R1. Introduction
We could hardly be more pleased with the commentaries. To be sure, many scholars who offered responses found fault with some of our reasoning or ideas. Still, we were prepared for – indeed, expected – a thoroughly different flavor of response, considerably more resistant to our proposals.
As context for our expectations, consider the impact of one of the central ideas with which we were taking issue, the claim that “willpower” is a resource that is consumed when self-control is exerted. To give a sense of the reach of this idea, in the same month that our target article was accepted for publication Michael Lewis reported in Vanity Fair that no less a figure than President Barack Obama was aware of, endorsed, and based his decision-making process on the general idea that “the simple act of making decisions degrades one's ability to make further decisions,” with Obama explaining: “I'm trying to pare down decisions. I don't want to make decisions about what I'm eating or wearing. Because I have too many other decisions to make” (Lewis Reference Lewis2012).
Add to this the fact that a book based on this idea became a New York Times bestseller (Baumeister & Tierney Reference Baumeister and Tierney2011), the fact that a central paper articulating the idea (Baumeister et al. Reference Baumeister, Bratslavsky, Muraven and Tice1998) has been cited more than 1,400 times, and, more broadly, the vast number of research programs using this idea as a foundation, and we can be forgiven for thinking that we would have kicked up something of a hornet's nest in suggesting that the willpower-as-resource model was wrong. So we anticipated no small amount of stings from the large number of scholars involved in this research enterprise. These were our expectations before receiving the commentaries.
R2. The big picture
R2.1. Non-barking dogs
Our expectations were not met. Take, for example, the reaction to our claim that the glucose version of the resource argument is false (Kurzban Reference Kurzban2010a). Inzlicht & Schmeichel, scholars who have published widely in the willpower-as-resource literature, more or less casually bury the model with the remark in their commentary that the “mounting evidence points to the conclusion that blood glucose is not the proximate mechanism of depletion.” (Malecek & Poldrack express a similar view.) Not a single voice has been raised to defend the glucose model, and, given the evidence that we advanced to support our view that this model is unlikely to be correct, we hope that researchers will take the fact that none of the impressive array of scholars submitting comments defended the view to be a good indication that perhaps the model is, in fact, indefensible. Even if the opportunity cost account of effort turns out not to be correct, we are pleased that the evidence from the commentaries – or the absence of evidence – will stand as an indication to audiences that it might be time to move to more profitable explanations of subjective effort.
While the silence on the glucose model is perhaps most obvious, we are similarly surprised by the remarkably light defense of the resource view more generally. As Kool & Botvinick put it, quite correctly in our perception: “Research on the dynamics of cognitive effort have been dominated, over recent decades, by accounts centering on the notion of a limited and depletable ‘resource’” (italics ours). It would seem to be quite surprising, then, that in the context of our critique of the dominant view, arguably the strongest pertinent remarks come from Carter & McCullough, who imply that the strength of the key phenomenon that underlies the resource model – two-task “ego depletion” studies – might be considerably less than previously thought or perhaps even nonexistent. Despite the confidence voiced by Inzlicht & Schmeichel about the two-task findings, the strongest voices surrounding the model, then, are raised against it, rather than for it. (See also Monterosso & Luo, who are similarly skeptical of the resource account.)
Indeed, what defenses there are of the resource account are not nearly as adamant as we had expected. Hagger wonders if there is “still room for a ‘resource’ account,” given the evidence that cuts against it, conceding that “[t]he ego-depletion literature is problematic.” Further, he relies largely on the argument that the opportunity cost model we offer might be incomplete, thus “leaving room” for other ideas. As is evident from the other commentaries, many alternatives beyond our own might fill the space he has in mind.
Harvey, although crediting that our argument is “convincing,” suggests that a model in which depletable resources are allocated along the lines we propose is no worse than the one we advance. In the absence of a candidate for such a depletable resource, we favor our proposal, and are encouraged by the fact that Harvey offers no reason in principle to favor the depletable resource view over ours; he suggests only that the criteria that we believe apply to such models are overly “stringent.” We of course cede his larger points. That is, we don't deny either that neurons need energy to function or that deficits in neurotransmitters have genuine, important effects on performance. We feel comfortable granting these points while retaining our opportunity cost view.
Bonato, Zorzi, & Umiltà (Bonato et al.) defend the view that the idea of “strictly depletable” resources is “the most economic explanation” for a set of findings in which brain-lesion patients demonstrate performance deficits when multi-tasking. Related, but departing from this “strict” view, Hofmann & Kotabe seem to favor what could be called a “husbanding” view, writing that “certain executive functions cannot be exerted infinitely without a state reduction in executive capacity,” so “people are motivated to monitor and conserve capacity.” We resist the husbanding view for reasons discussed in the target article.
On a similar note, Brzezicka, Kamiński, & Wróbel (Brzezicka et al.) consider a version of a resource account, but their version departs from the traditional model in at least one important way, as they construe the resource as neurophysiologically local, rather than the sort of general resource originally proposed. Their view refers to the possibility that what is being depleted are neurotransmitters or neuromodulators. Related, Holroyd proposes that the anterior cingulate cortex (ACC) regulates impulsive behavior “via an energy factor that depletes with use.” We believe that there are difficulties with these “local resource” views, and we address these ideas in more detail below.
