The Willpower Paradox: Possible and Impossible Conceptions of Self-Control

Conflict-monitoring theory

Although the recruitment of self-control is often conceived as an intentional, self-initiated, deliberate, conscious, and effortful process (Baumeister et al., 2016; Duckworth, Taxer, et al., 2019) that rests on a value-based decision (Kool et al., 2017; Kotabe & Hofmann, 2015; Shenhav, 2017; Shenhav et al., 2013; Westbrook & Braver, 2015), a number of theories in experimental psychology, cognitive neuroscience, and social psychology have endorsed the idea that self-control processes may be triggered automatically. An early version of this idea can be found in Narciss Ach’s classical theory of volition, in which he assumed that a conflict between an intention and a habitual response triggers automatically, without intervening deliberation, an enhanced exertion of “volitional effort” (Ach, 1910). A modern variant of this idea can be found in the initial formulation of the conflict-monitoring theory (Botvinick et al., 2001). The key assumption is that the detection of a conflict between competing action tendencies triggers an increased recruitment of cognitive control that enhances the impact of currently active goals or task sets on information processing and response selection. Consistent with this assumption, in reaction-time tasks involving response conflicts (e.g., Stroop or flanker tasks), interference from competing responses is often reduced in trials immediately after a response conflict, which suggests that conflicts automatically trigger the recruitment of enhanced control (the so-called conflict-adaptation effect; e.g., Egner & Hirsch, 2005; Fischer et al., 2008; Gratton et al., 1992; Kerns et al., 2004; Scherbaum et al., 2011; Stürmer et al., 2002). The conflict-monitoring theory has instigated an enormous amount of research that is beyond the scope of this article (for reviews and critical discussions, see Braem et al., 2019; Egner, 2007; Mansouri et al., 2017; Scherbaum et al., 2012; Schmidt, 2013, 2019). The important point for the current discussion is that in the computational implementation of the initial conflict-monitoring model (Botvinick et al., 2001) the recruitment of control did not depend on the expected utility of the to-be-supported goal but was directly triggered by the detection of a response conflict. ⁷

Counteractive-control theory

Although the conflict-monitoring theory has been mostly applied to response conflicts in simple reaction-time tasks, theories of self-control likewise postulate that the detection of a conflict between a goal and a desire is a trigger condition for the recruitment of self-control (Kotabe & Hofmann, 2015; Myrseth & Fishbach, 2009). Although awareness of a goal-desire conflict may induce a deliberate decision to recruit self-control to override the unwanted desire, research on the counteractive-control theory (Fishbach et al., 2010; Myrseth et al., 2009; Trope & Fishbach, 2000; Zhang & Fishbach, 2010) has yielded evidence that the recruitment of self-control can also be an automatic response to a goal-desire conflict. The key assumption of this theory is that a conflict between a goal and a temptation triggers self-control processes that asymmetrically shift motivation, thereby strengthening goal pursuit and decreasing the motivational strength of a temptation. Such asymmetric shifts in motivational strength may initially rest on conscious, deliberative processes, but the repeated mobilization of self-control in a given goal context may lead to the formation of direct associations between tempting stimuli and the superordinate goals they threaten, as well as to the formation of inhibitory links between goal representations and temptations. These facilitative and inhibitory associations can automatize, such that they are activated unconsciously, effortlessly, and without mediation by a value-based decision (Fishbach et al., 2003). As a result, self-control is no longer a motivated process but an automatic response to conflict that biases the decision process toward cognitions that promote superordinate or long-term goals.

In support of this assumption, studies have shown that temptations (when primed by respective stimuli) automatically trigger thoughts about higher order, long-term goals. For example, in a study using a sequential priming paradigm, participants recognized goal-related words faster when these followed a subliminally presented temptation-related word (Fishbach et al., 2003). In contrast, higher order goals do not increase the accessibility of temptation-related thoughts. On the contrary, activating a higher order goal that a person is committed to inhibits the accessibility of alternative goals (Shah et al., 2002). This speaks to the biased nature of conflict-induced self-control processes favoring superordinate goals over immediate desires and differentiating successful self-controllers from those frequently failing in self-control. In line with this assumption, as compared to successful dieters, those who frequently fail in their dieting attempts do not exhibit such automatic goal-activation when confronted with tempting food (Papies et al., 2008a, 2008b). Supporting the notion of automaticity, the described processes reportedly emerge even under cognitive load and appear not to require working memory (Fishbach & Shen, 2014).

In conclusion, the central point for the current discussion is that dropping the assumption that self-control is an intentional and value-based process avoids the willpower paradox. If the recruitment of control is caused automatically by the detection of conflicts or by situational cues, it is no longer paradoxical to assume that self-control serves to override a short-term desire that is currently stronger than a superordinate goal. In philosophical theories of self-control, this corresponds to the suggestion that “all exercises of synchronic self-control are non-actional . . . because there is no suitable strongest desire to set up an exercise of actional synchronic self-control” (Kennett & Smith, 1997, p. 128).

