2.1. Definition

The definition of motivation varies across different times and fields (e.g., Kleinginna & Kleinginna, Reference Kleinginna and Kleinginna1981; Madsen, Reference Madsen1974), but one common definition is that motivation energizes and directs behavior (Lewin, Reference Lewin1942; Locke & Latham, Reference Locke and Latham2004; Reeve, Reference Reeve2017; Simpson & Balsam, Reference Simpson and Balsam2016; VandenBos & American Psychological Association, Reference VandenBos2007; Weiner, Reference Weiner1992). Energization means that motivation instigates or initiates action or behavior (Elliot, Reference Elliot, Bong, Reeve and Kim2023), which shall be called the “spring to action” of behavior (James, Reference James1890), a force that impels behavior (Descartes, Reference Descartes1955), or “energy behind our actions” (Wigfield, Muenks, & Eccles, Reference Wigfield, Muenks and Eccles2021). Direction means that it guides and channels behavior in a certain way. The former aspect of motivation is often referred to as “motives” (Anderson et al., Reference Anderson, Hildreth and Howland2015; Atkinson & Raynor, Reference Atkinson and Raynor1978; McClelland et al., Reference McClelland, Atkinson, Clark and Lowell1976) or “needs” (Deci & Ryan, Reference Deci and Ryan1985; Dweck, Reference Dweck2017; Maslow, Reference Maslow1943; Stevens & Fiske, Reference Stevens and Fiske1995). The latter aspect of motivation is referred to as “goals” or “values” (Austin & Vancouver, Reference Austin and Vancouver1996; Eccles & Wigfield, Reference Eccles and Wigfield2002; Elliot & Fryer, Reference Elliot, Fryer, Shah and Gardner2008; Fishbach & Ferguson, Reference Fishbach, Ferguson, Kruglanski and Higgins2007). Combined together, motivation is conceptualized as a determinant of a certain set of behaviors (Fig. 1A).

Figure 1.

Motivation as the (black box) determinant of behavior (A) and motivation as psychological construction of underlying mental computational processes (B). Graphics are available under creative commons licenses (https://creativecommons.org/), downloaded from https://upload.wikimedia.org/wikipedia/commons/d/d9/A_black_box.svg, https://commons.wikimedia.org/wiki/File:Gear_-_Noun_project_7137.svg, and https://commons.wikimedia.org/wiki/File:Gears-686316.jpg.

Our critical analysis of motivation constructs mainly concerns the former aspect, the function of energization (see also Hinde, Reference Hinde1960, for another critique). Regardless of one's theory of motivation, motivation is almost always used to explain the initiation (or the intention of the initiation) of behavior; as such, energization is often regarded as the definitive aspect of motivation (e.g., Madsen, Reference Madsen1974). The direction aspect of motivation is also often considered a fundamental aspect of motivation, but we view this aspect as somewhat subsumed within the energization aspect (see also Elliot, Reference Elliot, Bong, Reeve and Kim2023). In fact, most of the extant constructs of high-level motives or needs emphasize energization in terms of what they are motivated for (i.e., direction). For example, an achievement motive represents the motivation toward a high standard of excellence (McClelland et al., Reference McClelland, Atkinson, Clark and Lowell1976) and a need for autonomy directs people toward fulfilling a sense of agency (Ryan & Deci, Reference Ryan and Deci2017). It is difficult to imagine motives or needs that instigate behavior in a completely nonspecific manner, other than a few limited examples (e.g., general Pavlovian instrumental transfer; Corbit & Balleine, Reference Corbit and Balleine2005). We will briefly revisit the direction aspect of motivation in a later section.

The concept of motivation to explain basic human behavior (e.g., mating, consumption of foods) was initially subject to criticisms concerning its operationalization (e.g., Koch, Reference Koch1941). Although these points overlap with our criticism to some degree, our main criticism is aimed at the motivation constructs that explain a broad range of higher-order behavior, which we shall call high-level motivation constructs.

2.2. An example

To render our criticism concrete and easy to understand, we first show one example of how high-level motivation constructs are used to explain behavior: The need for competence (Ryan & Deci, Reference Ryan and Deci2020). The need for competence is based on the concept of competence proposed by White (Reference White1959). According to White (Reference White1959), competence reflects the organism's capacity to effectively interact with its environment. Extending this idea, Deci and Ryan (Reference Deci and Ryan1985) proposed that humans have a basic psychological need for competence, which is defined as people's motivation to experience feelings of mastery and success. We use the need for competence as an example simply because it is one of the most accepted high-level motivation constructs in the field (19,800 hits in Google Scholar in September 2023), and recent theoretical progress has made it possible to instantiate our point (Murayama, Reference Murayama2022).

Humans and animals often exhibit behavior that seems to be aimed at mastering the environment. Research showed that monkeys are engaged with solving puzzles (Harlow, Reference Harlow1949). Humans have the capacity to engage in learning for a prolonged period of time (Hidi & Renninger, Reference Hidi and Renninger2019). Humans and animals also have a tendency to explore the environment without clear rewards (Berlyne, Reference Berlyne1966) and a tendency to seek positive feedback (Elliot & Moller, Reference Elliot and Moller2003). None of these behaviors can be explained by basic motivation such as a hunger drive, and are considered to be caused by the need for competence (Deci & Ryan, Reference Deci and Ryan1985). Using this construct, we can explain the behavior in the following way: “We have a tendency to explore because we have a basic motivation to master the environment,” and, “People can sustain their commitment to an activity because people have a fundamental motivation to seek mastery.” The satisfaction of this need for competence is theorized to enhance one's intrinsic motivation (i.e., motivation to work on a task without relying on explicit incentives), and to lead to many positive long-term outcomes such as higher well-being (Ryan & Deci, Reference Ryan and Deci2017, Reference Ryan and Deci2020).