More commonly, commentators have defended alternative models that explain subjective effort not by exhaustion of depletable resources, but rather by reference to certain tasks being inherently difficult and/or certain cognitive processes carrying intrinsic costs. Many of these alternative models refer in general terms to resources, energy, or capacity that are demanded by some mental tasks and that people are motivated to conserve. For example, both Gendolla & Richter and Wright & Pantaleo assume that effort is something that is expended in proportion to task difficulty or task demands. Hofmann & Kotabe similarly talk about a capacity that is exerted in proportion to task demands, and Huizenga, van der Molen, Bexkens, & van den Wildenberg (Huizenga et al.) propose a model in which resources are allocated to tasks in proportion to task difficulty. Hennecke & Freund argue that “subjective effort is a function of the resources a person perceives to invest into the pursuit of the target goal in relation to the subjectively available goal-relevant resources.” In all of these cases, people are averse to effort and avoid it if possible – they are motivated to use as little of their resources, energy, or capacity as possible.
Two other commentaries are more specific in locating the intrinsic costs of certain kinds of mental activity. Navon argues that cognitive operation of “decoupling” is inherently aversive; Kool & Botvinick argue that cognitive control carries intrinsic costs.
Below, we discuss some of the specific details of each of these commentaries, including clarifying our definition of effort and the fact that effort is both the output of some computations and an input to others. Relevant to all of these commentaries, however, is that we disagree with the notion that costs are intrinsic to certain kinds of mental activity or that difficulty is inherent to certain kinds of tasks. This kind of alternative model, in our view, does away with the notion of depletable resources, but fails to provide in its place any explanation for why certain kinds of mental activity are effortful. Our opportunity cost model is an attempt to provide such an explanation – that cognitive processes are costly or aversive to the extent that employing these processes carries substantial opportunity costs. Whether or not our model turns out to be the correct one, alternatives to the resource account must provide some explanation for why certain cognitive processes are costly, including the currency of the putative costs.
So, while there are traces of evidence of defenses of variants of the resource view, by and large these defenses are relatively mild and relatively rare. From this somewhat puzzling state of affairs – the contrast between our sense that many scholars in particular communities take the resource account more or less for granted, and the anemic defense of the account – we conclude that outside of the community of researchers currently working on this model, there is little appetite for a defense of it.
Related, but perhaps not as surprising, few commentators have taken serious issue with the assumptions that underlie our approach, especially the idea that this puzzle will be solved by invoking the language of computation and the evolutionary principle of function. While there are traces of resistance to these ideas in the comments, by and large the flavor of the remarks reflects an acceptance of our claim that these ideas will be useful in trying to understand and explain both the phenomenology and the behavioral data.
The context here is a disconnect between the literatures we are engaging, the “ego depletion” literature, on the one hand, and the vigilance literature, on the other (cf. Malecek & Poldrack). In the former, the language of cognition is nearly absent, with models built from metaphor – reservoirs, resources, and so on – whereas in the latter the building blocks of explanations are computational. By and large, members of the latter community we understood to welcome the idea that the language of computation ought to be brought to bear on the phenomena in the “ego depletion” literature. Recent thinking on decreases in performance on vigilance tasks over time is resonant with our approach, in particular in characterizing the vigilance decrement as a rational response to a task that demands sustained attention without rewarding attention (or punishing inattention) (Hancock Reference Hancock2013).
An exception is found in Iran-Nejad & Zengaro, who say that “the computation metaphor, if used for other than a mathematical tool of science, is an Achilles heel.” We find their example of where a computational account fails – explaining variation on moral judgments in the famous Trolley Problem (Greene & Haidt Reference Greene and Haidt2002) – ironic. In our view, Mikhail's (Reference Mikhail2007) formidable account of this variation represents a signal example of how the application of computational language can illuminate and explain previously puzzling patterns of data.
Given the controversy that continues around evolutionary approaches to psychology (Pinker Reference Pinker2002), we were surprised and gratified to find so little resistance to that element of our approach. The only serious worry along this front comes from Cohen & Saling, but their objection is founded on an unfortunate misunderstanding. Our claim is not that “being a utility-maximiser is adaptively optimal,” as they have rendered our view. Instead, our claim is that evolution selects for systems that guide adaptive behavior, and that the language of utility and maximization models are useful in the context of building computational models (cf. Cosmides & Tooby Reference Cosmides and Tooby1994). Outside of these brief worries, little mention is made of this key assumption, despite our expectations, themselves derived from the ambiance of debate that surrounds evolutionary approaches to psychology.
Similarly, there is very little objection to the general conclusions we reached from a summary of the neuroscience literature. As we have indicated above, no one rose to defend the idea that global levels of brain glucose serve as a resource that limits performance in mental tasks. No one questions that the current state of the neural evidence was broadly consistent with a cost-benefit type of account. As we discuss in section R6, several commentaries have made alternative proposals regarding the specific computational role that different neural systems might play in a cost-benefit framework, and a few have proposed hypotheses regarding other potential resources, but our basic framework has not been fundamentally questioned.
For the sake of completeness, we add that our assumptions surrounding phenomenology, surely a subject on which there is no shortage of diverse and strong opinion, has met with curiously little objection (but see Craig). To be sure, some commentators raise the issue of the extent to which the processes we have in mind are conscious versus non-conscious (see next section), but by and large we find it remarkable that so little attention was paid to what might have been a very basic objection to the worldview (Cosmides & Tooby Reference Cosmides, Tooby, Lewis and Haviland-Jones2000) that enrobes our proposal.
R2.2. Terms and assumptions
As is frequently the case in scholarly debate, some of the disagreements derive from differences in the meanings attached to terms and non-shared assumptions. In this subsection we discuss several such cases, with an emphasis less on resolving the disputes – people are of course free to use terms to mean whatever they wish – and more on clarifying our own uses and commitments.