Dual valuation systems

Theories postulating dual valuation systems assume that a reflective (“deliberative,” “cold,” “cognitive”) system mediates the computation of the subjective utility of goals and actions based on deliberative judgments about the subjective probability and expected value of anticipated future outcomes (Hofmann, Friese, & Strack, 2009; Lindgren et al., 2019; Loewenstein, 1996; McClure & Bickel, 2014; Metcalfe & Mischel, 1999; Strack & Deutsch, 2004; Volkow & Baler, 2015). This system is assumed to operate in a slow, controlled, and resource-demanding manner. By contrast, the impulsive (“hot,” “emotional,” “visceral”) system mediates automatic affective responses and the activation of behavioral impulses by current desires and the prospect of an immediate reward. Impulses are defined as urges to perform a certain behavior (e.g., to act on a temptation). They are evoked when a motivational state (e.g., a current desire that elicits a craving for nicotine) co-occurs with relevant stimuli with a strong incentive value (e.g., a pack of cigarettes) that elicit the prospect of a short-term gratification. In addition to the expected short-term gratification of satisfying an immediate desire, the impulse to give into a temptation may also be strengthened by additional factors such as the convenience of a behavior or social pressure. It is assumed that the impulsive system comprises innate or learned associations between stimuli, responses, and rewards; requires few attentional resources; and operates independently from a person’s conscious goals and expectancies (Hofmann, Friese, & Strack, 2009). Self-control conflicts emerge when the two systems generate inconsistent value signals, because the reflective system computes value signals on the basis of anticipated long-term consequences, whereas the impulsive system is myopic with respect to future outcomes and assigns values to actions on the basis of current needs and the previous learning and reinforcement history.

In support of dual-systems models it has been shown that individual differences in cognitive-control capacities moderate the relative impact of automatic affective reactions and reflective processes on behavioral choice. High control competencies related to executive attention, inhibitory control, and affect regulation promoted self-controlled choices by reducing the influence of automatic affective reactions to tempting stimuli (e.g., candies) and increasing the relative impact of explicit dietary goals (e.g., to reduce sugar consumption) on behavior (candy consumption; Hofmann et al., 2008, 2012; Hofmann, Friese, & Roefs, 2009). This indicates that control processes recruited by the reflective system reduce the impact of automatic valuations by the impulsive system and thereby enhance the impact of long-term goals (see also Wolff et al., 2016).

Further evidence for dual valuation systems stems from studies that used functional MRI to examine brain activation in intertemporal choice tasks, in which participants made choices between smaller-sooner and larger-later rewards (e.g., $5 now vs. $40 in 6 weeks). Initial studies (McClure et al., 2004, 2007) showed that brain regions in the mesolimbic dopamine system involved in reward processing (ventral striatum and medial orbitofrontal cortex) were predominantly activated when choices involved immediate rewards. By contrast, activations in regions involved in cognitive control (lateral prefrontal and posterior parietal cortex) were independent from the delay of the reward. Moreover, choices of later-larger rewards were associated with relatively larger activity in frontoparietal compared with limbic regions. These findings were interpreted as evidence for the competition between an impulsive system that discounts future rewards steeply and assigns disproportionate high value to immediate rewards and a reflective system that discounts delayed rewards only minimally and mediates farsighted choices. Consistent with this interpretation, a more recent study (Turner et al., 2019) found that when participants chose larger delayed rewards, activity in the frontoparietal control network increased with increasing attractiveness of the sooner-smaller reward (i.e., with an increasing need for self-control). Moreover, reduced activation in prefrontal brain regions involved in inhibitory control was associated with an increased propensity to commit daily self-control failures as assessed via experience sampling (Berkman et al., 2011; Krönke et al., 2018; Krönke, Wolff, Shi, et al., 2020; Lopez et al., 2014).

Although later we discuss evidence suggesting that (contrary to dual-systems models) choices rest on a common neural value signal (Hare et al., 2009, 2011; Kable & Glimcher, 2007, 2010), the important point for our current argument is that dual-systems theories can avoid the willpower paradox, provided they make the additional assumption that only the reflective system has the capacity to mobilize self-control, whereas the impulsive system has no access to self-control processes. If one assumes that the motivation to recruit self-control exclusively derives from the goal that is assigned the highest subjective utility based on value computations of the reflective system, one can maintain both the intentionality and synchronicity assumption without raising a conceptual paradox. Whereas in a unitary-agent account, self-control that is both intentional and synchronous implies that the agent most strongly prefers to execute a behavior and simultaneously most strongly prefers not to execute the behavior, this implication is avoided by dividing the agent into subsystems, each of which may assign different utilities to current desires and long-term goals. Even though the goal, which is assigned the highest subjective utility by the reflective system, may be motivationally “weaker” than the competing desire, it is not paradoxical to assume that this goal can motivate the intentional (value-based) recruitment of self-control to override a desire that would otherwise determine behavioral choice. ⁹

In a similar vein, philosophical accounts of synchronous self-control have proposed that divided-minds theories can avoid the willpower paradox, provided they make the additional assumption that self-control rests on proprietary strategies that are exclusively available to and controlled by a reflective or deliberative system (Altehenger, 2021; Sripada, 2014).

