H&C label all models in cognitive science as “cognitive,” as indicated by their blanket use of the term “cognitive model.” Because of this labeling scheme, they are free to explore and rely on any modeling strategy to make their case. However, just because cognitive science is so-called does not mean that all models are cognitive models in particular and that every model should be subject to their requirement in general. This is more than a semantic complaint. Distinguishing model types matters because any view that categorizes all models as a single type will fail to capture the complexity of explanation in the mind sciences (and how those models may work toegether). Moreover, when type differences are ignored, the argument runs the risk of prescribing a monist modeling strategy that is tethered to a single explanatory framework. In H&C’s case: mechanistic. Such an approach is empirically inadequate and overly restrictive as cognitive science relies on a diverse set of modeling types to triangluate on the phenomenon of interest.
Cognitive models is a particular model type that takes a cognitive system as their intended target where the features included purport to characterize the system in a way that captures its cognitive functioning. These models describe a system in terms of representations, processes, and resources (Samuels, Reference Samuels2018; Weiskopf, Reference Weiskopf and Kaplan2017). Cognitive models can be contrasted with mechanistic models, which represent a system in terms of its organization, demonstrating how component activities produce the phenomena of interest (Kaplan & Craver, Reference Kaplan and Craver2011). Canonical mechanistic models explain in terms of organized activities of and interactions among components, i.e., its “parts” (Craver, Reference Craver2006), while cognitive models focus on a functional explanation in terms of the process the system performs and/or what the mechanism may represent.
An important difference between cognitive models and mechanistic models is that the mechanistic strategy depends on the model making determinate claims about the localization of components (Weiskopf, Reference Weiskopf and Kaplan2017). Cognitive models don’t necessarily do this. Instead, they often lack commitments to particular spatial organization, even if the models include diagrams or texts that rely on spatial relations (Weiskopf, Reference Weiskopf2011). This difference is critical when it comes to H&C’s proposal that metabolic considerations should be integrated into all models. The idea that all models must incorporate metabolism inevitably requires a particular type of description and focus that shapes the model into a form of mechanistic explanation. This consequence is demonstrated by their critique of Kurzban et al., as we will see.
H&C introduce metabolic considerations as a normative constraint on modeling where explanatory adequacy comes, in part, from whether it includes those details. To demonstrate the reasonableness of their proposal, the authors provide two examples of successful inclusion of metabolic considerations and then argue that the strategy can be generalized to all models.
The first example is the cyclical model of synaptic memory consolidation (Karbowski, Reference Karbowski2019). The cyclical nature of this model accounts for the energy exchange between the system and its environment. Karbowski is interested in the molecular events in dendritic spines and is explicitly characterizing the model in terms of a chemical mechanism. The second example (Jaras et al., Reference Jaras, Harada, Orchard, Maldonado and Vergara2021) describes the leaky integrate-and-fire (LIF) model of the action potential, but with a reference to metabolic features. The resulting energy-dependent LIF model (EDLIF) captures the differences that ATP makes to the reset value of the membrane potential.
Both the cyclical model and the EDLIF are clear-cut examples of a mechanistic modeling strategy that provides a mechanistic explanation. Given this, it is no surprise that the models easily integrate metabolic aspects. After all, mechanistic models can lie anywhere on a continuum between a mechanism sketch and an ideally complete description (Craver, Reference Craver2006). What H&C show is that you can always fill in more details.
Contrast these examples with H&C’s critique of Kurzban et al., who are interested in understanding the phenomenology of mental effort. Their model consists of a narrative of the problem space along with an abstract description of how cognitive demand requires a cost/benefit analysis. This model (given as a description) is meant to explain the experiences of boredom and mental effort associated with vigilance tasks. As a bonafide cognitive model, it explains mental effort in terms of representational processes without localizing the posited mechanism within the brain – it provides a functional description. In Marrian terms, their strategy is deliberately silent on implementation details. However, this does not mean that they ignore them entirely.
In their artictle, H&C specifically take issue with a statement about the model given by the authors in another BBS target article. In a reply to H&C’s commentary, Kurzban et al. state that they agree that the brain is constrained by biophysical thermodynamic laws but disagree that such factors should be incorporated into their explanation (Kurzban et al., Reference Kurzban, Duckworth, Kable and Myers2013, p. 714). H&C argue that this is a problem because all models should explicitly reference metabolic features (their normative constraint). Because Kurzban et al. refuse to acquiesce, their model must be rejected. But notice what’s really going on. H&C actually reject the explanatory framework of cognitive modeling in favor of a mechanistic-style explanation. Indeed, incorporating metabolic details would transform the Kurzban model into a kind of cognitive–mechanistic hybrid model, if not mechanistic entirely. If this is the strategy that H&C want to take for all models, then cognitive models will be slowly picked off and either replaced or transformed into mechanistic models.
Properly distinguishing model types reveal that H&C’s metabolic constraint fails to capture the diverse modeling strategies operating within cognitive science. Once modeling types are made explicit, it becomes difficult to justify a metabolic constraint on all models. If, however, one insists on restricting model types based on the inclusion of metabolic constraints, it’s worth considering whether all there is to cognitive science is mechanistic explanation. My take is that there’s more to it than that (Williams, Reference Williams2023; Williams, Reference Williams2025).