R2.2.1. Terms
First, rational and the related term rationality have consistently posed conceptual challenges. To clarify, we resist the notion that our proposal should be viewed as suggesting “a mechanism that rationally allocates processors” as Harvey renders it. An even grosser mischaracterization of our view is Inzlicht & Schmeichel's assertion that our model “assumes that people calculate costs and benefits in an objective, dispassionate manner.” We made no claim about what “people,” broadly, do, let alone what passions influence their calculations; rather we made a proposal, narrowly, about the causal variables that underlie the phenomenology of effort and decisions in the context of a particular set of tasks. Our proposals are no more an endorsement of homo economicus, broadly, than are ideal observer models or optimal foraging models, as we indicated in the target article.
Relatedly, our claim is neither that “[people's] most basic motivation is to maximise utility,” nor that “being a utility-maximiser is adaptively optimal” (Cohen & Saling). The use of the language of utility might have, reasonably, recruited readers’ sense that our assumptions echoed those of economists, which is why we tried to be explicit about our assumptions in the target article. Our claim is that natural selection tends to fashion systems whose properties can be modeled as maximizing, as is frequently done in literatures ranging from visual perception (Simoncelli & Olhausen Reference Simoncelli and Olhausen2001) to foraging (Charnov Reference Charnov1976). This does not entail that maximizing utility is either the most basic motive, or that doing so is necessarily optimal.
In short, our claim was not one of rationality, which we would take to be a strong one, but rather a weaker claim that the explanation for the phenomena in which we are interested is to be located in the conceptual primitives of costs and benefits (more about which below). Mechanisms can operate in virtue of cost/benefit calculations while departing (systematically) from predictions derived from a normative model of rationality. Indeed, the only time the word “rational” appears in the target article is when we are describing our starting point for the enterprise, and in that passage we hastily assert that the cognitive mechanism in question might depart from rationality.
The terms effort and motivation are similarly potentially contentious. In our article, we tended to use the word “effort” in the context of our own model, as an aspect of phenomenology, as in the titular use of “subjective effort” – though, as Huizenga et al. point out, we ourselves were not entirely consistent. Still, we are uncertain what to make of it as a dependent measure, as in Figure 1 of Gendolla & Richter's commentary. From their later remarks that “the energy conservation principle is that organisms do not invest more resources than necessary for an action,” and that “motivational intensity theory posits that effort rises with subjective demand as long as success is possible and justified,” we take Gendolla & Richter to be equating effort and energy, an equation we would strenuously resist, as should be clear from the target article.
In addition, while our model is explicitly concerned with explaining phenomenology and performance during mental tasks, we readily concede Monterosso & Luo's point that the constructs of self-control and executive function are not synonymous. We conceptualize self-control as behavior consistent with valued long-term goals at the expense of less valued but more immediately attainable goals (i.e., temptations). Certainly, there are many instances in which deploying executive function in the service of a subjectively valued long-term goal (e.g., completing a manuscript) conflicts with using the same computational processes to attain a less valued but more immediate alternative (e.g., checking email). However, the exercise of executive function need not entail such a conflict. That is, executive functions are often used to manage lower-level computational processes in the absence of temptation. And, while executive function certainly facilitates self-control, so, too, do a variety of metacognitive strategies (e.g., commitment, selective attention, and psychological distancing).
Finally, we note that while different communities use the term “motivation” in multiple ways, we take our model to be a motivational model, with the opportunity cost calculation being a causal antecedent of the deployment of computational mechanisms; on our view, such causal pathways are the essence of “motivation.”
R2.2.2. Assumptions
As indicated above, we are gratified that many of our assumptions have gone unchallenged. Of course, not all of them were. Huizenga et al. reject our assumption that an important opportunity cost calculation stems from “daydreaming,” which they deny requires executive functions. We stand by our broader claim that many of the sorts of things that subjects might do when they release their attention from vigilance or self-control tasks do recruit executive functions, though of course we are open to evidence on this point. (For opposing views on this issue, see Smallwood & Schooler Reference Smallwood and Schooler2006, and, in response, McVay & Kane Reference McVay and Kane2010.)
This relates as well to Charney's doubts about our assumption that tasks cannot be inherently boring (or exciting, etc.). We would defend our assumption. We don't take boringness as something that inheres; rather, we take boringness to be a relationship between a nervous system and a task. We would make the same argument with respect to task “difficulty” (Bonato et al.), how “demanding” a task is (Gendolla & Richter), and “monotony” (Prudkov). Consider the computations involved with recovering a three-dimensional image from retinal data; computationally, this is a terribly difficult challenge (Marr Reference Marr1982). Yet, it is in no interesting sense “difficult” for people (with normal vision) to see. What is difficult (boring, monotonous) depends on the arrangement of the nervous system. That is, some tasks are difficult for our nervous system, and some are easy. We therefore resist the notion that difficulty – or boredom, and so on – inheres to tasks. In short, we take boringness as a relationship between a subject's mind and a task, and as such, a relationship to be explained, rather than an inherent property of a task that plays an explanatory role.
We take a similar position on the notion that executing tasks that require cognitive control carry “intrinsic” costs (Kool & Botvinick). That is, we resist the notion that costs are the sort of thing that can be “intrinsic.” We are of course sympathetic to the notion that cost computations accompany the sorts of tasks Kool & Botvinick have used, but we prefer to think of these cost representations as computational outputs to be explained, as opposed to “intrinsic” properties of the tasks or computational processes.
