Given a fixed amount of computational power in designing an intelligent system, there is a necessary tradeoff between how many resources are devoted solely to the current task, and how many resources are devoted to monitoring for information that may be important but is not necessarily relevant to the current task. If more resources are dedicated to the current task, it may be accomplished more quickly – but at the same time, this setting may make it more difficult to reorient and switch to a different task. On the other hand, if more resources are dedicated to monitoring or reorienting, then the current task may not be completed as accurately or quickly.
This tradeoff is a form of the “stability-flexibility dilemma” – much recent research in cognitive neuroscience has focused on how the stability-flexibility dilemma is managed by the brain. One theory is that an inhibitory process “kicks in” to help switch to a new task by actively suppressing the representations relevant to the old task. Consistent with this account, some work has shown that people are slower to switch to task A if they have performed it more recently relative to having performed it slightly longer ago (at least, within the context of a laboratory task-switching paradigm where task switch as follows ABACBACABCAB etc.) This is known as a “set alternation” or “backwards inhibition” cost, and is thought to reflect the residual inhibition of task A, even though subjects should now be activating it, rather than inhibiting it.
On the other hand, set-alternation costs might also reflect less efficient processing if the two instances of task A require different responses: in this case, it is not inhibition of A, but rather the particular stimulus-response mapping used on task A that is interfering with performance on the current trial, which may contain a different stimulus. To rule out this explanation, Ulrich Mayr performed an experiment in which the effects of repeating tasks, stimuli, and responses could all be dissociated.
If the mismatching stimulus-response mappings were to blame for set-alternation costs, subjects should not show set-alternation costs on a task where the neither the stimulus nor the response was repeated relative to the last time that task was completed. The results showed that when switching tasks, subjects were slower when switching to a more-recently performed task than to a less-recently performed task (the set-alternation cost). There was less slowing if people were given longer until the instructions of the next trial, but there was not less slowing if people were given longer after the instructions for the next trial. This result is consistent with previous work by the same author, and is interpreted to reflect that the inhibition of the previous task cannot be reduced through preparatory processing (and thus may not be under conscious control).
Importantly, the set-alternation cost also was not affected by whether the response repeated relative to the previous trial. In other words, people were slower to switch to a more recently performed task relative to a less recently performed task, even if everything about that task was the same as when they had last performed it. This result contradicts the theory that set-alternation costs are actually caused by mismatching stimulus-response mappings from previous iterations of the task.
Mayr argues that this supports the idea of inhibition as an enabling factor in the flexibility-stability dilemma in task-switching paradigms, but this argument portrays inhibition in a very strange light: it appears to be an uncontrollable process. It is therefore hard to understand how this automatic and un-overrideable process can be an agent of “high level control.”
Furthermore, some recent evidence suggests that task-switching is associated with a slowing of RTs not only because stimulus-response mappings must change (this is the theoretically interesting aspect of task-switches) but also because you are providing them with a cue that the task may have changed; it takes time to process that. Accordingly, I haven’t seen a replication of the backwards inhibition cost when controlling for the effects of cue switches.
In the end, however, Mayr has presented interesting evidence in favor of one theory about the stability-flexibility dilemma. There are reasons to think that the putative inhibition in this task is not the same as other forms of inhibition (as in Stroop or Stop Signal), primarily in that this appears to be an involuntary form of inhibition.