Goal-directed versus habit systems

A second class of dual-systems theories that avoid the willpower paradox distinguishes between systems mediating goal-directed (outcome-based) actions and systems mediating habitual responses. Habitual responses are usually defined as behaviors that (a) are acquired through repeated execution and/or reinforcement of actions under invariant stimulus conditions that lead to stable stimulus-response associations, (b) are activated by stimulus contexts or conditioned cues independently from goals or anticipated outcomes, and (c) are insensitive to rapid changes of response-outcome contingencies or outcome values (e.g., a reinforcer devaluation; Carden & Wood, 2018; K. J. Miller et al., 2019; Seger, 2018; Wood & Rünger, 2016). ¹⁰ Thus, self-control conflicts may arise when a stimulus context activates a habitual response that is incompatible with a person’s goals or outcome valuations and that would control behavior unless the goal-directed system intervenes and recruits cognitive control to override the habitual response (Evans & Stanovich, 2013).

Theories postulating distinct goal-directed and habitual systems are supported by evidence from animal research and neuroimaging studies in humans that indicate that the two types of behavior depend on distinct yet interacting neural circuits (Burton et al., 2015; Yin & Knowlton, 2006). Goal-directed behavior has been attributed to an “associative” loop that connects the prefrontal cortex with regions in the dorsomedial striatum, including the caudate nucleus. Habitual behavior is mediated by a “sensorimotor” loop that connects the somatosensory and motor cortex with the medial and posterior putamen (Amaya & Smith, 2018; Ceceli & Tricomi, 2018; Seger, 2018). Consistent with this distinction, the acquisition of a habitual response via extended training was associated with decreasing brain activation in prefrontal areas and associative basal ganglia regions involved in goal-directed control and with increasing activation (Tricomi et al., 2009) and connectivity (Zwosta et al., 2018) in the sensorimotor network.

Like dual-valuation-systems theories, theories postulating distinct goal-directed and habitual systems can maintain both the intentionality assumption and the synchronicity assumption without raising the willpower paradox. If the recruitment of self-control is exclusively controlled by the goal-directed system (and the intention to exert self-control derives its motivational force from the goal that is currently assigned the highest expected utility), and if habitual responses are triggered by a stimulus without mediation by a value-based decision process, it is no longer paradoxical that self-control may help to override an otherwise stronger habitual response. The reason is that there is simply no competing behavioral tendency that is assigned a higher subjective utility than the goal that motivates the recruitment of self-control.

It should be noted that habits may not exclusively be triggered by direct stimulus-response associations that result from mere frequent repetitions of a behavior in a stimulus context (K. J. Miller et al., 2019). Rather, it has been suggested that habits may also develop via a process in which the reward value of response outcomes is conditioned onto context cues accompanying those rewards. With sufficient repetition, such context cues acquire the power to trigger habitual responses (Wood & Neal, 2007; Wood & Rünger, 2016). With respect to the willpower paradox, the important point is that such theories also avoid the willpower paradox because they assume that such a “motivated cuing” of habits does not involve value-based decisions mediated by the goal-directed system. Instead, habitual responses rest on the cumulated (“cached”) reward value of states and actions rather than explicit representations of action-outcome contingencies. Consequently, like dual valuation systems, this view of habits circumvents the paradoxical implication of a unitary valuation system that simultaneously assigns the highest utility to an action and to the intention to prevent executing this action.

Proactive preventive self-control: precommitment, resistance to reconsideration, and implementation intentions

A paradigmatic example of proactive preventive self-control are precommitment strategies, which serve to reduce the likelihood that one will yield to an anticipated future temptation (Duckworth et al., 2016; Kuhl & Goschke, 1994; Kurth-Nelson & Redish, 2012; Soutschek et al., 2017; Studer et al., 2019). In the most extreme case, one can try to prevent self-control conflicts altogether by avoiding situations involving an anticipated temptation, or, if this is not possible, to restrict the space of one’s future behavioral options. The classical example for the latter strategy is Homer’s Ulysses who, in anticipation of the Sirens—dangerous creatures who lured sailors with their enchanting voices to shipwreck on the rocky coast of their island—had himself tied to the mast to prevent himself from giving into the temptation elicited by the Sirens’ song. A more mundane example is the person who knows that they will desperately crave candies late at night and therefore refrains from buying unhealthy snacks in the afternoon to prevent themselves from yielding to the temptation at night.

Precommitment strategies do not always succeed in completely avoiding temptations or making indulgence impossible but often merely increase the cost incurred by giving into a future temptation. For instance, the strategy of not buying candies at daytime to ensure that one sticks to a healthy diet when desperately craving sweets late at night may not eliminate the opportunity to walk to a nearby gas station selling candies until late at night but imposes an additional cost on indulgence in terms of time and effort. Likewise, telling your friends about your dieting plans will impose a cost on indulgence in terms of social embarrassment. An important implication is that precommitment will work only when one succeeds in increasing the cost of indulgence up to a point such that the overall subjective utility of the short-term reward will be lower than that of the long-term goal at the moment of choice.

Precommitment exemplifies a remarkable cognitive capacity of humans to anticipate their own future motivational states and take preventive measures to restrict the space of their future behavioral options. Of note, evidence indicates that precommitment decisions do not recruit the prefrontal-parietal control network typically activated by inhibitory control but are specifically associated with activation in the lateral frontopolar cortex (Crockett et al., 2013; Soutschek et al., 2017), a higher order cognitive region involved in prospection, counterfactual thinking, and metacognitive processes (Duverne & Koechlin, 2017; Soltani & Koechlin, 2021).