H&C label all models in cognitive science as “cognitive,” as indicated by their blanket use of the term “cognitive model.” Because of this labeling scheme, they are free to explore and rely on any modeling strategy to make their case. However, just because cognitive science is so-called does not mean that all models are cognitive models in particular and that every model should be subject to their requirement in general. This is more than a semantic complaint. Distinguishing model types matters because any view that categorizes all models as a single type will fail to capture the complexity of explanation in the mind sciences (and how those models may work toegether). Moreover, when type differences are ignored, the argument runs the risk of prescribing a monist modeling strategy that is tethered to a single explanatory framework. In H&C’s case: mechanistic. Such an approach is empirically inadequate and overly restrictive as cognitive science relies on a diverse set of modeling types to triangluate on the phenomenon of interest.
Cognitive models is a particular model type that takes a cognitive system as their intended target where the features included purport to characterize the system in a way that captures its cognitive functioning. These models describe a system in terms of representations, processes, and resources (Samuels, Reference Samuels2018; Weiskopf, Reference Weiskopf and Kaplan2017). Cognitive models can be contrasted with mechanistic models, which represent a system in terms of its organization, demonstrating how component activities produce the phenomena of interest (Kaplan & Craver, Reference Kaplan and Craver2011). Canonical mechanistic models explain in terms of organized activities of and interactions among components, i.e., its “parts” (Craver, Reference Craver2006), while cognitive models focus on a functional explanation in terms of the process the system performs and/or what the mechanism may represent.
An important difference between cognitive models and mechanistic models is that the mechanistic strategy depends on the model making determinate claims about the localization of components (Weiskopf, Reference Weiskopf and Kaplan2017). Cognitive models don’t necessarily do this. Instead, they often lack commitments to particular spatial organization, even if the models include diagrams or texts that rely on spatial relations (Weiskopf, Reference Weiskopf2011). This difference is critical when it comes to H&C’s proposal that metabolic considerations should be integrated into all models. The idea that all models must incorporate metabolism inevitably requires a particular type of description and focus that shapes the model into a form of mechanistic explanation. This consequence is demonstrated by their critique of Kurzban et al., as we will see.
H&C introduce metabolic considerations as a normative constraint on modeling where explanatory adequacy comes, in part, from whether it includes those details. To demonstrate the reasonableness of their proposal, the authors provide two examples of successful inclusion of metabolic considerations and then argue that the strategy can be generalized to all models.
The first example is the cyclical model of synaptic memory consolidation (Karbowski, Reference Karbowski2019). The cyclical nature of this model accounts for the energy exchange between the system and its environment. Karbowski is interested in the molecular events in dendritic spines and is explicitly characterizing the model in terms of a chemical mechanism. The second example (Jaras et al., Reference Jaras, Harada, Orchard, Maldonado and Vergara2021) describes the leaky integrate-and-fire (LIF) model of the action potential, but with a reference to metabolic features. The resulting energy-dependent LIF model (EDLIF) captures the differences that ATP makes to the reset value of the membrane potential.
Both the cyclical model and the EDLIF are clear-cut examples of a mechanistic modeling strategy that provides a mechanistic explanation. Given this, it is no surprise that the models easily integrate metabolic aspects. After all, mechanistic models can lie anywhere on a continuum between a mechanism sketch and an ideally complete description (Craver, Reference Craver2006). What H&C show is that you can always fill in more details.
Contrast these examples with H&C’s critique of Kurzban et al., who are interested in understanding the phenomenology of mental effort. Their model consists of a narrative of the problem space along with an abstract description of how cognitive demand requires a cost/benefit analysis. This model (given as a description) is meant to explain the experiences of boredom and mental effort associated with vigilance tasks. As a bonafide cognitive model, it explains mental effort in terms of representational processes without localizing the posited mechanism within the brain – it provides a functional description. In Marrian terms, their strategy is deliberately silent on implementation details. However, this does not mean that they ignore them entirely.
In their artictle, H&C specifically take issue with a statement about the model given by the authors in another BBS target article. In a reply to H&C’s commentary, Kurzban et al. state that they agree that the brain is constrained by biophysical thermodynamic laws but disagree that such factors should be incorporated into their explanation (Kurzban et al., Reference Kurzban, Duckworth, Kable and Myers2013, p. 714). H&C argue that this is a problem because all models should explicitly reference metabolic features (their normative constraint). Because Kurzban et al. refuse to acquiesce, their model must be rejected. But notice what’s really going on. H&C actually reject the explanatory framework of cognitive modeling in favor of a mechanistic-style explanation. Indeed, incorporating metabolic details would transform the Kurzban model into a kind of cognitive–mechanistic hybrid model, if not mechanistic entirely. If this is the strategy that H&C want to take for all models, then cognitive models will be slowly picked off and either replaced or transformed into mechanistic models.
Properly distinguishing model types reveal that H&C’s metabolic constraint fails to capture the diverse modeling strategies operating within cognitive science. Once modeling types are made explicit, it becomes difficult to justify a metabolic constraint on all models. If, however, one insists on restricting model types based on the inclusion of metabolic constraints, it’s worth considering whether all there is to cognitive science is mechanistic explanation. My take is that there’s more to it than that (Williams, Reference Williams2023; Williams, Reference Williams2025).
Financial support
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
Competing interests
The author declares no competing interests.