A major challenge for researchers trying to understand how humans engage in voluntary behavior is our flexibility. We respond differently to the same situation depending on our internal goals and task states. Complicating things further, we often rapidly switch between one set of actions and another while remaining in the same environment and exposed to the same stimuli. This shifting between tasks, although the term ‘tasks’ is ill-defined (see Box 1), is often framed as a symptom of the modern world. We may (but shouldn’t) talk on our phones as we drive or check email during meetings. However, even when separated from technology, we engage in task switching as we contemplate dinner while on an afternoon walk or monitor children while gardening. In fact, when we are focused on what we think of as a single activity, switching is often required: a basketball player defending an opponent must track the opponent’s movements, watch for screens, and prepare to box out and switch to offense, so the act of playing basketball includes many tasks with competing demands and conflicting actions, sometimes signaled by the same stimulus (e.g. going for the steal or preventing the opponent’s driving to the basket). Moreover, switching is not a uniquely human activity (e.g. [1]); animals may forage while watching for predators and tracking other members of their group [2]. In short, task switching may describe the typical means by which we choose our behaviors in environments where opportunities and risks are hard to predict (see 3, 4•, 5).

To capture this aspect of behavior — that different internal states may rapidly change responses to the same environment — researchers have developed task switching procedures that attempt to recreate this demand. Typical task switching experiments involve the presentation of stimuli associated with more than one task, called ‘bivalent’ or ‘multivalent’ stimuli. For example, subjects might be presented with a colored shape whose color and shape are mapped to distinct responses (e.g. green -> key 1; square -> key 4). Which response should be produced may be determined by a cue presented near in time to the stimulus (e.g. 6, 7, 8••, 9••), a memorized sequence of tasks (e.g. [10]), or up to the subject (e.g. 11•, 12•) (see Box 2). The general finding is that, in each of these procedures, performance is worse when the task on the previous trial is different (e.g. respond to shape and then respond to color) than when the task is the same (e.g. response to color and then respond to color) (for reviews see 13••, 14••, 15••). This decrement in performance is called a switch cost.

Here, I selectively review some conceptual and empirical issues relating to switch costs, treating them as distinct from other behavioral phenomena associated with performing two or more tasks close together in time, such as dual-task costs (i.e. the costs associated with performing tasks in a temporally overlapping fashion) [16] and mixing costs (sometimes called global switch costs, i.e. the costs associated with performing a task when other tasks must be kept in mind) [17]. However, these other forms of costs may be equally important for understanding how action is selected and controlled and can be affected differently by various factors. For example, both dual-task costs [18] and mixing costs [19•] appear to be more sensitive to the cognitive changes associated with aging than switch costs.