3.1. The black-box problem

As the definition of motivation as an energizer and director of behavior attests, researchers typically regard motivation as the initial cause (i.e., origin) of a certain set of behaviors (Fig. 1A). Of course, researchers have assumed that motivation constructs are influenced by many external factors, such as environmental changes, learning, socialization, and development, and that they have genetic origins as well as neural bases (McClelland, Reference McClelland1987). In early research, motivation was conceptualized as an “intervening variable” (Hull, Reference Hull1943), meaning that it has both external antecedents (e.g., deprivation of food) and outcomes (e.g., increased response). However, many motivation constructs are regarded as the origin of behavior in the sense that they are the internal variable which is supposed to generate the willpower to initiate the action in the first place.

This property of motivation constructs is particularly useful for researchers to understand higher-order behavior. As noted in the example above, we can explain exploratory behavior by proposing that we have a basic motivation to master the environment. Crucially, however, the high-level motivation construct does not truly explain what it is or how this behavioral tendency is generated (for historical arguments, see Bindra, Reference Bindra1959; Koch, Reference Koch and Jones1956). Instead, motivation is like a black box, where the process that generates the behavior is unknown. By supposing motivation to explain behavior, we implicitly fall prey to the so-called “motivational homunculus” problem (see also Gladwin, Figner, Crone, & Wiers, Reference Gladwin, Figner, Crone and Wiers2011). That is, to explain how motivated behavior is generated, we posit a construct that works as a “generator” of motivated behavior, but this logic may suffer from the issue of infinite regress (Kenny, Reference Kenny, Grene and Prigogine1971). The real danger here is that, because the constructs seemingly explain the set of higher-order behaviors well, we may think that all questions are answered, and soon stop investigating what these constructs are and how they work.

Historically speaking, there was a time when researchers tried to eliminate the black-box property of motivation constructs by proposing physiological or biological causes. For example, Hull's concept of drive such as hunger drive or thirst drive was directly linked to the physiological deficits of food or water (Hull, Reference Hull1943). There are also contemporary theories that connect some motivation constructs with simple physiological or biological factors (e.g., testosterone and power motivation; Schultheiss, Campbell, & McClelland, Reference Schultheiss, Campbell and McClelland1999). However, recent studies suggest that such simple one-to-one correspondence between a physiological factor and motivation is not plausible to explain motivation constructs for higher-order complex behavior (e.g., Kim, Reference Kim2013; Murayama, Izuma, Aoki, & Matsumoto, Reference Murayama, Izuma, Aoki, Matsumoto, Kim, Reeve and Bong2017; Steinman, Duque-Wilckens, & Trainor, Reference Steinman, Duque-Wilckens and Trainor2019). We believe that high-level motivation constructs partly gained popularity because they paralleled motivations for basic behavior. With the analogy of motivation for food, the statement, “We have a tendency to do X because we have a fundamental motivation for it,” sounds intuitive and convincing. However, this parallism is misleading, because it provides a false impression that high-level motivation has clear physiological or biological causes.

Note that our argument differs from the issue of circular explanation that is often discussed in the classical literature (e.g., Bindra, Reference Bindra1959; Seward, Reference Seward1939; i.e., motivation constructs explain a particular type of behavior by arguing that people have a motivation for that behavior). Motivation constructs do have great utility in that they can make generalizable predictions (Berridge, Reference Berridge2004) – for example, by supposing humans have a need for competence, one can predict that humans perform a range of epistemic behaviors, even if these behaviors were not part of the original observation. The constructs also make our explanation parsimonious – now we can conceptualize these behaviors as manifestations of the single motivational construct of the need to belong. By assuming general properties for motivation constructs, we can even make novel predictions for behavior (Baumeister & Leary, Reference Baumeister and Leary1995). Therefore, circularity is not an issue in our opinion. Our argument is, rather, that motivation is considered a useful explanatory variable (i.e., explanans), but it does not have explanatory variables itself (i.e., explanandum).

3.2. Lack of consensus on the definition of high-level motivation constructs

The problem of the motivational black box gives rise to another critical issue: Challenges in defining high-level motivation constructs. Because motivation constructs are created to explain a certain set of behaviors without specifying their internal properties, they can be defined only in terms of the behaviors explained by the constructs (see Fig. 1A). As a result, there is always room for ambiguity when one tries to define them based on their internal properties. In other words, high-level motivation constructs have an inherent challenge for precise definition because of their black-box property.

