of boiling water. But if we trace the activity of a single water molecule as it breaks into its component pieces, we don't find a rolling pattern; we find random motions. If we trace the individual activity of many water molecules, we still only find lots of individual random patterns. So where does the rolling pattern come from? Answer: tracing the motion of individual atoms and molecules is looking at the wrong level of analysis for rolling. One needs to look at the cumulative effects of all the water at the same time. The only way to do that is to look at the water at a higher level of description and describe the motion at that level. The problem of where rolling patterns comes from disappears at the higher level, for it is just the pattern that emerges when water boils. Maybe binding works in the same way. Looking at the activity of single cells or of several cells taken at once is looking at the wrong level of description. If we look at the cumulative effects of all these cells taken at once, then we see a pattern emerging that just is our perception of some stimulus. No binding need take place.
This is not to say that this higher level patterns are not "causal," as Jim Newman suggests ["the brain ... perniciously refuses to function in a linear, causal fashion"], but it is to say that the effects we see are nonlinear. It is causal; it is just very complicated and probably beyond our capacity to track at any lower level. Nor is it to say that as a result of taking this higher level view, we don't need feature detectors or any of the other cells hypothesized in the lower level hierarchical theories, as Keith Sutherland did. The lower level is still there and still useful for explaining many phenomena, feature detection included. There is nothing wrong, I don't think, with being a pluralist about levels. I would explain potassium influx and eflux at one level of analysis and aphasia and agraphia at another. Similarly, I might explain feautre detection at one level and then the unified nature of our perceptions at another.
Finally, I think that my perspective does away with the implicit and sometimes explicit distinction between software and hardware. There aren1t simply two levels that interact or not, depending on how you think about computers. There are lots of levels, each of which contains items that can be described as a function or process (the "software") and as a structure (the "hardware"). As Hayes argues, thinking in terms of a bipartate theory just doesn't work for the brain.
Valerie Gray Hardcastle email@example.com