With respect to the willpower paradox, the important point is that precommitment strategies are recruited when an expectancy-value deliberation still assigns the maximum expected utility to a long-term goal. At this moment, the desire that would normally elicit a preference for a short-term reward and lead to a self-control failure is not yet present but merely anticipated. Consequently, a long-term goal can motivate the recruitment of preventive strategies to avoid a temptation or prevent oneself from giving into a future temptation without raising the willpower paradox.

A further preventive strategy is resistance to reconsideration (i.e., the attempt to avoid reevaluating one’s earlier decisions and commitments). Once one starts to reconsider a previous decision or commitment when being confronted with a temptation elicited by a current desire, it is likely that one will revise one’s earlier choice. This simply follows from the fact that a key feature of a self-control conflict is precisely that the short-term desire does transiently become a stronger motivational force than the long-term goal and may thus cause a preference reversal. For instance, reconsidering my earlier decision to prevent myself from eating candies when experiencing an intense craving for chocolate late at night bears the risk that I come to the conclusion that walking to a store selling candies at night is the most valuable option at this moment. Thus, reconsidering the subjective expected utilities of the short- and long-term rewards during a self-control conflict will often result in a self-control failure.

Consequently, when a future temptation cannot be avoided by a precommitment decision or if the additional cost imposed on indulgence does not suffice to make it the less preferred option at the moment of choice, preventive self-control will be effective only if one resists reconsideration. In philosophical action theory, Bratman (1987, 2009) has argued that resistance to reconsideration is a genuine consequence of committing oneself to future-directed intentions, which promotes persistence in the pursuit of long-term goals in the face of competing motivational tendencies (cf. Holton, 2009).

One effective strategy that helps to increase the threshold for reconsidering a previous commitment is the formation of implementation intentions (Achtziger et al., 2008; Bieleke et al., 2021; Eder, 2011; Gollwitzer, 1999; Gollwitzer & Oettingen, 2013; Wieber et al., 2015). Implementation intentions denote “if-then” plans that specify under which conditions intended actions are to be executed. By establishing such if-then plans, individuals pass the execution of behavior over to situational cues. For example, a dieter may plan that if they are in a restaurant, then they will order a salad instead of a steak. As a consequence, even if there might be more tempting dishes on the menu, the goal-conducive behavior is supposed to be triggered quasi-automatically by the situational context (the “if” part of the implementation intention). An impressive body of research documents that this specific form of planning indeed promotes the realization of intentions in laboratory tasks and supports goal striving in the face of real-life temptations, as indicated, for instance, by the promotion of healthy behavior and reduced procrastination in academic contexts (for reviews, see Bieleke et al., 2021; Gollwitzer & Sheeran, 2006). The effects of implementation intentions are mediated by a range of mechanisms, including more efficient perceptual processing of information specified in the “if” part of the intention, enhanced accessibility of opportunities, and automated activation of goal-directed responses (Bieleke et al., 2021).

Implementation intentions should not be confused with automatic self-control as described in Resolution 1. The function of implementation intentions is to promote the quasi-automatic triggering of an intended action by stimulus conditions specified in the “if” part of the intention. For example, a dieter may plan that if they are in a restaurant, then they will order a salad instead of a steak. Consequently, even if there might be more tempting dishes on the menu, the goal-conducive behavior is supposed to be triggered quasi-automatically by the situational context. Thus, by “delegating control to the environment” and promoting the quasi-automatic initiation of self-controlled behavior, implementation intentions render synchronous self-control superfluous (Bieleke et al., 2021; Gollwitzer, 1999; Gollwitzer & Bargh, 2005). However, in contrast to the automatic recruitment of self-control as described in Resolution 1, the initial formation of an implementation intention is usually the result of a deliberate choice and therefore constitutes an intentional self-control strategy, even though it serves to subsequently promote the automatic realization of self-controlled behavior.

Reactive preventive self-control: precautionary down-regulation of emerging desires

Proactive preventive self-control strategies avoid the willpower paradox because they are intentionally recruited at a time when a temptation has not yet emerged but is merely anticipated to emerge in the future. Thus, a long-term goal that is assigned the highest expected utility can motivate proactive preventive self-control strategies because there is no current desire that is more preferred. However, preventive self-control can also be exerted in a reactive manner when a conflicting motivational tendency already emerges and starts to challenge a previously formed intention. As long as the motivation to execute the prior intention is still stronger than the competing motivational tendency, the perception of the emerging conflict can induce the intentional mobilization of self-control to down-regulate the emerging urge and to prevent it from overruling the currently dominant goal. Like a totalitarian leader who uses the state’s power to silence oppositionist movements, self-control might be mobilized to prevent emerging conflicting motivational tendencies from threatening a previously set intention. ¹² The intentional recruitment of reactive preventive self-control does not raise the willpower paradox because it occurs at a stage of an emerging conflict when the long-term goal is still the strongest motivational tendency with the highest subjective utility.