This issue has manifested in various forms in the literature of motivation. In early years, researchers tried to create a comprehensive list of human needs (McDougall, Reference McDougall and McDougall1909; Murray, Reference Murray1938), but there was always a question of how we could be certain that two similar needs were distinct and not the same, or whether certain needs were or were not fundamental (Pittman & Zeigler, Reference Pittman, Zeigler, Kruglanski and Higgins2007). In recent years, this issue has often been discussed in the context of jingle–jangle fallacies of motivation constructs (e.g., Bong, Reference Bong1996; Pekrun, Reference Pekrun, Bong, Reeve and Kim2023; Pekrun & Marsh, Reference Pekrun and Marsh2022), where the same construct label is used to denote different constructs or different construct labels are used for the same construct (Kelley, Reference Kelley1927). Taking the example of need for competence again, the construct is considered to be a source of intrinsic motivation (Deci & Ryan, Reference Deci and Ryan1985; i.e., if this need is satisfied, intrinsic motivation increases). Does this mean they are separate constructs? Or is need for competence a constituent part of intrinsic motivation? There are also other constructs related to need for competence. For example, self-efficacy is a belief in one's capacity to competently control the environment (Bandura, Reference Bandura1997) and is clearly related to need for competence. Perceived control (Skinner, Reference Skinner1996), self-esteem (Baumeister, Reference Baumeister1993), and self-concept (Marsh & Shavelson, Reference Marsh and Shavelson1985) are also similar constructs clearly connected to need for competence. Are they different constructs and, if not, what are the relationships? Great effort has been devoted to resolving these issues, and the work helped researchers to deepen our thoughts about these motivation constructs. However, given the black-box property, it is virtually impossible to make a conclusive judgment on the difference between constructs, other than arguing “the originator of the theory said so” (see also Murayama, FitzGibbon, & Sakaki, Reference Murayama, FitzGibbon and Sakaki2019).

This issue of defining high-level motivation constructs is especially problematic when researchers want to make causal inferences. For example, in survey studies, researchers use a variety of designs (e.g., a longitudinal study) to estimate the causal effect of a motivation construct on outcome variables. But does that mean that motivation has a causal effect? To make a causal inference, we estimate the effect when motivation is (hypothetically) “intervened on,” holding other factors constant. However, given the black-box property of high-level motivation constructs, researchers often have difficulty in judging whether or not some potential controlling variables are the inherent property of the motivation constructs (see Eronen, Reference Eronen2020; Hernán & VanderWeele, Reference Hernán and VanderWeele2011). For example, think of a situation in which researchers want to learn the causal effect of (satisfaction of) need for competence on math exam performance using longitudinal survey data. One may want to treat self-esteem as a controlling variable, but one could also argue that self-esteem is a constituent part of need for competence. In such cases, controlling for self-esteem does not make sense. But such a decision is often difficult because the constructs are underidentified. Generally speaking, when the target construct is not unambiguously defined, we can never make a solid causal inference from empirical data (Rohrer & Murayama, Reference Rohrer and Murayama2023).

Some may argue that the issue could be addressed empirically. For example, researchers often use the strategy of testing “incremental validity,” in which motivation construct A has predictive power over an outcome variable above and beyond a similar motivation construct B (Smith, Fischer, & Fister, Reference Smith, Fischer and Fister2003). Some other researchers use factor analysis and show that items representing motivation A and B form distinctive factors (Byrne, Reference Byrne2001; based on certain statistical criteria). Positive results from these analyses often lead researchers to conclude that the two motivation constructs “are overlapping but distinct.” However, the analysis does not directly answer the question of definitions, because the evidence simply shows that the two measurements are assessing something different, and is mute to exactly how the two theoretical constructs are differently defined. In fact, after such a conclusion, it is often not entirely clear what it really means that two motivational constructs are overlapping but still distinct.

4.1. High-level motivation as a psychological construction

We offer an alternative perspective on high-level motivation constructs. Specifically, we propose that we should not view high-level motivation constructs as the original causal determinant of behavior. Rather, we argue that such high-level motivation constructs are an emergent property of underlying mental computational processes (Fig. 1B). To make sense of these emergent properties, humans construe the construct of motivation. In other words, high-level motivation constructs are a consequence of psychological construction.

We define mental computational processes as concrete internal mechanisms which produce behavior (see also Marr, Reference Marr1982). Consider a robot that behaves like humans. The robot employs particular computational mechanisms to process external input (i.e., sensory input) and decide on actions (i.e., output). Oftentimes some stored information in the robot (memory) plays a critical role for this computation. We call all of these mechanisms mental computational processes. Of course, humans are not robots. Although these mental computational processes determine people's behavior, humans also have subjective experiences such as positive and negative feelings, and these subjective experiences should be influenced by mental computational processes (LeDoux, Reference LeDoux2014).Footnote ² Figure 1B provides a schematic picture.

Importantly, humans are capable of recognizing regularities in and creating mental categories from their own behaviors and subjective experiences (Reeder, Reference Reeder2009). “Motivation” may be a convenient term to explain these categories. Many studies suggest that we are indeed naturally inclined to infer motivations or intentions from a variety of observations (e.g., Baillargeon et al., Reference Baillargeon, Scott, He, Sloane, Setoh, Jin, Bian, Mikulincer, Shaver, Borgida and Bargh2015). As a result, if we have a tendency to be affiliated with certain social groups, for example, people may be naturally convinced that we have an affiliative or social motivation. However, such inference does not necessarily mean that social motivation is itself represented in our mental computational processes – instead, social motivation could be a consequence of interpreting and categorizing the regularities that exist in behavioral patterns and subjective experiences. As such, motivation is the subjective interpretation of or label for an emergent property arising from underlying mental computational processes.

