David Ganzhorn
ganzhorn@cs.rochester.edu
Paper 7
CS240
10/31/03
Jeremy Wolf and Kyle Cave survey the evidence for a binding problem, as well as propose a partial theoretical understanding of it, in their article (Cave and Wolf). One example of the existence of binding is found in a visual search task. In a certain image, there are horizontal green bars and one vertical green bar, all on the left. On the right, there are horizontal green bars, vertical red bars, and one vertical red bar. Searching for the vertical green bar on the left occurs just as fast regardless of how many distractor horizontal bars there are, but the task on the right takes more time proportionate to the amount of distractor bars; this is because the brain can search in parallel for features, but it must pay attention serially to spatial locations to bind features together, in order to identify combinations of features. So, the visual system of the brain must only find where the single vertical feature is to find the target on the left, whereas the visual syste! m must find the combination of vertical and red to find the target on the right. Finding combinations is believed to be a serial process because it takes people much longer (proportional to the input) to find combinations of features as opposed to single features (which take about as long regardless of input). I believe a person can only pay attention in serial because of the complexity involved in combining features; if there were numerous types and quantities of features in a given visual input, then there would be a huge amount of combinations of features to calculate throughout the whole visual field. Because it would take an incredibly long time to process the entire visual field for combinations, it makes sense to only process the part that is being attended to. The visual search task I gave as an example has one complication; one will only pay attention to the areas where the component features of the desired combination exist. Thus, the parallel processing of fe! atures can greatly narrow down the places were serial attenti! on must be applied, so long as the entire visual field is not equally convoluted with the target features.
The answer to whether the brain behaves in a parallel or serial manner appears to be, "both", but I feel that the answer can be elaborated on. If one makes the analogy of comparing a brain to a computer, then the question "what is the brain^Òs operating system?" becomes a reasonable wondering. By operating system I am not referring to the underlying processing components of the system, only the overall, clearly observable behavior of the system. Although identifying where some blue is in the visual field can be done in parallel, I see this as a very limited and trivial example of the brain^Òs capacities. For any truly interesting and meaningful visual search task, such as trying to find someone^Òs face in a crowd, the visual systems of the brain must work serially to combine the features that have been detected. Thus, at least for visual search tasks, I would say the brain can only operate serially.
The obvious next question is whether any other parts of the brain operate serially. In fact, a recent study shows that even a visual decision task and an aural decision task must be done in serial (Hazeltine, Pashler, and Ruthruff). When required to do an action depending on visual input, and a different action depending on aural input, most subjects would do one action then quickly do the other. Rather than doing both simultaneously, the two had to be done in parallel, or the response had to be delayed until both had been processed.
Taken together, these studies indicate that much of the brain^Òs functioning must occur serially; complex visual searches must be done serially, and even combinations of tasks such as a visual decision task and an aural decision task must be done serially. Although it may seem that paying attention necessitates serial processing, it is possible that one only pays attention to the tasks that require serial processing. Either way, attention is deeply tied to serial processes.
It is quite significant for substantial tasks in the brain, such as visual searching, decision making, and possibly paying attention in general, to be serial processes. Much of what people do all day requires these two tasks; navigating about and interacting with a visual world and deciding what to do along the way. This has major implications for computation models of the brain; every time the overwhelmingly complex processes in the brain can be broken down into smaller components, especially serial components, it becomes clearer how a model of the brain can be made. When the interactions of specific functions of the brain become clear, then their place in models also becomes clear. However, knowing the place of a function within a model of the brain does not necessarily mean that the function is understood; even if the brain can only make decisions serially, and only do so a quite limited number of times per second, each decision can be extraordinarily complex. Althou! gh the brain may not be able to beat computers in quantity of decisions, it certainly is beating them in quality. However, time will only widen the gap in terms of quantity, and the future looks promising for improving the quality of decisions made my computational models of the brain.