Controlled processing

All behavior varies along a continuum from mindless and automatic to willful and purposeful (i.e., controlled) behaviors (Barsalou, 1992). Automatic processes are reflexive. If a car or a predator is bearing down on us, we leap out of the way before we have even had a chance to form the intention to do so. Automatic processes thus seem to depend on relatively straightforward, hardwired relationships between the brain's input and output systems. In neural terms, it seems that they depend on strong, well-established neural pathways just waiting to be triggered by the right sensory cue. That is, automatic processes are driven in a bottom-up fashion: they are determined largely by the nature of the sensory stimuli and by whatever reaction they are most strongly wired to.

These relatively simple, hardwired, cue-response mappings are useful in many situations. Because they can simply be fired off by the environment, they are quick and can be executed without taxing the limited capacity of our conscious thoughts. But a creature that is at the mercy of its immediate environment is not well equipped to deal with the ambiguity, and exploit the opportunity, of a complex and dynamic world. Sometimes, the course of action is not obvious because cues are ambiguous (i.e., they activate more than one possible internal representation), or multiple behaviors might be triggered and the one that is optimal for a given situation is at a disadvantage relative to better-established (more automatic) alternatives. In these situations, ambiguity needs to be resolved by our internal states and intentions, by knowledge of possible and desired future outcomes (goals), and by knowledge of what means have been successful at achieving them in the past. This information is used to activate the same basic sensory, memory, and motor processes that are engaged during automatic behaviors. Only now, they are not triggered by cues from the environment. They are orchestrated in a top-down fashion by our expectations about goals and means. So, this mode of behavior is controlled in the sense that we (our knowledge and intensions) are in charge, not the environment.

A tenet of modern studies of behavior is that this knowledge is obtained by learning mechanisms that detect and store associations between cues, internal states, and actions that predict goal attainment (reward) (Dickinson, 1980). Insofar as primates are capable of navigating situations that involve diverse relationships across a wide range of informational domains, it follows that a neural system for cognitive control must have access to information from many brain systems and the ability to encode the goal-relevant relationships between them.

As implied above, the fundamental operation of a cognitive control system is selection, choosing which basic sensory, memory, and motor processes are activated at a given moment. Selection is central because of our very limited capacity to engage in controlled behaviors. This is evident to anyone who has tried to talk on the phone and answer e-mail at the same time; we can only think about a limited number of things at a time. This is in contrast to our high capacity for automatic processes. Because they can be triggered by the environment, any number of them can be fired off at once, as long as they do not come into direct conflict with one another.

It makes sense that brain mechanisms for cognitive control would evolve with this capacity limitation. First, there is the trade-off between amount of information and depth of analysis; focusing processing on a narrow band of information relevant to a current goal allows a much more elaborate analysis of a situation and available options. This single-mindedness also allows us to stay on track; processing information not relevant to a current goal increases the chance for distraction. In fact, in many views of cognition, control and attention are virtual synonyms. It follows that a neural system for cognitive control must have the infrastructure and mechanisms for selecting goal-relevant, and suppressing goal-irrelevant, processes throughout the cerebral cortex.

Also, as previously implied, the cognitive control system must be highly plastic. Virtually all voluntary, goal-directed behaviors are acquired by experience, and given that humans and other primates are capable of rapid learning of new volitional behaviors, the cognitive control system must have an equal capacity for rapid learning.

The cognitive control system must also have a way to deal with the gaps in time that are inevitable with goal-directed behaviors. Information about predicted goals and means must be brought on-line before the behaviors are executed and maintained until the current task is completed. Also, because goal-directed behaviors typically are extended over time, we often must wait for additional information or for a more opportune time to execute an action. The ability to keep goal-relevant information on-line and available for processing is called working memory (Baddeley, 1986). It is less critical for automatic processes because they are immediate: a stimulus triggers an already established circuit.

To bring the role of the cognitive control system into further focus, consider a model proposed by Norman and Shallice (1986). It illustrates a widely accepted view of the architecture of cognition (Fig. 104.1). At the lower level, there are the automatic processes (sensory analysis, memories, details of motor acts, well-learned skills, etc.). The system is in automatic mode when processing flows through the lower level, from input to output, along established pathways without any hindrance or modification: sensory inputs trigger their analysis, which triggers long-term memories, actions, and so on.

Figure 104.1..  

Two levels of cognitive processes proposed by Norman and Shallice (1986), among others. Represented here is the notion that specialized functions that acquire information about goals and means select and coordinate among innate and well-established routines.

A brain that contained only this lower level of processing, however, would belong to an automaton. It would only be capable of reacting reflexively to stimuli that happen to flow into it. To allow the system to engage in proactive, goal-directed behavior, most theories of cognition posit the need for “executive” functions specialized for acquiring and representing goals and means. These functions come on-line when a course of action is uncertain or if a reflexive, automatic reaction associated with a given input would produce an undesirable outcome. They send top-down signals that control the flow of processing through the lower (main) level by favoring goal-relevant processes and suppressing irrelevant ones. Because of this role, Norman and Shallice refer to the executive functions as the Supervisory Attention System. Another example of this two-level processing hierarchy is Baddeley's Working Memory model, which posits an executive controller that selects the information that is held in lower-level short-term sketchpads (Baddeley, 1986).

To summarize, we have some expectations about the properties of a neural system for cognitive control. It should be a brain region where diverse information from other brain systems can converge and be synthesized into representations of goals and means. It should have a great capacity to learn from experience. It should have the ability to maintain that information on-line and a means for it to influence other brain systems. A rapidly accumulating body of evidence regarding the PFC suggests that it meets these requirements.