From this alternative perspective, the key solution to the black-box problem is simple: To unpack the black box. Specifically, researchers on motivation should take further steps to unravel the mental computational processes underlying high-level motivation constructs. In the following section, we demonstrate how this principle can be applied to the motivation construct need for competence.

4.2. A case for reward-learning models of information seeking

We propose that the black box of need for competence can begin to be unpacked with reward-learning models of information-seeking behavior (e.g., FitzGibbon, Lau, & Murayama, Reference FitzGibbon, Lau and Murayama2020; Gottlieb, Oudeyer, Lopes, & Baranes, Reference Gottlieb, Oudeyer, Lopes and Baranes2013; Gruber & Ranganath, Reference Gruber and Ranganath2019; Marvin & Shohamy, Reference Marvin and Shohamy2016). In our view, information seeking is an excellent lens from which we can unpack some types of motivation, including need for competence, because the act of seeking information describes effort taken to acquire knowledge, a fundamental process to master the environment. We are not claiming that the reward-learning model is the best model to instantiate need for competence with mental computational processes; there are several alternatives (e.g., Patankar et al., Reference Patankar, Zhou, Lynn, Kim, Ouellet, Ju and Bassett2023) and there are many different versions of reward-learning models (e.g., Oudeyer & Kaplan, Reference Oudeyer and Kaplan2007). Our purpose is not to compare these models. We use this model simply for demonstration purposes as it provides a useful example of how our perspective can explain certain types of motivation constructs.

Although there are many different versions of reward-learning models to explain human information-seeking behavior, a common assumption is that information is an intrinsic reward. Murayama (Reference Murayama2022; see also Murayama et al., Reference Murayama, FitzGibbon and Sakaki2019) summarized and expanded on these common aspects of reward-learning models in the “reward-learning framework of knowledge acquisition” (Fig. 2). According to the framework, when an agent identifies some uncertainty in its knowledge (often called “knowledge gap”), the agent computes the expected rewarding value of upcoming information, and if it is deemed valuable, the agent initiates information-seeking behavior. When the agent successfully acquires the information, the agent experiences a positive rewarding feeling, which in turn strengthens the value of the same sort of information. This is a well-known reinforcement principle (Hull, Reference Hull1943) but the critical point is that the framework assumes that information itself can have rewarding value. There are different algorithms proposed to accurately quantify the rewarding value of information (e.g., uncertainty; Bennett, Bode, Brydevall, Warren, & Murawski, Reference Bennett, Bode, Brydevall, Warren and Murawski2016; van Lieshout et al., Reference van Lieshout, Vandenbroucke, Müller, Cools and de Lange2018). Furthermore, Murayama (Reference Murayama2022) argued that acquired information is consolidated into the existing knowledge base, and this expanded knowledge could prompt the agent to become more aware of further knowledge gaps (i.e., “the more we know about a topic, the more likely we realize that there are things that we do not know”). As a result, this system creates a positive feedback loop, sustaining long-lasting information-seeking behavior.

Figure 2.

Reward-learning framework of knowledge acquisition (adapted from Murayama, Reference Murayama2022).

The framework aims to provide a rough summary of current information-seeking models. In these models, we can see several mental computational processes included, such as assessment of knowledge gap, computation of expected reward value of information, selection of information, integration of the information into existing knowledge, and so on. These computations do not normally operate consciously, but rather implicitly (Murayama et al., Reference Murayama, FitzGibbon and Sakaki2019). Some of these information-seeking models are conceptual (e.g., Gruber & Ranganath, Reference Gruber and Ranganath2019; Jach, DeYoung, & Smillie, Reference Jach, DeYoung and Smillie2022). Some other researchers propose computational models to accurately describe people's information-seeking behavior (for a review, see Baldassarre & Mirolli, Reference Baldassarre and Mirolli2013) and some other models are even implemented in robots with the aim to replicate people's (especially infants' or young children's) information-seeking behavior (e.g., Baranes & Oudeyer, Reference Baranes and Oudeyer2013). This means that, by implementing these mental computational processes, we can create an agent that actively seeks information to expand its knowledge.

Critically, the agent which implements the reward-learning models actively and continuously searches for information that it does not know to expand its knowledge, as if it has the motivations for mastery and competence, despite there being no need for competence featured in the mental computational processes (i.e., there are no boxes or parameters directly representing need for competence in the model). Need for competence thus appears as an emergent property of reward-learning models of information-seeking behavior. Several other motivational concepts can be explained in a similar way. For example, the motivational concept of interest often refers to people's enduring tendency to engage in particular learning content over time (Hidi & Renninger, Reference Hidi and Renninger2019; Renninger & Hidi, Reference Renninger and Hidi2016). The agent can realize this enduring information-seeking behavior by incorporating their knowledge base and resultant positive feedback loop into the system (Murayama, Reference Murayama2022; Murayama et al., Reference Murayama, FitzGibbon and Sakaki2019). The agent also appears “intrinsically motivated” in the sense that it actively searches its environment without extrinsic incentives.

Of course, there are many motivational constructs that cannot be explained by the presented reward-learning framework of knowledge acquisition (e.g., need to belong). Additionally, although reward-learning models are dominant in the fields of cognitive science and neuroscience, this is not the only way to specify mental computational processing of motivated behavior. The example is simply intended to demonstrate that high-level motivation construct can be a consequence of the subjective construction from behavioral regularities.