We reiterate that the sensation of effort, according to our view, depends on the systems recruited by the task in question. So, contrary to Charney's view, even if the identical alternative tasks are available to two subjects, they will experience different sensations of effort if the two tasks recruit different computational mechanisms. It is the alternative uses to which recruited computational mechanisms, together, can be put that, according to our proposal, produce the experienced sensation of effort.
R3. Critiques of the opportunity cost model
A number of commentators have granted many or most of our assumptions but have taken issue with various elements of the substance of the proposal. In this section we address some of these challenges, with an emphasis on computational inputs/outputs, potential alternative conceptualizations, and our interpretation of some of the results to which commentators have drawn our attention.
R3.1. Inputs and outputs
According to our proposal (see, e.g., Fig. 1 in the target article), sensations of subjective effort should be understood to be the output of mechanisms computing opportunity costs, as well as inputs to decision processes designed to guide adaptive behavior with respect to the decision regarding whether to continue to pursue the present task. This dual role corresponds to our conception of phenomenology in other domains. For example, fear is both the output of a set of mechanisms monitoring risks in the environment (e.g., predators, violence, dangerous heights) and an input to mechanisms designed to motivate adaptive avoidance. Similarly, hunger is both an output of a set of mechanisms monitoring energetic requirements and an input into decision mechanisms that motivate food search and consumption behaviors.
We noted this dual nature of subjective effort in multiple instances. Despite our attempts to be explicit about our commitments, as when we suggested that “sensations are the outputs of mechanisms designed to produce inputs to decision-making systems” (sect. 2.3.2, para. 1), some commentaries reflect a certain amount of confusion on this point. For example, Hennecke & Freund suggest an alternative model whereby “[r]ather than being an output of computations that compare costs and benefits of the target and competing goals, effort enters these computations as an input.” Hofmann & Kotabe similarly misconstrue our model of subjective effort as reflecting only the output of opportunity cost–monitoring mechanisms, suggesting that we take into consideration the idea that “subjective task effort may also enter as a cost input into the cost-benefit analysis that underlies the utility calculation of each activity involved.” We, of course, agree.
Related to the point above, we did not intend to give the impression that opportunity cost computations would be the only input to decision-making systems. Molden helpfully points out numerous plausible inputs, including voluntarily chosen (vs. coerced) task engagement and beliefs about the source of experienced effort. Similarly, Harrison & McKay point out that religious priming could potentially change the experience of effort. We agree that beliefs (as Molden puts it, “top-down orientations”) about the world should affect both experienced subjective effort and motivation to persist. Beliefs – and even metabeliefs (cf. Hofmann & Kotabe) – could do this in multiple ways, including altering the perceived probability of success, altering the perceived utility of task completion, and altering the perceived utilities of alternative activities. The number of inputs is likely substantial – and certainly would include computations surrounding reputational effects of the sort that Harrison & McKay allude to – and, given our ability to arbitrarily coin reward (see Ainslie), potentially a great deal more.
R3.2. Necessity and sufficiency
A number of commentaries have endorsed elements of our approach while taking issue with the details of our proposal. For instance, Kool & Botvinick broadly endorse our approach, in particular the “motivational turn” in the allocation of mental processes and our emphasis on a “value-based perspective.” Along these same lines, several commentaries emphasize the fields of behavioral economics and neuroeconomics, in particular as sources of methods for testing our model. Westbrook & Braver note that the model might be “untestable without objective cost measures.” We agree, and think this points to a substantive advantage of cost-benefit models over resource-based models, which seem difficult to falsify (Navon Reference Navon1984). Indeed, we are currently using methods drawn from behavioral and experimental economics to test various aspects of the model.
Still, despite agreement about the value of a motivational approach, broadly, several commentaries suggest that our proposal surrounding opportunity costs is unlikely to be able to account for all cases of subjective effort (Ainslie; Gendolla & Richter; Hennecke & Freund; Kool & Botvinick; Molden). We welcome these comments, and we agree that we failed to draw an important distinction in our discussion of opportunity cost. First, because executive systems can be put to use for activities (e.g., planning) more or less independent of opportunities afforded by the environment, there might be opportunity costs that inhere whenever these systems are put to use. Second, potential opportunities in the environment – using a cell phone, for example – represent a second, distinguishable opportunity cost. We are committed to the view that the first sort of opportunity costs explains why tasks that recruit executive systems are perceived as effortful even in environments in which there are no tempting alternative activities. (See Harrison & McKay for a dissenting view.)
Ainslie similarly endorses the “motivational turn” (as does Molden) in explaining subjective effort, but argues that opportunity costs are unnecessary to explain subjective effort, preferring instead the notion of “endogenous reward.” We are broadly sympathetic to this notion, and suspect that our perspectives might be compatible. For instance, Ainslie suggests that there exists a “baseline level of reward that does not depend on external contingencies,” and that when idle (e.g., daydreaming), individuals appear to generate their own rewards, potentially even arbitrarily. This echoes examples we used in the target article of mind wandering and daydreaming as potentially valuable alternatives precluded by use of executive function–related processes.
Navon also raises the question of how people can “rank on-line, albeit implicitly, the costs/benefits of all alternatives (or even the most salient ones) sufficiently for estimating opportunity costs.” Navon also raises the important point that some sources of distraction (e.g., a flying bird stuck in an office) make it hard to maintain executive function–related processes but represent very low or even zero opportunity cost.
We think this is a critical question, which we did not address for reasons of space, though we do not think the problem is insurmountable. Under conditions of uncertainty and incomplete information, people must perforce estimate the expected value of alternatives – such estimates need not be perfectly accurate to be useful guides to behavior. In the case of the trapped bird, our guess is that the formal properties of the stimulus, by and large, are the sorts of things these systems were designed to attend to. Historically, nearby animals or objects – especially fast-moving ones – were sources of threat or opportunity, and it seems reasonable to design the mind to value attending to them, even if not every instance results in an incurred cost or benefit.
