According to some perspectives, anterior cingulate cortex (ACC) may become activate in situations where the reward value of given representation or stimulus has decreased, resulting in more competition between representations. Activation of this region may help increase tonic norepinephrine, resulting in more exploratory behavior, and thus more variable responding.
A different perspective on ACC is advocated in this 2001 article by Braver, Barch, Gray, Molfese and Snyder. They also suggest that ACC is important for resolving conflict between multiple responses - and would therefore be particularly important for making low-frequency responses during which a different, prepotent reponse must be resisted.
Braver et al. begin by reviewing the variety of circumstances known to elicit ACC activity: difficult tasks, those with large numbers of errors, those involving response competition (i.e., Stroop) and those where there are many correct answers (i.e., under-determined responding). All of these tasks seem to involve conflict between mutiple possible responses.
The authors suggest that low-frequency responding is a "minimal condition" for this kind of conflict, and that it matters little what the low frequency response actually is: infrequently withholding a response should show similar ACC activity as infrequently selecting a response among many, which should show similar ACC activity as infrequently giving a response where only one response is possible.
To test this hypothesis, Braver et al. administered three types of tasks to 14 subjects in an fMRI scanner. Casual readers may wish to skip the following methodological details, which are in italics.
The first task was a version of go/nogo in which subjects had to withold responding to an "X" (17% of trials) while responding to any of 25 other letters (together occuring with 83% frequency).
In contrast, the second "target detection" task required subjects to refrain from responding to "X" (occuring on 83% of trials) while providing a response to any of 25 other letters (altogether occurring with 17% frequency).
Finally, the third "response selection" task involved responding with the left hand to an "X" and responding with the right hand to any of 25 other letters. This task had four conditions: a "low X" condition (where X's occurred with 17% frequency and the other 25 letters with 83% frequency altogether), a "high X" condition (where X's occurred with 83% frequency and the other 25 letters with 17% frequency combined).
There were also two additional tasks: an "equal frequency response inhibition" task (where the X occurred with equal probability as all 25 other letters combined, and subjects had to respond only to X) and an "equal frequency response selection" task (where the X occurred with equal probability as all 25 other letters combined, and subjects had to give a left-handed response to X and a right-handed response to any of the 25 other letters). These tasks were presented in blocked fashion, with a 35 second passive fixation baseline.
As expected, the subjects were less accurate and slower on trials where the correct response was less frequent. Activity in the ACC corresponding to low-frequency responses was equivalent in magnitude across the three tasks, and was significantly higher than ACC activity in the tasks with equal frequency. This confirms the hypothesis that ACC is sensitive to response frequency, rather than more specific task demands (for example, response inhibition).
Braver et al. also discovered that right dorsolateral and ventrolateral PFC showed selective activation in response to No-Go stimuli, but only in the conditions where No-Go responses were infrequent. Similarly, Braver et al. suggest that the dlPFC activation could reflect reorienting of attention towards "rare or novel events." The authors also note that their results are consistent with a fractionation of the ACC in which specific subregions are more sensitive to low frequency responding or more sensitive to errors.
Previous computational models of ACC suggest that it becomes active in situations of response conflict, where more than one response becomes activated. Yet other models have focused on the possible role of ACC in detecting errors, and have provided a convincing fit to both behavioral and neuroimaging evidence. According to either framework, activity in ACC upregulates activity in prefrontal regions, in order to amplify the representations that are consistent with a currently relevant goal. The newest computational models of this cognitive control network suggest that norepinephrine and the locus coereleus may have a central role in this "gain modulation" of goal-relevant representations.