We used need for competence and the reward-learning framework of knowledge acquisition as an illustrative example, but there are other potential examples which demonstrate that higher-level motivation constructs can be an emergent property of mental computational processes. For example, Shultz and Lepper (Reference Shultz and Lepper1996) proposed a connectionist model to explain various experimental findings of cognitive dissonance (e.g., Festinger & Carlsmith, Reference Festinger and Carlsmith1959). These findings are often explained by positing that humans have a fundamental motive to maintain cognitive consistency or reduce cognitive dissonance (Festinger, Reference Festinger1957; Heider, Reference Heider1958). However, the model explained the experimental findings without explicitly supposing such a motivation. In addition, O'Reilly (Reference O'Reilly2020) built a multi-layered connectionist model (inspired by neuroscientific findings) and argued that the model could explain motivated behavior (i.e., the dynamic nature of goal-directed behavior) without explicitly incorporating motivation into the model.

4.3. Strengths of considering mental computational processes

By specifying the mental computational processes underlying higher-order motivated behavior, high-level motivation constructs are no longer black boxes. Instead, they clearly explain what the motivation construct is and how this behavioral tendency is generated. Importantly, the proposed perspective explicitly refutes the idea that high-level motivation constructs themselves cause wide-ranging higher-order human behavior (see Fig. 1A). Rather, our behavior is governed by the mental computational processes, which form a collective dynamic system of interacting elements. Different types of higher-order behavior (e.g., exploring the environment, sustaining epistemic engagement, seeking competence-relevant information) are the consequences of the integration or parts of this collective system, not the consequence of a unitary construct of motivation.

The proposed perspective also neatly sidesteps the fundamental challenge of defining high-level motivation constructs. This is because it is the mental computational processes, not the motivation constructs themselves, that are necessary to understand human behavior. For example, as we presented in the previous section, a description of the mental computational processes that explain how people decide to seek information over time is sufficient to explain the behaviors related to need for competence, intrinsic motivation, and interest. There is no further need to discuss which components of this process represent need for competence, intrinsic motivation, or interest, because we already explained the mechanism and behavior (see also Kidd & Hayden, Reference Kidd and Hayden2015; Murayama et al., Reference Murayama, FitzGibbon and Sakaki2019). The proposed perspective indicates that the priority should be given to understanding the underlying computational mechanisms, not discussing the boundaries of (inherently ill-defined) constructs.

An important benefit of the proposed perspective is that it provides a different way of theorizing about and examining motivation: Describing how a high-level motivation construct is realized by mental computational processes (for a discussion on the merit of this approach in psychology, see van Rooij & Baggio, Reference van Rooij and Baggio2021). Traditionally, researchers start by positing a specific high-level motivation construct and develop a theory by specifying the factors related to the construct (e.g., personality traits, well-being). As a consequence, empirical research has been mainly interested in examining external antecedents (e.g., family environment) and outcomes (e.g., well-being) of motivation (establishing the so-called “nomological network”; Cronbach & Meehl, Reference Cronbach and Meehl1955). As detailed later, we see merit in this approach. However, this is just one way of understanding motivated behavior. An alternative process-focused approach can create new sets of research questions to further advance our understanding of motivation.

For example, we showed that the reward-learning framework of knowledge acquisition (Murayama, Reference Murayama2022) can also provide a theory of how need for competence is realized through reward-learning mechanisms. Once such a theory is developed, researchers can empirically test its validity by examining various parts of the mental processes (e.g., “Does a knowledge gap really facilitate information-seeking behavior?”). The theory can also help us evaluate the antecedents and outcomes identified in the previous empirical literature (e.g., which component of the mental computational processes can be altered by family environment?). Furthermore, the model prompts researchers to critically analyze the assumptions underlying the model. For example, the reward-learning framework assumes that information works as rewards, but the model does not specify the type of information that is perceived as rewarding. Then researchers can further theorize and empirically test the nature of information that people actively seek, further pinning down the origin of motivated behavior. In fact, this is currently a hot topic of the field (Fitzgibbon & Murayama, Reference Fitzgibbon and Murayama2022; Gottlieb & Oudeyer, Reference Gottlieb and Oudeyer2018; Kidd & Hayden, Reference Kidd and Hayden2015).

4.4. Reinterpreting past empirical findings

It is important to restress that we do not disregard voluminous empirical survey and experimental studies of motivation in the past century. Much research has found that motivation, as assessed by survey questions or manipulated in experiments, is related to human behavioral and psychological outcomes. These results clearly show the usefulness of motivational constructs in predicting behavior. In fact, “motivation constructs” have been popular in psychology because they can predict various important outcomes, such as well-being, achievement scores, career choice, and so on (e.g., Robbins et al., Reference Robbins, Lauver, Le, Davis, Langley and Carlstrom2004). These empirical studies can also inform researchers of potential intervention programs (e.g., Lazowski & Hulleman, Reference Lazowski and Hulleman2016). However, our argument is that we should not interpret these empirical findings as evidence that high-level motivation constructs directly cause behavior.