R3.3. Devil in the details (of costs/benefits)
Other commentaries, while also broadly in agreement with the cost-benefit approach, take issue with specific details of the nature of costs/benefits in task persistence and task switching. Nicolle & Riggs suggest that the anticipated regret of switching might bias people toward staying on task. We find this plausible, and, as we have indicated above, concede that opportunity costs are only one input into the decision-making systems that govern task switching.
Inzlicht & Schmeichel similarly agree with the starting point of costs/benefits – “[s]ome version of this view seems likely to be correct” – but argue that our model failed to capture the dynamics of costs/benefits over time. We respectfully disagree. In the target article we argued that people acquire information about the value of a task as they perform it, accounting, we think, for phenomenological and performance dynamics. On this note, we take seriously the point that feedback is important to task persistence, as is information about beliefs about whether task success is possible in the first place. Gendolla & Richter cogently make this second point, noting that we failed to distinguish task choice from task execution, which, they argue, led us to neglect task demand. We broadly agree that efficacy is an important consideration and might be another input into decision-making systems. The way we think about this issue is that representations of one's ability to achieve success on a task enter into the expected benefit computation of continued deployment of computational resources on the task in question. As indicated above, we take “task demand” to be a relationship between the person doing the task and the details of the task, as opposed to a property of the task in itself.
Relatedly, Bruyneel & Dewitte, while sympathizing with our broad approach, note that cost/benefit computations might themselves depend on executive function–related processes. This carries the entailment that the quality of cost/benefit calculations depends in part on how much executive function–related processes are devoted to some task other than cost/benefit calculating. We certainly allow that this is a possibility, though our views veer more toward Ainslie's and Kool & Botvinick's. That is, we think it plausible that executive processes might not be required for the sorts of opportunity cost computations we have in mind, though we take Bruyneel & Dewitte's point about the empirics. Still, in terms of the particular empirical pattern they point to – the two-task paradigm – as Carter & McCullough indicate, these results might not be as robust as previously believed (Hagger et al. Reference Hagger, Wood, Stiff and Chatzisarantis2010a), inclining us toward caution about what to infer from this line of work. The suggestion Bruyneel & Dewitte offer regarding the interpretation of these results is intriguing, however; and we look forward to additional work disentangling alternative explanations.
R4. Individual and group differences
Several commentaries raise the issue of individual or group differences, either in the sustained performance of mental tasks over time or in the concomitant phenomenology associated with performing these tasks. For example, Hagger points to evidence in his meta-analysis (see Hagger et al. Reference Hagger, Wood, Stiff and Chatzisarantis2010a) that individual difference variables moderate the effect sizes observed in “ego depletion” experiments, suggesting we elaborate how individual difference variables could “bias individuals’ tendency to interpret the opportunity costs of their responses relative to the next most desirable action.” We agree that any complete account of subjective effort should explain both changes within individuals over time as well as between-individual differences in these trajectories. Indeed, exploring the extent to which our model can explain individual and group differences constitutes an important direction for future research. Our hope is that our necessarily brief remarks on this topic highlight useful first steps toward that end. Extending the logic of our model, individuals might vary (a) in their valuation (i.e., implicit estimation of the expected value) of the target task, (b) in their valuation of the next-best alternative task to which the same cognitive processes may be deployed, and (c) in how they appraise or interpret feelings of subjective effort.
Why might individuals derive different benefits from performing identical mental activities? As Westbrook & Braver point out, it is an empirical fact that people vary in their willingness to perform tasks that generate feelings of mental effort (see also Cacioppo & Petty Reference Cacioppo and Petty1982). Likewise, for sensation-seeking individuals, mental tasks might be perceived as higher in value to the extent they are novel or unpredictable and lower in value to the extent they are repetitive and monotonous. More conscientious individuals might assign higher value to task performance to the extent that their standards for performance are higher throughout the task. Likewise, more compliant or agreeable individuals may assign greater benefits than others might to fulfilling an experimenter's expectations.
Estimates of opportunity costs might also differ between individuals. For instance, some of us might be more inclined to daydream (e.g., to remember past events, to prospect into the future), an activity that conflicts with task-oriented processing. Opportunity cost estimates might also differ as a function of how vividly individuals tend to generate counterfactuals, whether consciously (Frederick et al. Reference Frederick, Novemsky, Wang, Dhar and Nowlis2009) or, as we have argued, implicitly. Moreover, opportunity costs or sensitivity thereto might vary across individuals because of differences in cognitive processing capacity. As Westbrook & Braver point out, in a willingness-to-pay paradigm older adults must be paid more to engage with challenging mental tasks than younger adults. Likewise, limitations in cognitive processing capacity might explain the evidence summarized by Bonato et al. In their experiments, brain-damaged patients showed increasingly compromised performance in one task, detecting visual stimuli in the periphery, as the demands of a second task performed simultaneously increased. Healthy controls did not show this decrement. Brain-damaged patients likely have more limited cognitive processing capacity, and as more of this limited (but, in our view, not depletable) capacity is allocated to the second task, performance on the first task starts to fail.
