Cognitive caching promotes flexibility in task switching: evidence from event-related potentials

Abstract

Time-consuming processes of task-set reconfiguration have been shown to contribute to the costs of switching between cognitive tasks. We describe and probe a novel mechanism serving to reduce the costs of task-set reconfiguration. We propose that when individuals are uncertain about the currently valid task, one task set is activated for execution while other task sets are maintained at a pre-active state in cognitive cache. We tested this idea by assessing an event-related potential (ERP) index of task-set reconfiguration in a three-rule task-switching paradigm involving varying degrees of task uncertainty. In high-uncertainty conditions, two viable tasks were equally likely to be correct whereas in low-uncertainty conditions, one task was more likely than the other. ERP and performance measures indicated substantial costs of task-set reconfiguration when participants were required to switch away from a task that had been likely to be correct. In contrast, task-set-reconfiguration costs were markedly reduced when the previous task set was chosen under high task uncertainty. These results suggest that cognitive caching of alternative task sets adds to human cognitive flexibility under high task uncertainty.

Figure 1. Illustration of our manipulation of task uncertainty (top) and its hypothesized effects on the activation of task sets (bottom).

Under high task uncertainty, participants had to decide between two tasks that were equally likely to be correct. We propose that in these situations, individuals activate one task set for execution and keep the unchosen task set in a pre-active state in cognitive cache, thereby facilitating later retrieval of this task set. Under low task uncertainty, participants knew that one task was more likely to be correct than the other. We propose that in these situations, individuals largely refrain from cognitive caching, rendering later retrieval of the alternative task set slow and inefficient.

Figure 2. Task flow of the task-switching paradigm modeled after the Wisconsin Card Sorting Test.

Participants were instructed to match a target card (by pressing a respective key) to one of four key cards according to the correct task. Feedback cues following each sorting response indicated whether the applied task had to be switched or repeated. Following a switch cue, participants were required to guess which of the two remaining tasks was now correct. If they had guessed correctly, they were presented with positive post-switch feedback (positive PF) initiating a first repetition trial. If they had guessed incorrectly, they were presented with negative post-switch feedback (negative PF) initiating an addendum switch trial. Both positive and negative PF allowed inducing the correct task rule, but only negative PF required an additional switch of tasks.

Figure 3. Mean latency and accuracy of responses on switch, addendum switch and first repetition trials.

Error bars indicate standard error of the mean.

Figure 4. Grand average ERP waves at midline sites elicited by switch cues and positive post-switch feedback (pos PF) cues.

Waveforms are depicted as a function of switch probability (low probability [low prob] vs. high probability [high prob] (left) and task uncertainty (low task uncertainty [low uncert] vs. high task uncertainty [high uncert] (right).

Figure 5. Grand average ERP waves at midline sites elicited by negative post-switch feedback (neg PF) cues and positive post-switch feedback (pos PF) cues.

Waveforms are depicted as a function of switch probability (low probability [low prob] vs. high probability [high prob]) (left) and task uncertainty (low task uncertainty [low uncert] vs. high task uncertainty [high uncert]) (right).

Figure 6. Difference waves and scalp maps illustrating the effect of task uncertainty on neural addendum switch costs.

ERP waveforms elicited by positive post-switch feedback cues were subtracted from ERP waveforms elicited by negative post-switch feedback cues to obtain the neural activity specific to addendum switch operations (left). Scalp maps (right) depict the topography of this addendum-switch-specific activity.

Figure 7. Behavioral and electrophysiological switch costs (switch trial – first repetition trial) and addendum switch costs (addendum switch trial – first repetition trial), separately for low-uncertainty conditions and high-uncertainty conditions.

Error bars indicate standard error of the mean.