For example, in many studies of high-level motivation constructs, researchers assess certain types of motivation using established survey questions. These survey questions often assess certain aspects of mental computational processes (e.g., the positive rewarding feeling following knowledge acquisition) or overall behavior (e.g., the frequency of active information-seeking behavior) related to the motivational construct. When the aggregated scores are related to outcomes and potential confounders are well controlled (Fig. 3, top “Statistical results from data”), we tend to argue that “motivation,” assessed by the survey questions, caused the outcomes (Fig. 3, left, “Motivation as determinant”). This way of thinking is common, and is often strengthened by the recent proliferation of latent variable modeling (Jöreskog, Reference Jöreskog1969), which visually shows path diagrams where boxes representing motivation predict or are predicted by other variables.

Figure 3.

Different interpretations of statistical results according to different perspectives.

However, the association between aggregated scores and the outcome could happen in a different causal scenario. Specifically, these survey items which reflect parts of mental computational processes (which interact with each other) may directly influence other variables without mediating motivation constructs (Fig. 3, right, “Motivation as psychological construction”). This way of viewing measurement has gained increasing attention in the methodological literature (Donnellan, Usami, & Murayama Reference Donnellan, Usami and Murayama2023; e.g., McClure, Jacobucci, & Ammerman, Reference McClure, Jacobucci and Ammerman2021; VanderWeele, Reference VanderWeele2022). For example, Donnellan et al. (Reference Donnellan, Usami and Murayama2023) argued that such interpretations are more realistic than latent variable modeling when items represent a wide range of behavior and psychological processes, and proposed a new mixed-effects model to analyze survey data without relying on latent variables (see also Rhemtulla, van Bork, & Borsboom, Reference Rhemtulla, van Bork and Borsboom2020). From this perspective, aggregated survey scores capture the extent to which the mental computational processes efficiently work as a whole for a particular individual or a situation (see van der Maas et al., Reference van der Maas, Dolan, Grasman, Wicherts, Huizenga and Raijmakers2006, for a similar argument in case of intelligence). Such efficiency scores are parsimonious and useful for prediction, but we do not need to assume that these scores capture the motivation construct itself.

We believe this reframing of empirical studies on motivation is particularly important because researchers often face situations in which empirical findings disagree depending on how we measure or manipulate motivation (e.g., Hulleman, Schrager, Bodmann, & Harackiewicz, Reference Hulleman, Schrager, Bodmann and Harackiewicz2010; Utman, Reference Utman1997). Recent years have also seen many studies indicating that the same constructs assessed by survey questions and experimental tasks have different correlates (Dang, King, & Inzlicht, Reference Dang, King and Inzlicht2020; Eisenberg et al., Reference Eisenberg, Bissett, Zeynep Enkavi, Li, MacKinnon, Marsch and Poldrack2019), and this has also been a recurring issue in motivation constructs (Deci, Koestner, & Ryan, Reference Deci, Koestner and Ryan1999; McClelland, Koestner, & Weinberger, Reference McClelland, Koestner and Weinberger1989; Schultheiss, Yankova, Dirlikov, & Schad, Reference Schultheiss, Yankova, Dirlikov and Schad2009). In such situations, we are often encouraged to accurately define the construct or establish a valid measurement. It is indeed true that we need precise definitions of constructs and valid measurement, but at the same time, the argument rests on an implicit assumption that there is a single construct of motivation that determines behavior and thus the inconsistent results are problematic. From the perspective of psychological construction, on the contrary, such divergent results are natural when different assessments or manipulations tap different aspects of the mental computational processes. Having a precise process model in mind, researchers can then make more fine-grained predictions about how different types of assessments or manipulations result in different outcomes. Such a process-oriented perspective can also help researchers develop targeted interventions that aim to change specific outcomes.

5.1. What is the distinction between basic motivation constructs and high-level motivation constructs?

The high-level motivation constructs evaluated in this article explain a wide-range of higher-order behavior, whereas basic motivation constructs have a relatively narrow set of behaviors as explanandum (e.g., the motivation for sex to explain people's desire to copulate). However, although we made a distinction in this article for the purpose of simplicity, the distinction is continuous rather than dichotomous. We do not argue that motivation for basic behavior does not suffer from the issue of black box at all – this is a matter of degree. Our point is that the problem is exacerbated for high-level motivation constructs.

5.2. Are mental computational processes free from motivation constructs?

No, they are not. As mental computational processes produce behaviors, we must logically assume something which initiates behavior either explicitly or implicitly. In the reward-learning framework of knowledge acquisition, there is an implicit assumption that people choose to seek information that has a high reward value. Therefore, one could argue that the framework posits a “motivation to seek rewarding information,” perhaps just before the box of information-seeking behavior in Figure 2. Unless we understand the basic biological/physical mechanisms to produce behavior, we can never eliminate motivation-like constructs.

But our argument is not to eliminate motivation constructs from explanation. Our proposal is that theorizing mental computational processes (i.e., analyzing motivation from a different level) can provide new insights into what researchers have called “motivation” in the literature, enabling a deeper understanding of our motivated behavior. For example, both exploration behavior and long-term intellectual engagement are thought to be caused by need for competence, but there is no explanation for how these outcomes are related. By specifying the mental computational processes, we can propose that exploration behavior is caused by the rewarding value of information, whereas long-term engagement is a consequence of a positive feedback loop created by expanding existing knowledge from new information (Fig. 2). Both behaviors are produced by the same mental computational system (so it makes sense that they are explained by the same construct) but the way they are produced within the system is different.