Finally, individuals might differ in how they interpret the sensation of subjective effort. For instance, people who believe that mental resources are depletable might construe such qualia as evidence they are “running out of willpower.” Such an attribution might incline these individuals to give themselves a break, allocating computational processes away from the target task to some easier alternative (see Clarkson et al. Reference Clarkson, Hirt, Jia and Alexander2010; Reference Clarkson, Hirt, Austin Chapman and Jia2011; Job et al. Reference Job, Dweck and Walton2010). In contrast, individuals might learn to interpret feelings of subjective effort as indicating their willpower is being challenged, an inference which, when experimentally induced, improves performance (Magen & Gross Reference Magen and Gross2007). Similarly, individuals might infer from the experience of subjective effort that they are engaged in something worthwhile (i.e., that reward is imminent), perhaps from the repeated association of subjective effort with eventual reward (Eisenberger Reference Eisenberger1992). Or, individuals well-practiced in mindfulness meditation might be more aware of feelings of subjective effort yet be less inclined to react to them (see Holzel et al. [Reference Holzel, Lazar, Gard, Schuman-Olivier, Vago and Ott2011] for a review).
R5. Why do people persist?
We discern two concerns regarding our model's predictions of task persistence. First, why don't people persist indefinitely on mental tasks whose value exceeds that of all other possible uses of the same computational resources? Second, why do people persist at all on tasks that provide no immediate reward when immediately pleasurable alternatives (i.e., temptations) are available? We believe that our proposal addresses both of these concerns: Persistence depends on favorable cost/benefit calculations of the task relative to alternative uses of the same computational processes.
Considering the first question, Hagger worries that “in the absence of ‘next-best’ tasks, task persistence will be indefinite, which seems unfeasible” (see Hagger's Abstract). Likewise, Hoffman & Kotabe suggest that in our model, “because effort is treated as the result of a relative utility comparison of opportunity costs, people should go on almost infinitely (experiencing virtually no effort) pursuing a cognitively demanding option A when the value of this option is very high and no alternative option B comes close in utility.”
Our position is that, indeed, some individuals do sometimes persist on mental tasks for hours and hours, stopping perhaps only to attend to basic physiological needs such as sleep or food, when these activities yield consistently greater utility than next-best alternatives. As an extreme case, so-called idiots savants (to use Michael Howe's term) are known to devote hours and hours to a single task, reaching exceptional levels of performance despite subnormal general intelligence (Howe & Smith Reference Howe and Smith1988). Furthermore, our specific claim is that a monopoly on computational processes can be sustained over time without feelings of subjective effort when the perceived benefits are also maintained over time. The possibility of sustained, “effortless” mental activity over time is consistent with the substantial literature on flow, the subjective state of being completely involved in a challenging, intrinsically rewarding activity “to the point of forgetting time, fatigue, and everything else but the activity itself” (Csikszentmihalyi et al. Reference Csikszentmihalyi, Abuhamdeh, Nakamura, Elliot and Dweck2005, p. 600).
It is also true, however, that flow is rare. We see two reasons for its rarity. First, over time, there is almost always some alternative activity that becomes more valuable than the task at which we have been laboring. Moreover, for most of us, most of the time, there are diminishing marginal returns for mental tasks. Speaking for ourselves, we have rarely sat down to write a paper and found that our good ideas kept flowing at full force hour after hour after hour. As the rate at which our good ideas come diminishes (and may even become negative insofar as we end up mangling previously well-argued passages), the relative value of alternative activities increases. Instead of making (slower and slower) progress on our paper, we could, for instance, answer email, talk to colleagues, make plans for dinner, and so on. Our view is that the estimation that we could better use our limited cognitive processing capacity for better ends leads to feelings of fatigue and restlessness, prompting us to direct our attention to a rival task of greater expected utility.
Turning to the second question, why people persist at all, Hillman & Bilkey wonder how our model might explain the incontrovertible fact that animals (e.g., scientists) persist through subjectively effortful tasks over extended time periods (e.g., “months of executive function–demanding writing, research, and teaching”; Hillman & Bilkey) when alternatives (whose varieties are all too familiar to the present readership) present significant benefits. In a similar spirit, Inzlicht & Schmeichel suggest that our model does not adequately explain “why people sometimes engage in seemingly costly and effortful behavior following periods of high subjective effort; for example, going to lengths to aggress against others or to find and consume drugs.” As Zayas, Günayadin, & Pandey (Zayas et al.) point out, the benefits of common effortful mental tasks (e.g., working on math homework) are often abstract and delayed in time, whereas temptations (e.g., texting) are associated with immediate, salient rewards.
We thank Hillman & Bilkey for, in essence, answering the same question they pose. They reason that when individuals persist on a task whose benefits are removed in time, “the discounting of the primary goal that normally occurs under conditions of temporal distance, uncertainty, or exertion, could be attenuated during the cost/benefit evaluation. Alternatively, the degree of discounting of competing tasks could be increased.” We could not agree more. Persisting at a task that provides delayed rather than immediate benefits is facilitated by mentally representing the task as valuable, either because it is part of a valued pattern of behavior or because its execution is instrumental to some superordinate goal (Rachlin Reference Rachlin2000). Likewise, resisting a temptation is facilitated by diminishing its subjective value, for example, by using attentional or reappraisal strategies (Magen & Gross Reference Magen, Gross, Hassin, Ochsner and Trope2010; Mischel et al. Reference Mischel, Shoda and Rodriguez1989).
R6. Neuroscience of resources and motivation
In the target article, we reviewed the relevant neurophysiological evidence and concluded that the current state of the evidence is consistent with a cost-benefit account of subjective effort, but not a resource account. In particular, the evidence is inconsistent with the idea that the depletion of global levels of brain glucose leads to reductions in performance of mental tasks and the feeling of subjective effort (Kurzban Reference Kurzban2010a), and no evidence has yet been mustered for any other proposed resource. In contrast, many of the elements necessary for a cost-benefit account have been established, including neural representations of costs and benefits, neural systems involved in executive function whose engagement entails substantial opportunity costs, and potential mechanisms through which the former can influence the engagement of the latter.