Phenomenologically, reactive preventive self-control is the form of nonsynchronous self-control that comes closest to what has been termed “interventive” self-control (Hofmann & Kotabe, 2012). It is typically exerted during a temptation episode in close proximity to possible indulgence and when an urge is actually felt and not merely anticipated. However, we argue that it should nevertheless be considered a preventive self-control strategy because it is exerted when the long-term goal is still the dominant motivational tendency and serves to prevent an emerging desire from becoming the strongest motivational tendency and causing a preference reversal.

Compared with other preventive strategies, reactive preventive self-control operates on a much shorter time scale. The underlying mechanisms thus likely differ qualitatively from those mediating proactive preventive strategies. In particular, reactive preventive self-control requires mechanisms that continuously monitor emerging conflicting urges (like a dictator’s surveillance agencies). Recent evidence from studies combining functional neuroimaging with smartphone-based ecological momentary assessment of daily self-control failures have indeed shown that self-control depends on the efficient functioning of brain systems involved in conflict and performance monitoring and the detection of errors (including the anterior midcingulate cortex, presupplementary motor area, and anterior insula). Individuals who showed low error-related activation in these brain regions in a Stroop-like conflict task were more prone to commit self-control failures in their daily lives (Krönke et al., 2018; Overmeyer et al., 2021). Of note, activation in these brain areas predicted individual differences in real-life self-control over and above self-reported trait self-control (Krönke et al., 2018). These findings are consistent with the assumption that continuous efficient monitoring of emerging conflicts is an important precondition for mobilizing self-control in a timely manner to be able to down-regulate emerging conflict-laden desires before they supersede a long-term goal.

In conclusion, we propose that reactive preventive self-control—like automatic self-control described in Resolution 1—is triggered by the detection of a conflict between a long-term goal and an emerging short-term urge. However, reactive preventive self-control differs in important respects from automatic self-control. Whereas Resolution 1 rests on the assumption that self-control strategies can be triggered automatically by a conflict, even when a competing desire is already stronger than the to-be-supported superordinate goal, reactive preventive self-control can be initiated intentionally as long as the competing desire has not yet become the strongest motivational tendency.

Phenomenological challenges

A challenge for the automaticity resolution derives from the observation that exercising self-control is often experienced subjectively as an intentional, self-initiated effort to resist temptation. For the automaticity resolution this raises the question of how to account for the fact that self-control is often accompanied by a distinctive agentic quality and feeling of effort. One possible explanation could be that such phenomenological manifestations do not reflect the intentional exertion of control but the perception of a conflict. The simultaneous activation of two action tendencies with similar strength may evoke the subjective feeling of an internal struggle between a goal and the lure of an immediate reward and explain why self-control episodes feel effortful, agentic, and intentional, even if the recruitment of self-control is automatic. Moreover, the specific phenomenological quality of self-control may arise because one feels that one is about to override a strong desire in the service of a long-term goal that may be more entrenched in a network of self-relevant norms and moral values, and this may be experienced subjectively as the victory of one’s genuine “self” over unwanted impulses (cf. Berkman, Hutcherson, et al., 2017). Testing this assumption would require experimental designs that manipulate conflict strength independently from the degree of effort required to exert self-control.

The generality challenge

Although we pointed to evidence that self-control processes can under certain circumstances be triggered automatically, it is unlikely that all instances of self-control are nonintentional in the sense of not being mediated by a value-based choice. As noted earlier, many prominent self-control theories assume that the recruitment of self-control strategies is something that we usually bring about intentionally; is self-initiated, deliberate, and conscious (Baumeister et al., 2016; Duckworth, Taxer, et al., 2019); and is the result of a “decision to inhibit or to override an unwanted behavior” (Hofmann, Friese, & Strack, 2009, p. 165). Likewise, a key assumption in recent theories of cognitive control, notably the expected value of control theory (Shenhav, 2017; Shenhav et al., 2013; Silvestrini et al., 2022), is that the mobilization and allocation of control strategies is the result of a value-based decision-making process, whereby costs and benefits of control are integrated into an estimate of its expected value (Kool et al., 2017; Westbrook & Braver, 2015). The assumption that individuals intentionally choose to exert willpower is further supported by the fact that such choices are often taken as indicating which goals persons consider as most important for their core values, self-concept, and identity, as postulated in the identity-value model of Berkman and colleagues (Berkman et al., 2017a). Thus, although automatic self-control provides a tenable resolution of the willpower paradox, the generality of this resolution is limited because in many instances of goal-desire conflicts the recruitment of self-control appears to be an intentional and deliberate process.

Implications for future research

For theories abandoning the intentionality assumption it will be important to obtain further evidence under which conditions the recruitment of self-control can occur automatically and to specify in more detail in what sense self-control can be automatic. So far, empirical evidence for automatic self-control stems mainly from laboratory experiments demonstrating that response conflicts in interference tasks can trigger the mobilization of enhanced cognitive control (for reviews, see Botvinick et al., 2004; Braem et al., 2019; Egner, 2007; Mansouri et al., 2017) and that temptation-related stimuli can automatically activate representations of superordinate goals (Fishbach et al., 2003, 2010; Myrseth et al., 2009). For theories abandoning the intentionality assumption it will thus be important to obtain converging evidence for the automatic recruitment of self-control using a wider range of experimental paradigms, in particular tasks involving conflicts between long-term goals and immediate desires or cravings. This appears particularly important given that the evidence for the conflict-triggered mobilization of control in simple response-time tasks has become the target of empirical and methodological critiques (for a review, see Schmidt, 2019).