We can go even further down the line to unpack motivation-like constructs in the specified mental computational processes. As noted above, the reward-learning framework implicitly assumes a motivation to seek rewarding information. But what are the basic building blocks of rewarding value for information? Is that novelty (Poli, Meyer, Mars, & Hunnius, Reference Poli, Meyer, Mars and Hunnius2022), entropy (Loewenstein, Reference Loewenstein1994), or Kullback–Leibler divergence (Ningombam, Yoo, Kim, Song, & Yi, Reference Ningombam, Yoo, Kim, Song and Yi2022)? As we do not have direct access to such information in reality, how can such a metric be computed in the real world? (Gottlieb & Oudeyer, Reference Gottlieb and Oudeyer2018). Note that as we go down the level, motivation-like constructs become narrower and narrower (or more and more specific), and at some point we may no longer feel comfortable to call it motivation. When we orient attention to an object that suddenly comes into our visual field, for example, is this a manifestation of motivation? We can in theory suppose such a motivation that causes the orientation behavior, but many researchers would say it is not necessary or useful (Murayama, Reference Murayama, Bong, Reeve and Kim2023b).

In a sense our proposal is consistent with the idea that the functioning of human behavior can be understood at different levels of granularity with the lowest being the biological or physical level (e.g., Dennett, Reference Dennett1987; Marr, Reference Marr1982; Newell, Reference Newell1994; in case of motivation, see Nagengast & Trautwein, Reference Nagengast, Trautwein, Bong, Reeve and Kim2023). Our proposed perspective indicates that high-level motivation constructs reflect higher-level explanations whereas mental computational processes represent lower-level explanations. No level of understanding should be dismissed as “wrong” (i.e., one level of explanation should not be replaced with a lower-level explanation), because they explain the behavior for different purposes, but the problem of motivation literature is that most researchers are satisfied with higher-level explanations (i.e., supposing high-level motivation constructs to explain behavior) and little effort has been made to pursue lower-level explanations.

5.3. Has your perspective truly not been discussed in the prior motivation literature?

The idea of psychological construction is not novel at a general level – it has been discussed in various forms in psychological and cognitive science as well as philosophy of science (Brick, Hood, Ekroll, & de-Wit, Reference Brick, Hood, Ekroll and de-Wit2022; Churchland, Reference Churchland1979; Dalege et al., Reference Dalege, Borsboom, van Harreveld, van den Berg, Conner and van der Maas2016; Danziger, Reference Danziger1990; MacCorquodale & Meehl, Reference MacCorquodale and Meehl1948; Pessoa, Medina, & Desfilis, Reference Pessoa, Medina and Desfilis2022; Stich & Ravenscroft, Reference Stich and Ravenscroft1994). Other psychological constructs such as emotions, personality, intelligence, and consciousness have also received similar scrutiny (Boag, Reference Boag2018; Fiske, Reference Fiske2020; Lau, Reference Lau, Murphy, Ellis and O'Connor2009; Russell, Reference Russell2003; van der Maas et al., Reference van der Maas, Dolan, Grasman, Wicherts, Huizenga and Raijmakers2006). Recent theoretical developments in psychological network analysis (Borsboom & Cramer, Reference Borsboom and Cramer2013) have a similar root (for application to motivational constructs, see Sachisthal et al., Reference Sachisthal, Jansen, Peetsma, Dalege, van der Maas and Raijmakers2019; Tamura et al., Reference Tamura, Ishii, Yagi, Fukuzumi, Hatano, Sakaki and Murayama2022). However, the construct of motivation can be uniquely positioned in this context because motivation is characterized as the causal determinant of behavior. As we showed earlier, this unique property poses fundamental challenges when theorizing or interpreting motivated behavior, and the psychological construction perspective has clear utility to circumvent the issue.

Nevertheless, there is some (albeit limited) work related to our psychological construction perspective in the literature of motivation. For example, Bem (Reference Bem1967) suggested that dissonance reduction motive (Festinger, Reference Festinger1957) can be interpreted as the consequence of self-perception. The idea is that people are not motivated by the dissonance reduction motive, but simply act and infer attitudes by observing/interpreting what they did (i.e., attribution process). This idea is seemingly similar to the psychological construction perspective in that it does not assume a motivation to explain higher-order behavior. However, the difference is that Bem's (Reference Bem1967) theory uses the attribution process to explain people's actual behavior or attitude (e.g., “people changed their behavior because of attribution”). On the contrary, our proposed perspective would suggest an interplay between different mental computational processes that together produced something we would categorize as a “dissonance reduction motive.” The attribution process could be part of it, but not the whole (in fact, it is not plausible that higher-order behavior is largely explained by attribution processes). Therefore, although we acknowledge that there were similar ideas in the literature of motivation, our proposed psychological construction perspective is critically distinct in that we stress the importance of specifying mental computational processes.