Although several commentaries have addressed the neuroscience section of the target article, very few have challenged these general conclusions. Rather, the commentaries focus on two areas: (1) alternative proposals for a physical resource whose depletion leads to the sensation of subjective effort, and (2) alternative proposals regarding the specific computational role that particular neural systems play in a cost-benefit account.
R6.1. Alternative proposals for a physical resource
Three commentaries offer alternative proposals regarding a physical resource whose management constrains subjective effort and performance. In the target article, we allowed that there might be other candidate resources beyond glucose, though we knew of no explicit proposals. Brzezicka et al. provide one, arguing that the pool of readily releasable neurotransmitters might be a local resource that gets depleted with continued activity in the same circuit. In the target article we listed several questions that any resource proposal would need to answer, and Brzezicka et al.'s “local resource depletion hypothesis” does not yet seem to address the one that is perhaps most central: Why do some forms of mental activity but not others feel effortful? For example, why does the fairly continuous operation of the visual system during waking hours not run afoul of this constraint? Why is doing math problems effortful, but not watching a sunset? Maintaining a pool of available neurotransmitters would seem to be a constraint that affects all neural circuits, not just the ones whose activity is associated with the phenomenology of subjective effort. Absent a satisfying answer to such questions, we remain very skeptical of local resource explanations.
Holroyd provides a more general defense of the resource position, arguing that resource and computational accounts are not mutually exclusive. We do not disagree, though we see little reason to explore hybrid accounts if computational accounts prove sufficient (Navon Reference Navon1984). Holroyd also states that: “Doubts about glucose utilization notwithstanding (Schimmack Reference Schimmack2012), mental costs must reflect in part the simple fact that the brain is a biophysical system that obeys thermodynamic laws.” Again, we do not dispute that the brain is a biophysical system that obeys thermodynamic laws, nor do we dispute that the design of the brain is at some level constrained by energetic concerns (Lennie Reference Lennie2003; Montague Reference Montague2006b). Where we disagree is with the notion that these principles are important to (or possibly even relevant to) explaining the phenomenology of subjective effort and task performance changes. The brain consumes a large amount of energy at “rest” and the amount of energy consumed does not increase dramatically for different kinds of mental tasks. The largest local changes in energy consumption that we know of are in the visual cortex when one opens one's eyes, yet that activity is not generally perceived as effortful. Thermodynamic principles certainly provide a general constraint on brain design, but we see no evidence that energetic differences are the primary factor driving transitions between different patterns of neural activity.
As we understand their proposal, Tops, Boksem, & Koole (Tops et al.) argue that the energetic resource being conserved is not in the brain at all, but rather in peripheral systems. They argue that novel and unpredictable environments disrupt predictive homeostatic regulation and thereby engender physiological costs that need to be monitored, and that the accompanying mental state of “reactive control” feels effortful. This proposal accords with the evidence that to the extent that subjectively effortful mental tasks do in fact consume additional glucose, then the relevant increases are in the periphery rather than the brain. (See Gibson Reference Gibson2007, for a discussion of this claim.) However, there are several aspects of this proposal that are not entirely clear to us. Is “reactive control” effortful even when the physiological costs being monitored do not increase? Or can the subjective effort of monitoring and the amount of peripheral energy consumption be dissociated? If the latter, then the proposal seems to depart in important ways from a “depletion” framework. We also question whether the sensation of subjective effort only occurs in novel or unpredictable environments, which seems to be an important prediction of this hypothesis. We look forward to increasingly specific proposals regarding the possibility of a role for a physical resource so that such proposals can be empirically distinguished from the sorts of computational accounts that we favor.
R6.2. Alternative mappings of a computational account of neural systems
In the target article, we argued that several elements necessary for a neural implementation of our opportunity cost model, or cost-benefit computational accounts generally, had already been established. This includes neural representations of costs and benefits of the type required, neural systems involved in executive functions whose engagement entails opportunity costs, and potential mechanisms by which the former can influence the deployment of the latter. This last piece was by necessity speculative given the limited number of studies on subjective effort to date in the neuroscience literature. Several commentaries have added alternative proposals regarding neural systems that influence task engagement based on costs and benefits. We are excited by this wealth of hypotheses, seeing it as a sign of the promise of cost-benefit computational accounts generally, and look forward to the future studies necessary to distinguish among them.
In the target article, one of the linking mechanisms we proposed involved the anterior cingulate cortex. This region, because it both contains neural representations of costs and benefits and is engaged during executive tasks, is well placed to influence task engagement based on opportunity costs. We pointed to several studies that had reported anterior cingulate activity that was lower after performance of a mentally demanding task. We proposed that the higher activity during early task engagement might represent an opportunity cost signal, which subsequently led to task disengagement and declines in performance (Boksem et al. Reference Boksem, Meijman and Lorist2006; Inzlicht & Gutsell Reference Inzlicht and Gutsell2007; Lorist et al. Reference Lorist, Boksem and Ridderinkhof2005). Both Holroyd and Hillman & Bilkey argue for an alternative interpretation of this pattern, which essentially inverts the valence of the anterior cingulate signal. In their view, higher anterior cingulate activity early leads to continued persistence on the task, and activity declines later when people have disengaged from the task. These commentators argue that high levels of anterior cingulate activity signal that the benefits of the current behavior outweigh the costs, and this activity motivates or energizes continued persistence on the current task. We think a simple comparison between early and late performance does not discriminate yet between these alternatives. What is needed is either a trial-by-trial comparison, where our proposal predicts that high anterior cingulate activity on one trial will predict decreased performance on the subsequent trial, or an analysis of individual differences, where our proposal predicts that individuals with the highest anterior cingulate activity will show the largest decrements in performance. Holroyd and Hillman & Bilkey would predict the opposite in these two cases. Alternatively, human studies that examine the effect of anterior cingulate disruption on subjective effort and performance could distinguish the two accounts. We think it is a positive aspect of our theory – and of the cost-benefit framework more generally – that it generates further empirical predictions that remain to be tested.