The feature of automatic self-control that is critical for avoiding the willpower paradox is that the recruitment of self-processes is nonintentional in the sense that it does not rest on a value-based decision. Thus, the key empirical challenge is to show that self-control can be triggered directly by conflicts without mediation by an evaluation of its expected value. One way to obtain such evidence might be to investigate whether self-control processes can be activated by conflicts even under high cognitive load, which should limit the possibility to engage in deliberative processes (cf. Fishbach & Shen, 2014). Another possibility might be to examine whether self-control processes can under certain circumstances be recruited in the absence of activation in brain regions encoding expected value of control. This will be particularly challenging, however, given that regions such as the anterior cingulate cortex appear to be involved both in the detection of conflicts and the evaluation of the costs of effortful control (Botvinick & Braver, 2015; Westbrook et al., 2019; Yee & Braver, 2018; Yee et al., 2021).

A further point of relevance for future research on automatic self-control concerns the fact that automaticity is a multidimensional construct that can refer to whether a process is unconscious, effortless, stimulus-triggered, or independent from anticipated outcomes (Moors, 2016). These different criteria for automaticity can dissociate. For instance, a process may be triggered by a stimulus but nevertheless depend on a prior intention, a process may be effortless although being initiated intentionally, or a process may run off unconscious but nevertheless be modulated by goals (Goschke, 2003; Hommel, 2000, 2019; Moors & De Houwer, 2006). Consequently, the view of automatic and intentional processes as a dichotomy of two distinct types of processes has been replaced by models that assume that intentions “configure” response dispositions and modulate their readiness to be triggered when execution conditions are satisfied (Bargh, 1989; Goschke, 2003, 2013; Hommel, 2000, 2019). Of note, this view of “conditional automaticity” relates to the concept of implementation intentions we discussed above in the context of Resolution 3. There we focused on implementation intentions as a means to prevent reconsideration and to promote the automatic triggering of an intended behavior when its execution conditions are met. However, it is also conceivable that persons form implementation intentions that refer to the recruitment of self-control itself. For example, a dieter might form the plan “If I feel tempted by an unhealthy treat then I will stop and first count backward for 30 seconds,” and this strategy may be activated automatically when facing the temptation and help to down-regulate the desire.

Finally, it would be important to investigate how automatic self-control emerges and how people acquire the ability to mobilize self-control automatically. One possibility would be to conduct learning experiments to examine whether and under which conditions the mobilization of self-control processes becomes automatic or habit-like with extended practice. For instance, to our knowledge it has not been investigated experimentally whether repeatedly associating temptation-eliciting stimuli with primes that activate superordinate goals leads to the formation of automatic links between temptations and goals as postulated in the counteractive-control theory (Fishbach et al., 2010). Likewise, one could investigate whether participants who repeatedly exert self-control when making intertemporal choices gradually develop a habit-like disposition to recruit control when being confronted with conflicts between immediate and delayed rewards. A particularly strong test of Resolution 1 would be to demonstrate that such a habit-like disposition to recruit control transfers to contexts in which it may even be advantageous to choose immediate rewards.

The asymmetrical-control challenge

Dual-systems theories have been the target of a range of general theoretical critiques that are beyond the scope of this article (De Houwer, 2019; Evans, 2019; Goschke, 2003, 2013; Grayot, 2019; Hommel, 2019; Melnikoff & Bargh, 2018a, 2018b; Pennycook et al., 2018). A central challenge for dual-systems theories as a possible resolution of the paradox is that they avoid the paradox only if they make the additional assumption that only the reflective system has exclusive access to self-control strategies and that it recruits these strategies to selectively support goals that are motivated by expected utilities computed by the reflective system. Conversely, it must be assumed that the competing impulsive or habitual responses are generated by separate systems that have no access to self-control strategies. These assumptions require independent justification, however, because it is prima facie not obvious why a current desire may not also trigger self-control. This appears conceivable especially when a desire does not automatically activate an impulsive response tendency but merely increases the subjective value of an immediate reward and thereby activates a respective short-term goal. It is an open question whether such short-term goals may motivate the recruitment of self-control strategies that may, for instance, serve to suppress thoughts about the adverse long-term consequences of satisfying an immediate desire.

Moreover, there is evidence that self-control can under certain circumstances even be undermined by deliberative reasoning. This research suggests that individuals who find ways to justify or rationalize impulsive choices to reduce feelings of guilt are more prone to give in to temptations (De Witt Huberts et al., 2013). Thus, self-control failures may result not only from a lack of deliberation but also from deliberation that generates reasons for why a smaller short-term reward should be preferred over a larger long-term reward. This raises the question: Under which conditions and based on which evaluations does the reflective system “decide” whether to recruit self-control to strengthen a long-term goal or whether to generate reasons to justify indulgence? In essence, we are back at facing the willpower paradox, albeit now with respect to the reflective system rather than the person conceived as a unitary agent. Thus, for dual-valuation-system theories to qualify as a tenable resolution of the willpower paradox, it will be critical to obtain further evidence about whether and under which conditions the impulsive system may also have access to self-control processes and under which conditions reflective processes are recruited to undermine rather than support self-controlled behavior.

