We all get tired. Sometimes we get so tired that we find it almost impossible to stay awake. Especially if we are in a meeting of the parking committee, and perhaps even if we are a meerkat:

Although we are all familiar with the feeling of fatigue, we may not always realize that fatigue comes in different variants. The feeling of physical fatigue after a marathon is very different from the fatigue that we experience after programming in R for several hours in a stretch. The latter fatigue is known as cognitive fatigue and kicks in even in the absence of any physical exertion. Driving for hours, programming all day, or even reading James Joyce’s Ulysses can leave us feel spent and unable to reply to another email until tomorrow.

A large body of literature has localized the brain regions involved in experiencing cognitive fatigue. We now know that the cortico-striatal network is implicated in fatigue. This network involves the striatum, the ventromedial prefrontal cortex (vmPFC), and the anterior cingulate cortex (ACC). Many neuroimaging studies have found that the ACC was crucially involved in cognitive fatigue.

Aside from localizing an important phenomenon in the brain, the identification of the ACC is also of some theoretical interest because of its prominent role in effort, reward processing, and error monitoring. For example, when participants make an error, they experience a discrepancy between what they expect (namely, to be correct) and the actual outcome (oops…), and this discrepancy has a neural signature in the ACC.

Is it mere coincidence that the same brain region, namely the ACC, responds to a violation of expectancy as well as to cognitive fatigue? Or might there be a common functional foundation, for example because fatigue might reflect an imbalance between effort and reward?

A recent article in the Psychonomic Society’s journal Cognitive, Affective, & Behavioral Neuroscience examined this question.

Researcher Glenn Wylie, with colleagues Genova, DeLuca, and Dobryakova, presented participants with a working memory task that varied in difficulty for extended periods (four “runs” of trials). This manipulation was expected to induce cognitive fatigue, especially for the more difficult variant of the task. Activity in the ACC was monitored via functional imaging (fMRI), and the main focus during analysis was on how ACC activation related to error processing and self-ratings of fatigue.

The figure below provides a snapshot on the behavioural data, showing latency and accuracy for the working-memory task across the four runs:

It can be seen that although accuracy remained at a (nearly) constant level across runs, performance slowed down as participants became more fatigued. The two different lines in each panel correspond to the two versions of the task: The “0-back” task (red) involved participants responding to each occurrence of the letter “K” in a long sequence of stimuli. Clearly, this was a very easy task as revealed by the near-perfect accuracy and short response latencies. The “2-back” version of the task required participants to monitor the same sequence of stimuli, but they had to respond when a letter (any letter) appeared that had been presented two events previously (e.g., respond to the second “N” in the sequence “R N Q N”). This version of the task was clearly more difficult, as indicated by the lower performance. It also led to increasing cognitive fatigue, as indicated by the slowing of response times.

The self-ratings of fatigue obtained at the end of each task confirm the pattern in the latency data, as shown by the next figure:

It is clear that the more fatigued people were, the more slowly they had responded in the “2-back” task, whereas there was no such relationship for the simple “0-back” task.

A similar pattern emerged when fatigue ratings were related to the fMRI signal change in the ACC, as shown in the figure below. Note that only trials with correct responses were included in the analysis of ACC activation. The pattern remained unchanged if trials following an error were also excluded from analysis. By excluding error trials (and trials following errors), the analysis isolated the involvement of the ACC in cognitive fatigue without a potential contribution from error processing.

The next question of interest is what brain regions were involved in error processing. To answer this question, Wylie and colleagues separated trials into correct and incorrect responses and contrasted brain activity between those two types of trials. The largest difference between correct and incorrect trials was observed in the ACC, with a few other areas also showing differences that were smaller in magnitude.

The fMRI results are therefore readily summarized: Areas that were activated by error processing were also activated when participants reported that they were fatigued. This confirmed the researchers’ expectation that outcome prediction and monitoring might represent processes that are also involved in cognitive fatigue.

How exactly might cognitive fatigue represent an issue of cognitive control? Unlike physical exertion, which leads to fatigue because of depletion of the energy reserves that are stored in our muscles (in the form of glycogen), cognitive fatigue is not commonly associated with depletion of an identifiable biological agent. Instead, as Wylie and colleagues put it:

“… the feeling of fatigue may be one way that the brain signals itself that the payoff matrix has changed and that the reward being received no longer merits the effort being expended—an important aspect of self-regulation.”

So perhaps programming in Python is cognitively exhausting not because you run out of some “brain fuel” but because putting the final touches on your new randomization algorithm is not as rewarding as it was to get the algorithm to work in the first place.

One implication of this interpretation is that cognitive fatigue ought to be fairly readily reversible. If this fatigue is the result of a mismatch between effort and expected rewards, could we somehow “reset” this mismatch? In line with this expectation, here is a list of a few small and easy things you can do to avoid or reverse cognitive fatigue by changing the payoff matrix. For example, give yourself a reward after a few hours of R programming, such as a pint at the pub or a piece of chocolate or even just a brief look out the window at a nice garden. Alternatively, on the other side of the equation, keep the effort lower by minimizing the number of decisions that are unrelated to your programming project—stick to a routine by ordering the same coffee as always, get the same Indian takeaway if it’s Wednesday, and so on.

Psychonomics article highlighted in this post:

Wylie, G. R., Genova, H. M., DeLuca, J., & Dobryakova, E. (2017). The relationship between outcome prediction and cognitive fatigue: A convergence of paradigms. Cognitive, Affective, & Behavioral Neuroscience, 17, 838-849. DOI: 10.3758/s13415-017-0515-y.