We concur with Murayama and Jach (M&J) that motivation is an emergent property of a collective dynamic system of interacting elements. However, the principles and the model these authors develop do not fall within the ontological and paradigmatic framework of dynamic systems and emergent phenomena. This ambiguity needs to be clarified as it has important implications for how motivational processes can and should be conceptualized and investigated.

By considering that their model lends itself to testing its various parts, as well as the classic antecedents and outcomes of motivation, M&J seem to conceptualize motivational processes as driven by component-dominant dynamics, that is, as decomposable into isolable parts (e.g., Hausman & Woodward, Reference Hausman and Woodward1999). However, according to the dynamic systems perspective, psychological phenomena are patterns that emerge from interaction-dominant dynamics (Den Hartigh, Cox, & van Geert, Reference Den Hartigh, Cox, van Geert, Magnani and Bertolotti2017; Van Orden, Holden, & Turvey, Reference Van Orden, Holden and Turvey2003; Wallot & Kelty-Stephen, Reference Wallot and Kelty-Stephen2018) that are non-decomposable and non-isolable (Bechtel & Richardson, Reference Bechtel and Richardson2010). Thus, the emergent properties of a dynamic system cannot be deduced from the properties of its components, just as the fluidity, viscosity, and transparency of water cannot be deduced from the aggregate properties of oxygen and hydrogen (Bunge, Reference Bunge1977).

Moreover, the principle of M&J's reward-learning model is a reinforcement loop consisting of a causal chain that unfolds among its components, with very few interactions to modulate the causal relationships. The system is self-boosting in that an interest-based engagement promotes a positive feedback loop that sustains long-lasting information-seeking behavior. Strictly speaking, this behavior cannot be considered emergent, since it can be predicted on the basis of the value of its immediate determinants in the causal chain. Unlike a causal chain, even in loop form, a dynamic system involves complex interactions among components, which lead – through a process of self-organization – to the emergence of a global behavior pattern for the system. This pattern tends to stabilize by contributing to the formation of an attractor landscape, which in turn constrains the states of the system's components and their interactions, and so forth (e.g., Kelso, Reference Kelso1995). This attractor dynamics implies non-proportionality between variations of the system's components and those of the emergent behavior, which results in nonlinear dynamics of that behavior. This nonlinear dynamics can account for some well-documented typical motivational patterns, such as persistence of effort despite negative experiences, oscillation between motivated and unmotivated states, and abrupt shifts in motivation following a tiny variation in one of its putative determinants (Carver & Scheier, Reference Carver and Scheier1998; Gernigon, Vallacher, Nowak, & Conroy, Reference Gernigon, Vallacher, Nowak and Conroy2015; López-Pernas & Saqr, Reference López-Pernas and Saqr2024). In its current form, M&J's feedback loop could neither explain nor simulate such dynamics.

How motivational processes are conceptualized has, in turn, important consequences for how they can be investigated. M&J consider mathematical formulations of mental computational processes to be useful, but neither necessary nor sufficient. Surprisingly, however, they do refer to van der Maas et al. (Reference van der Maas, Dolan, Grasman, Wicherts, Huizenga and Raijmakers2006) as a case example, whose dynamic model of the emergence of general intelligence is typically based on mathematical formalizations of interactions – governed by evolution rules – among many components that evolve over time. As the example of van der Maas et al. illustrates, patterns emerging from dynamic systems typically follow evolution rules that can be expressed mathematically, generally with logistic equations, or with coupled differential or difference equations (e.g., Guastello & Liebovitch, Reference Guastello, Liebovitch, Guastello, Koopmans and Pincus2009). Whether they are parameterized directly or indirectly via software interfaces, these equations model the underlying processes and can thus account for the emergence of higher-level motivational patterns. Unlike M&J's reinforcement loop, specific computer simulation methods, such as dynamic networks, dynamic field models, agent-based models, cellular automata, and genetic algorithms, are designed to implement the self-organization processes that lead to the emergence of particular psychological phenomena (Gernigon, Den Hartigh, Vallacher, & van Geert, Reference Gernigon, Den Hartigh, Vallacher and van Geert2024). Hence, these methods make it possible to observe and test how these phenomena identifiable at a higher-order level emerge from rules modeled at a lower-order level (Nowak, Reference Nowak2004; Smaldino, Reference Smaldino2023; Vallacher, Read, & Nowak, Reference Vallacher, Read and Nowak2017).

A reinforcement loop and the modeling of self-organization processes are also substantially different in terms of the type of prediction that can be tested. M&J contend that their “precise process model” can help researchers make more fine-grained predictions about how different types of assessments or manipulations result in different outcomes. This contention reflects an interventionist or manipulative conception of causality that is currently prevalent (e.g., Hausman & Woodward, Reference Hausman and Woodward1999), but which yields poorly reproducible results (Open Science Collaboration, 2015) and which cannot account for a process causality based on principles of emergence (e.g., Gernigon et al., Reference Gernigon, Den Hartigh, Vallacher and van Geert2024; van Geert & de Ruiter, Reference van Geert and de Ruiter2022). More realistically, though perhaps frustratingly for some researchers, the complexity of emergence processes and their idiosyncratic nature casts doubt on any promise of fine-grained prediction in terms of interventionist causality. The predictions permitted by process causality, on the other hand, concern typical statistical signatures of complexity and (nonlinear) dynamics that can be detected at the idiosyncratic level by specific time series analyses, such as Recurrence Quantification Analysis and Detrended Fluctuation Analysis, of both simulation and ecological data. In addition, dynamic models of individual cases are expected to yield, at population level, comparable descriptive statistics between simulation and ecological data. Ultimately, for the sake of convergent validity, a dynamic model of motivation should account for both the consistencies and inconsistencies of the field's literature (e.g., Gernigon et al., Reference Gernigon, Vallacher, Nowak and Conroy2015, Reference Gernigon, Den Hartigh, Vallacher and van Geert2024). In doing so, we may come one step closer to understanding how intra-individual processes can give rise to different motivational trajectories.

To conclude, we agree with M&J that an explanatory account of motivation requires a focus on the lower-level mechanisms that give rise to higher-order motivated behavior. The lens of dynamic systems is best suited to providing this focus, as it captures the complexity of motivational processes better than traditional approaches. However, embracing this perspective is a paradigmatic choice that is conceptually and methodologically constraining. While still to be made, this promising choice is within reach of M&J and other motivational researchers.