The goal-dependent habit challenge

With respect to the second class of dual-systems theories, which postulate distinct goal-directed and habit systems, it is important to note that they avoid the willpower paradox only if habitual responses are indeed independent from goals and outcome valuations. If this assumption holds, it is not paradoxical to state that self-control is recruited intentionally to override an otherwise dominant habitual response because the response is not an outcome- and value-based action and there is thus no competing response that is assigned a higher subjective utility than the goal that motivates the recruitment of self-control. However, this assumption has recently been challenged, and it has been argued that many instances of seemingly habitual behavior do in fact depend on goals and are sensitive to changes in the reward value of action outcomes (Hogarth, 2020; Kruglanski & Szumowska, 2020). For instance, there is evidence that insensitivity of well-practiced responses to outcome devaluation (as a critical indicator of habits) may reflect attachment to a different goal or dysfunctions in goal-directed control rather than habitual behavior (de Wit et al., 2018). Although this does not exclude the possibility of value-free and goal-independent habits that are acquired through the repeated strengthening of stimulus-response associations (K. J. Miller et al., 2019), the evidence for value-free habits in humans appears less compelling than often assumed (De Houwer et al., 2018). Thus, for goal-habit theories to qualify as a tenable resolution of the willpower paradox, it will be critical to obtain more convincing evidence that habitual responses in humans are (at least under certain circumstances) independent from goals and outcome valuations.

Implications for future research

For dual valuation systems a key question for future research concerns the ongoing debate about whether behavioral choice is the outcome of a competition between a reflective and an impulsive valuation system or whether behavioral choices rest on a common value signal that is computed by a unitary valuation system (Berkman et al., 2017a, 2017b; Berkman, Hutcherson, et al., 2017; O’Leary et al., 2017; Shenhav, 2017; Sklar & Fujita, 2017).

As mentioned, neuroimaging findings on neural correlates of intertemporal choices (McClure et al., 2004, 2007) were initially interpreted as strong evidence for distinct impulsive and reflective valuation systems in reward-related “limbic” brain areas and prefrontal-parietal control regions, respectively. However, subsequent studies found that activity in so-called limbic brain areas is not specifically elicited by the prospect of an immediate reward but tracks the subjective value of both immediate and (discounted) delayed rewards, and no evidence was found for an impulsive system that disproportionally values immediate reward (Kable, 2014; Kable & Glimcher, 2007, 2010). Meta-analyses of neuroimaging studies likewise suggest that the ventromedial prefrontal cortex encodes a common value signal that integrates multiple attributes of choice options and correlates with subjective preferences as revealed in behavioral choices (Bartra et al., 2013; Berkman, Hutcherson, et al., 2017; Clithero & Rangel, 2014; Hare et al., 2009, 2011; Kable, 2014; Kable & Glimcher, 2007, 2010; Krönke, Wolff, Mohr, et al., 2020).

In light of these findings, it has been suggested that self-control does not reflect the suppression of an impulsive by a reflective system but rather rests on the modulation of a common value signal by anticipated long-term outcomes. In line with this idea, it has been shown that individuals in whom value signals in the ventromedial prefrontal cortex are strongly modulated by long-term (relative to short-term) outcomes make more self-controlled choices in laboratory tasks (Hare et al., 2009, 2011) and commit fewer daily self-control failures during real-life conflicts (Krönke, Wolff, Mohr, et al., 2020). From this perspective, any process that increases the weight of future outcomes in the value-integration process (e.g., a cue-induced focusing of attention on long-term consequences) will enhance the likelihood of a self-controlled (farsighted) choice (Berkman, Hutcherson, et al., 2017; Hare et al., 2009, 2011; Kable & Glimcher, 2007; Krönke, Wolff, Mohr, et al., 2020).

Given the available evidence, it would be premature to predict which type of theory will ultimately turn out to be a more realistic account of value-based choice and self-control. With respect to the willpower paradox, the important point is that if theories postulating a common value signal should turn out to be a better account of behavioral choice (or if divided-mind theories would fail as a resolution of the willpower paradox for the reasons discussed above), the only remaining options to avoid the paradox would be to abandon either the intentionality or the synchronicity assumption and thus adopt Resolution 1 (automatic self-control) or Resolution 3 (nonsynchronous control).

Diachronic selves, control dilemmas, and the metacontrol challenge

As noted above, if the cost imposed on indulgence by a precommitment decision does not suffice to make the tempting option the less preferred option at the moment of choice, preventive self-control will be effective only if the person resists reconsideration and thereby avoids the risk of a preference reversal. To date little is known about the mechanisms and conditions that determine whether precommitment decisions do or do not become the subject of reconsideration and when previously formed goals determine action selection without a reevaluation of their subjective value (cf. Hayden & Niv, 2021).

