Why are some tasks experienced as more effortful than others? To address this question, it is useful to distinguish between reactive action control in unpredictable environments and predictive control in predictable environments. These different types of action control are supported by different brain systems. Predictive control areas are associated with the dorsal prefrontal cortex, dorsal anterior cingulate cortex (dACC), and dorsal striatum, which sustain feedforward action control in tasks that are familiar and predictable (Luu et al. Reference Luu, Jiang, Poulsen, Mattson, Smith and Tucker2011; Tops & Boksem Reference Tops and Boksem2011; Reference Tops and Boksem2012). By contrast, reactive control areas include the inferior frontal gyrus (IFG) and anterior insula (AI), which sustain momentary feedback-guided control when tasks are performed that are novel, urgent, or unpredictable. Reactive control thus represents a specialized mode of operation for detecting new information, encoding it in memory, and assimilating it into preexisting knowledge structures, and for changing earlier schemata, thereby facilitating future predictive control (Hasher & Zacks Reference Hasher and Zacks1979; Tops & Boksem Reference Tops and Boksem2011). Because reactive control reduces predictive homeostatic regulation of the internal milieu (discussed below), such cognitive control requires the momentary tracking of physiological costs and resources and is experienced as effortful. The experience of effort is hence an adaptive motivational mechanism that limits the (re-)initiation and prolonged performance of tasks that demand reactive control, especially when there are insufficient perceived benefits, threats, or resources to compensate for the physiological costs of reactive control (Boksem & Tops Reference Boksem and Tops2008).

The notions of predictability and controllability are central to understanding which challenges trigger a physiological stress response (Sapolsky Reference Sapolsky2005). Physiological responses to challenge parallel the two forms of action control: Reactive homeostatic responses arise in relation to changes in physiological variables that have already occurred, and predictive homeostatic responses emerge in anticipation of predictably timed challenges (Moore-Ede Reference Moore-Ede1986; Romero et al. Reference Romero, Dickens and Cyr2009; cf. Landys et al. Reference Landys, Ramenofsky and Wingfield2006). When a challenge or task is perceived as predictable and controllable, because resources are perceived to be sufficient for the task (e.g., enough muscle strength), predictive homeostasis is maintained and the task may not be experienced as effortful. By contrast, situational novelty (e.g., Hasher & Zacks Reference Hasher and Zacks1979; Shiffrin & Schneider Reference Shiffrin and Schneider1977) and unpredictability of cognitive operations (Ackerman Reference Ackerman1987; Fisk & Schneider Reference Fisk and Schneider1983) require effortful processing and can trigger reactive physiological responses that potentially incur health costs (Romero et al. Reference Romero, Dickens and Cyr2009). Importantly, reactive homeostatic control may decrease less urgent predictive homeostatic regulation, causing “somatic neglect” of, for example, circadian variation in appetite (Koole et al., in press).

Neuroimaging evidence supports our thesis that reactive control systems translate information about action costs and resources into a motivational feeling of effort. Through its reciprocal connections with autonomic and visceral centers of the nervous system such as the hypothalamus (Carmichael & Price Reference Carmichael and Price1995), the AI may be involved in the monitoring and regulation of peripheral resources such as glucose levels (Allport et al. Reference Allport, Butcher, Baird, MacGregor, Desmond, Tress, Colman and Davis2004), muscle condition (Craig Reference Craig2003), autonomic activation (Critchley et al. Reference Critchley, Wiens, Rotshtein, Ohman and Dolan2004), and the processing of aversive bodily states (Paulus & Stein Reference Paulus and Stein2006). In addition, insula activation has been related to the subjective perception of physical effort and exertion (de Graaf et al. Reference de Graaf, Gallea, Pailhous, Anton, Roth and Bonnard2004; Williamson et al. Reference Williamson, McColl, Mathews, Ginsburg and Mitchell1999; Reference Williamson, McColl and Mathews2003). The IFG/AI areas that are active when people experience subjective effort are also implicated in compensatory effort allocation with time on task. One study found the bilateral AI to be involved in assessing the level of energy expenditure required to reach a proposed effort (Prévost et al. Reference Prévost, Pessiglione, Metereau, Clery-Melin and Dreher2010), while several other studies suggested that increased attentional effort during performance over extended periods of time or after sleep deprivation is associated with increased activation of right-hemisphere ventral cortical areas including IFG/AI, and sometimes in the context of activity declines in dACC and/or the dorsolateral prefrontal cortex (Bell-McGinty et al. Reference Bell-McGinty, Habeck, Hilton, Rakitin, Scarmeas, Zarahn, Flynn, DeLaPaz, Basner and Stern2004; Chuah et al. Reference Chuah, Venkatraman, Dinges and Chee2006; Coull et al. Reference Coull, Frackowiak and Frith1998; Paus et al. Reference Paus, Zatorre, Hofle, Caramanos, Gotman, Petrides and Evans1997; Walker et al. Reference Walker, Stickgold, Alsop, Gaab and Schlaug2005). Moreover, momentary lapses in attention, which increase with time on task and fatigue, are associated with reduced activity in this right ventral attentional network, whereas its compensatory recruitment during subsequent trials is associated with recovery from lapses in attention (Weissman et al. Reference Weissman, Robets, Visscher and Woldorff2006).

Thus, the AI may influence action-selection by monitoring the availability of resources and the physiological costs associated with actions. The readout of this monitor may be experienced as feelings of effort, resistance, and discomfort that influence choices to initiate or (dis)continue task performance (Tops & de Jong Reference Tops and de Jong2006). Unlike what Kurzban et al. propose, increased subjective effort does not necessarily shift engagement towards alternative, more rewarding options, but may also stimulate disengagement, inactivity, and recuperation when perceived resources (as signaled by the AI) are low (Boksem & Tops Reference Boksem and Tops2008). In our view, this is the most important role of subjective effort in decision-making. Indeed, effort may be considered as an adaptive signal that the present behavioral strategy is no longer appropriate, because it continues to demand reactive control that usurps costly physiological resources when substantial resources have already been invested and the goal evidently has not yet been achieved. Feelings of effort may provide the cognitive system with a signal that stimulates lowering of current goals and/or seeking of less demanding alternative strategies.

A major advantage of our account over Kurzban et al.'s is that ours more precisely explains which tasks trigger subjective effort and fatigue (i.e., those that require reactive control, such as tasks that are novel or urgent). Moreover, our account is able to address the transition of prolonged effortful demand into persistent forms of fatigue. When the situation is uncontrollable, individuals are forced to rely on reactive control, associated with feelings of effort, up-regulation of reactive homeostatic responses, and decreased predictive homeostatic regulation. Although adaptive in the short-term when dealing with important and urgent situations, prolonged reactive homeostatic control can lead to enduring physiological changes (Romero et al. Reference Romero, Dickens and Cyr2009), which may give rise to chronic fatigue.