Humans are often praised for their remarkable ability to maintain task focus, and quickly switch back and forth between different tasks, goals, or strategies — the skills or phenomena of which are often referred to as cognitive flexibility. Research on cognitive flexibility has spawned a vast set of empirical studies, and several theories and models aiming for a more mechanistic understanding of the cognitive processes that allow us to be flexible (for reviews, see 1, 2, 3•, 4, 5). A recurring idea in this literature is that we can flexibly adjust control parameters when asked, such as the degree to which we pay attention to task- relevant versus -irrelevant information, or the extent to which we keep different tasks coactive to flexibly switch between them (see Figure 1). In this review, we challenge the idea that humans can intentionally up- or downregulate these control processes on the fly. Instead, we demonstrate that humans often need considerable training and experience to adapt or fine-tune control process parameters.

When thinking of cognitive flexibility and control in the context of cognitive models, cognitive flexibility is achieved by exerting control over arguably ‘lower-level’ processing pathways (e.g. stimulus-response mappings) through task or goal representations (e.g. 6, 7••, 8, see Figure 2). Humans are clearly efficient at setting, and switching between, such different, concrete task goals on demand (e.g. 9, 10), or at quickly reconfiguring a new task based on instructions (e.g. 11, 12, 13). In this review, however, instead of focusing on this ability per se, we focus mostly on the degree to which people activate these task goals, or the degree of switch readiness when switching between them (e.g. sometimes referred to as ‘control signal intensities,’ as opposed to ‘control signal identities,’ [14]). Variations in the degree of task focus are often investigated by measuring congruency effects in conflict tasks (Figure 1 a and c; for a review, see [11]), assessing the negative impact of conflicting, task-irrelevant features on performance. Fluctuations in the degree of switch readiness are often investigated by measuring switch costs in task-switching designs (Figure 1 b and d; for a review, see 9, 10), which measure the performance cost of switching between tasks compared with repeating a task. Both of these measures have been shown to vary depending on the need for control, often taken as evidence for the idea that people can ‘set’ different levels of task focus or switch readiness at will. This regulation of control parameters is sometimes also referred to as ‘meta-control’ [15], and, by analogy, can be thought of as parametrically turning a knob or a fader (e.g. [14]).

Theories on cognitive control or executive functions converge on the idea that control processes, although applied flexibly, are nontrivial and demanding. For example, in a review on cognitive flexibility and other executive functions, Diamond [16] argued that the functions that allow for cognitive flexibility are arguably the most demanding of all executive functions. Similarly, the expected value of control theory [14] starts from the idea that there is a cost to the control processes required for flexible cognition. Also, in the dual mechanisms of control framework [17], it is postulated that maintaining task goals or adjusting control processes is resource-consuming and -demanding. Nonetheless, these, and virtually all other theories on cognitive control, share the assumption that different control parameter settings can be prepared for, as long as the benefits outweigh the costs (e.g. when promised high rewards, 18, 14). Here, we would like to argue that there are more limitations to adjusting these control parameters than most might assume, or at least more than the labels ‘executive’ functions or ‘control’ seem to hint at. Specifically, in what follows, we will argue (1) that humans are not effective in parametrically regulating and setting different control parameters on demand, and (2) that changing these control parameters requires extensive practice and experience with the task.