This question relates to a more general challenge for Resolution 3 that we term the “metacontrol problem” of preventive self-control. Although Resolution 3 is compatible with the assumption of a unitary valuation system, it can also be interpreted as a special instance of a divided-mind account. In contrast to the dual-systems theories we discussed in the context of Resolution 2, which focus on synchronous conflicts between competing valuation or control systems within a person at a given moment in time, preventive self-control involves diachronic conflicts between one’s current preferences and the preferences of one’s past or future self. The function of preventive strategies can thus be considered to make binding decisions for one’s future self or stick to commitments made by one’s past self. ¹³

This interpretation of preventive self-control as a diachronic conflict between preferences of one’s current and future self makes particularly obvious why preventive self-control poses a metacontrol challenge. The reason is that successful resistance to reconsideration may lead to choices that do not reflect what one would judge to be the best option had one considered all options at the moment of choice. A central feature of self-control conflicts is that an unbiased (re)consideration of one’s current preferences would lead to a preference reversal (i.e., a self-control failure) because an immediate desire or temptation does have the highest subjective utility at the moment of choice. Consequently, when an implementation intention or a commitment that has been formed by an agent’s earlier self successfully prevents the agent from reconsidering preferences, this bears the risk of promoting overly rigid persistence even in situations in which it would be adaptive to abandon prior commitments to satisfy an important immediate desire, exploit an unexpected reward opportunity, or respond to a sudden threat. ¹⁴

Resistance to reconsideration thus confronts agents with a “control dilemma” that raises a fundamental metacontrol problem (Eppinger et al., 2021; Goschke, 2003, 2013; Goschke & Bolte, 2014, 2017). On the one hand, attaining long-term goals requires a persistent commitment and the ability to shield goals from transient changes of motivational states. On the other hand, unexpected changes of external conditions (e.g., reward contingencies) or internal states (e.g., vital needs) may afford that one abandons prior commitments and revises preferences. Agents must thus not only decide which action is best suited to attain a goal but also when it is adaptive to ignore unexpected changes, resist reconsideration, and shield goals from competing motivations and when it is better to respond to a change by abandoning prior commitments, reconsidering preferences, and exploring alternative options. Thus, preventive self-control strategies aimed at shielding goals and preventing reconsideration confront agents with the metacontrol problem of how to decide whether to stick to commitments of one’s past self even when a current desire has become a more preferred option.

Such metadecisions pose a nontrivial computational challenge. In changing and uncertain environments, it is usually impossible to constantly evaluate the expected values and probabilities of all short- and long-term outcomes of all behavioral options given limited time, incomplete information, and scarce computational resources. Moreover, forming a stable commitment to an intention functions precisely to prevent agents from reevaluating the costs and benefits of competing goals and desires to avoid a premature preference reversal, which raises the question of how an agent can decide whether to resist reconsideration without reconsidering in the first place. ¹⁵ Feasible solutions of such metacontrol problems must therefore rely on learning processes that enable agents to infer from prior experience in which contexts it is adaptive (and maximizes reward in the long run) to stick to prior commitments and when it is more adaptive to adjust preferences more myopically in light of one’s current motivational state. Initial computational-modeling studies indicate that experience-based learning in conjunction with Bayesian inference may provide a basis for solving such metacontrol problems (Markovic et al., 2021). However, we are not aware of any attempts to date to apply such models to the domain of self-control, and it will thus be a key challenge for theories favoring Resolution 3 to apply such models to preventive self-control and test them empirically.

Implications for future research

In contrast to proactive preventive self-control, reactive preventive self-control processes are intentionally recruited within a temptation episode when a long-term goal is still the most preferred option but the person anticipates that an emerging desire will at some point supersede the goal and therefore decides to down-regulate the desire. In contrast to proactive precommitment strategies, reactive preventive self-control operates at much shorter time scales and when a competing desire is already emerging. A main challenge for future research will thus be to develop experimental paradigms that allow examining the fine-grained time course of competing motivational tendencies in parallel with the recruitment of self-control processes. One promising avenue is the use of continuous measures such as mouse tracking (Dshemuchadse et al., 2013) or eye tracking (Steimke et al., 2016), which allow examining the dynamics of evidence accumulation and the integration of choice attributes. For instance, a study using mouse tracking in a dietary-choice task revealed that the tastiness of food items is processed earlier than healthfulness and that individual differences in self-control could partly be accounted for by the relative speed with which tastiness and healthfulness were processed (Sullivan et al., 2015). Another study using a multi-attribute drift-diffusion model to fit data from a dietary-choice task showed that prime stimuli, which supposedly activating dietary goals, influenced the decision process by modulating the strength and timing of taste and health attributes during the evidence-accumulation process (Sullivan & Huettel, 2021). Moreover, in a study using a computational model that sampled reward and delay information stochastically on a moment-by-moment basis it was shown that, as preferences for the choice alternatives evolved, asymmetric lateral inhibition was triggered, which accounted for the contribution of frontoparietal control regions to intertemporal choices (Turner et al., 2019). In conclusion, the combination of dynamic measures and computational models appears to be a promising approach to investigate when reactive preventive self-control is recruited during an emerging conflict and thereby to provide evidence for a key mechanism postulated in Resolution 3.