That said, this line of work suggests another interesting line of work for future research. Specifically, we could examine the processes of how people construct motivation from the regularities of observable behavior and subjective feelings. Although early research on attribution provided many insights into this psychological construction process (e.g., Bem, Reference Bem1967; Nisbett & Wilson, Reference Nisbett and Wilson1977), the proliferation of high-level motivation constructs in recent years seems to have suppressed this tradition of work. But by combining the work on mental computational processes and psychological construction processes, we may be able to achieve deeper insights into motivated behavior. For example, there is a possibility that such attributional process of motivation works as a reinforcer of motivated behavior itself, influencing mental computational processes (e.g., “I like to study because I think I have high achievement motivation”; see Palminteri & Lebreton, Reference Palminteri and Lebreton2022). Future studies could examine such interactive processes.

5.4. Does the criticism only apply to the energization aspect of motivation and not the direction aspect?

Our discussion focused mainly on the energization aspect of motivation, but motivation is said to additionally have a directional aspect. The directional aspect of motivation channels people's behavior in a certain way, and researchers argue that values (Eccles & Wigfield, Reference Eccles and Wigfield2002) or goals (Elliot & Fryer, Reference Elliot, Fryer, Shah and Gardner2008) play a critical role here. Does our proposed perspective have implications for the directional aspect?

As noted earlier, the distinction between energization and direction aspects in motivation is somewhat ambiguous. To clarify the distinction, Niv, Joel, and Dayan (Reference Niv, Joel and Dayan2006) introduced a normative account of motivation, by defining the directional aspect of motivation as a modulation of the mapping between reward outcome and utility (Fig. 4). When one is in a state of hunger, the utility of food reward would be increased. When one is in a state of thirst, the utility of food may not be as high but the utility of water would be increased. This means that hunger motivation altered the utility of food. This is one of the commonly accepted definitions of motivation in the classic literature (Berridge, Reference Berridge2004; Dickinson & Balleine, Reference Dickinson, Balleine, Pashler and Gallistel2002; Hull, Reference Hull1943). By viewing motivation as the modulator of reward utility, we can independently define motivation as separate from the energization aspect (Niv et al., Reference Niv, Joel and Dayan2006) – it simply changes the mapping of utility without directly causing behavior. In this respect, therefore, our criticism does not apply to the directional aspect of motivation as exemplified by hunger or thirst.

Figure 4.

Direction aspect of motivation: Motivation functions as utility mapping.

Goals and values could be considered in the same manner. Like hunger or thirst, goals and values function as the modulator of utility; when one values acquiring mathematics competence, for example, they would assign high utility or experience stronger positive feelings when succeeding in a math exam than those who do not value math. When one pursues performance goals, outperforming others would give stronger positive feelings than those pursuing mastery goals. Goals and values are also not viewed as initial determinants of behavior; rather, we usually think that goals and values are set in people's mind through certain mechanisms. They are often externally provided (e.g., external goal setting; Harackiewicz & Sansone, Reference Harackiewicz, Sansone, Maehr and Pintrich1991; Locke & Latham, Reference Locke and Latham1990; value intervention; Rozek, Svoboda, Harackiewicz, Hulleman, & Hyde, Reference Rozek, Svoboda, Harackiewicz, Hulleman and Hyde2017) or internally generated (e.g., self-set goals; Barron & Harackiewicz, Reference Barron and Harackiewicz2001; Locke, Reference Locke2001; internalization of values; Renninger & Hidi, Reference Renninger and Hidi2016). Therefore, our criticism does not directly apply to these concepts.

However, unlike hunger or thirst, theories of such higher-order constructs such as goals and values are still unclear about the mental computational mechanism of how they are internally generated (for some exceptions, see Ballard, Palada, Griffin, & Neal, Reference Ballard, Palada, Griffin and Neal2021; Vancouver, Wang, & Li, Reference Vancouver, Wang and Li2020). For example, prominent theories, such as expectancy-value theory (Eccles & Wigfield, Reference Eccles and Wigfield2020) and the hierarchical model of achievement motivation (Elliot, Reference Elliot, Pintrich and Maehr1997), identified a number of factors that influence goals or values (e.g., implicit beliefs, personal characteristics, subjective perceptions, and affective states). However, these theories do not answer how these factors lead to the adoption of goals and values. In fact, the hierarchical model does not identify what kind of decision-making process is involved when one adopts mastery goals over performance goals. Expectancy-value theory does not specify how value is incorporated and represented into the existing knowledge structure. Like the case of needs and motives, we believe that specifying the computational mechanisms of goal/value adoption and transformation would provide a new landscape of understanding these concepts even better. In sum, although directional aspects of motivation are indeed less immune to our criticism, higher-order motivational concepts such as goals and values still have large room to benefit from the proposed perspective.Footnote ³

5.5. What about the evolutionary account of motivation?

Researchers often use evolution to justify motivation constructs, for example, people are motivated to understand the environment because this is crucial for survival. It is true that sets of behavior explained by need for competence can have survival value. But this perspective only addresses the why question (“why do people have motivation X”), not the what (“what is this motivation X?”) or how (“how motivation X is realized”) questions. In addition, an evolutionary perspective does not mean that we should keep motivation constructs a black box. In fact, the evolutionary perspective is open to the possibility that evolution shaped our mental computational processes in a particular manner. In other words, thinking about adaptive value and ecological constraints would help researchers specify more realistic mental computational processes (Anderson & Milson, Reference Anderson and Milson1989; Lieder & Griffiths, Reference Lieder and Griffiths2020). Therefore, evolution is not incompatible with our proposal that we should unpack the black-box property of high-level motivation constructs – evolution is, rather, a useful tool to unpack the black-box.