Both Craig and Tops et al. point to the insula as an alternative region that might encode the costs of mental activity and act to promote disengagement from the current task when these costs are high. As these commentators nicely summarize, the insula receives information about bodily states and has long been implicated in emotion and subjective feeling. As Craig notes, the insula is also linked to the anterior cingulate, and the two regions are often coactive – including in response to feedback (Bartra et al. Reference Bartra, McGuire and Kable2013). We think it is quite plausible, then, that both regions could play a role in promoting changes in task engagement based on costs and benefits.
The second linking mechanism we proposed in the target article was through dopaminergic projections from the brainstem to the lateral prefrontal cortex. Malecek & Poldrack propose another brainstem neuromodulatory system, the noradrenergic system, as a candidate for this role. This is another plausible hypothesis, which comports with early evidence that subjective effort is associated with increases in pupil diameter (Kahneman & Beatty Reference Kahneman and Beatty1966), which are now presumed to be mediated by noradrenergic signals (Nassar et al. Reference Nassar, Rumsey, Wilson, Parikh, Heasly and Gold2012). We look forward to continued research that distinguishes the precise computational roles played by different neuromodulatory systems.
R7. Conclusion
In summary, while there are many points of contention and substantial empirical and conceptual issues to be resolved in the years ahead, we wish to return to where we began – the substantial and, to our thinking, surprising overlap in views surrounding how to go about understanding and explaining the subjective sense of effort that arises during certain kinds of mental tasks. In the recent history of this literature the prevailing view has been, at least within certain communities, that subjective effort and task performance reductions could be explained, at least in large part, with reference to diminishing resources. Although there are traces of continued embracing of such a view, our overall impression of the commentaries is that the bulk of scholars are open to, if not enthusiastic about, the computational approach that we favor.
To be sure, there remains disagreement about the details, but we feel that there is, with some important exceptions, relatively widespread agreement with (or conspicuous lack of criticism of) several propositions that we advanced in order to try to explain subjective effort:
-
1. A resource account is unlikely to be correct (in particular, the glucose version of a resource account).
-
2. Computation and function are necessary components of the explanation of subjective effort.
-
3. A cost-benefit approach is a promising general framework to understand the computations that underlie subjective effort.
-
4. Subjective effort should be understood as a motivational phenomenon.
We look forward to future work that illuminates questions that remain open, including the nature and details of the computations that underlie mental effort, the neurophysiological structures involved, and, of course, whether computations of opportunity costs play the sort of central role that we propose. Our hope is that our proposal will serve to focus debate on these open questions. While resource models have stimulated substantial amounts of research effort, our hope is that by moving beyond resource accounts, further progress can be made in understanding the origins and function of sensations of mental effort. We believe that situating this work in the context of evolved function and the language of computation might go some way towards giving the various communities working on this important question common ground from which to operate and collaborate productively in the years ahead.
Target article
An opportunity cost model of subjective effort and task performance
Related commentaries (30)
An addition to Kurzban et al.'s model: Thoroughness of cost-benefit analyses depends on the executive tasks at hand
An expanded perspective on the role of effort phenomenology in motivation and performance
An interoceptive neuroanatomical perspective on feelings, energy, and effort
Beyond dopamine: The noradrenergic system and mental effort
Beyond simple utility in predicting self-control fatigue: A proximate alternative to the opportunity cost model
Can tasks be inherently boring?
Competing goals draw attention to effort, which then enters cost-benefit computations as input
Depletable resources: Necessary, in need of fair treatment, and multi-functional
Difficulty matters: Unspecific attentional demands as a major determinant of performance highlighted by clinical studies
Effort aversiveness may be functional, but does it reflect opportunity cost?
Effort processes in achieving performance outcomes: Interrelations among and roles of core constructs
Formal models of “resource depletion”
Give me strength or give me a reason: Self-control, religion, and the currency of reputation
Is ego depletion too incredible? Evidence for the overestimation of the depletion effect
Local resource depletion hypothesis as a mechanism for action selection in the brain
Maximising utility does not promote survival
Mental effort and fatigue as consequences of monotony
Monotonous tasks require self-control because they interfere with endogenous reward
On treating effort as a dynamically varying cost input
Opportunity cost calculations only determine justified effort – Or, What happened to the resource conservation principle?
Opportunity prioritization, biofunctional simultaneity, and psychological mutual exclusion
Persistence: What does research on self-regulation and delay of gratification have to say?
Persisting through subjective effort: A key role for the anterior cingulate cortex?
Subjective effort derives from a neurological monitor of performance costs and physiological resources
The costs of giving up: Action versus inaction asymmetries in regret
The economics of cognitive effort
The intrinsic cost of cognitive control
The opportunity cost model: Automaticity, individual differences, and self-control resources
Theories of anterior cingulate cortex function: Opportunity cost
Willpower is not synonymous with “executive function”
Author response
Cost-benefit models as the next, best option for understanding subjective effort