In their target article, Haueis and Colaço acknowledge that there is a tension between the “abstract, formal” characterization of a computational physical system in terms of its medium-independence and a physical characterization of the same system which would include its metabolic properties. Their attempt to identify and address this tension is hampered, however, by their focus on Piccinini’s idiosyncratic notion of physical computation. Piccinini (Reference Piccinini2015) specifically attempts to define and individuate the computational properties of a physical system in terms of concrete mechanisms, without invoking the notion of a formal property. It is widely acknowledged, however, that such an account of physical computation cannot make sense of Haugeland’s (Reference Haugeland1985) concept of medium-independence, on which the formal procedure being executed by a physical system can be identified and described independently of any physical medium. (See Coelho Mollo (Reference Coelho Mollo2017), Drayson (Reference Drayson2025), Haimovici (Reference Haimovici2013), and Williams (Reference Williams2025) for further discussion).

The real tension for Haueis and Colaço, we take it, is that if a particular type of cognitive process can be individuated formally (e.g. algorithmically), then non-formal properties of a physical system which implements the formalism are irrelevant to the cognitive process qua formal computation. If Haueis and Colaço have shown that an appeal to metabolic properties of the physical system is necessary to identify the cognitive process in question, however, then they have demonstrated that there is a non-formal way to individuate that type of cognitive process. This would be an important result because it would show that the cognitive process is not computational (in the formal sense): details of its physical implementation are essential to its identity as that type of cognitive process.

Whether Haueis and Colaço will be able to achieve this result at an appreciable scope depends on their ability to secure a role for metabolic considerations in the evaluation of competing cognitive models. Haueis and Colaço show how this can be done when comparing models of synaptic memory storage. The case is compelling, but it is unclear how the lessons can be carried forward to broader models of memory. Their example is highly constrained: the cognitive capacity (memory), neural system (dentate gyrus), and computational operation (synaptic weighting) are fixed. This is rarely the case in the evaluative comparison of cognitive models. How to individuate capacities, operations, and vehicles is central to debates between cognitive models, and it is much less clear how to incorporate metabolism when one or more of these are in flux.

To illustrate this, consider memory models at a broader scale. Even granting that changes in synaptic weighting are the best model for synaptic memory storage, vehicle indeterminacy for memories remains. After all, a memory is not stored at a single synapse. How should changes in synaptic weighting be organized into vehicles of memory storage? Competing models often propose disparate structures or emphasize different properties of those structures, resulting in an indeterminacy of vehicles (Mace, Reference Mace2025b). Spike trains, ensemble level properties, and field potentials are all contenders, and competing models of each emphasize different properties. For example, debates persist about when rate, timing, and synchrony of spike trains are important, as well as whether those properties can be sampled from individual neurons in different trials (Cariani & Baker, Reference Cariani and Baker2022; Ebitz & Hayden, Reference Becket and Hayden2021). Without having worked out vehicle indeterminacy issues in advance, it is difficult to determine which non-generic metabolic considerations matter and which specific neural, cognitive, or computational system to apply them to.

Similar issues arise for thinking about the operation being performed by these changes in synaptic weight. It is memory storage, but should this storage be thought of as storage of the memory, or storage of the index for information distributed elsewhere? Synaptic structures in the dentate gyrus may be interpreted as storing representations about the past for later retrieval or providing an index to cortically stored engrams so that they can be activated in instances of memory retrieval (see Borzello et al., Reference Borzello, Ramirez, Treves, Lee, Scharfman, Stark, Knierim and Rangel2023). Experimental contexts generally permit a number of disparate ascriptions for what is represented and how (Mace, Reference Mace2025a). It is unclear how metabolic considerations can factor into adjudications about the operations of any particular structure, in which case, it is unclear whether metabolic processes should be considered.

The problem becomes even more pressing as the cognitive models under consideration scale up to the systems level and debates about cognitive ontology. Given Haueis and Colaço’s remarks about cognitive effort, it is clear that they intend to have such reach. It is, however, unclear whether metabolic considerations can get a substantive grip in debates at this level of remove from neural implementation. Models of episodic memory, for example, are often divided by which feature they consider primary: the storage and/or recreation of rich, experientially detailed representations of past events, or the storage and reactivation of discrete, event-specific representations. As part of his general approach to cognitive ontology, Khalidi (Reference Khalidi2023) argues that these should be collapsed into a single cognitive kind. Others (Robins, Reference Robins2024) argue for splitting them into two distinct kinds, episodic simulation and event memory, respectively.

It is difficult to see how metabolic considerations get much of an evaluative grip here. Neither denies that the underlying neural mechanisms are relevant for cognitive individuation. They instead reflect the orientation of much recent work in cognitive ontology that resists aligning cognitive taxonomy too closely with neural details (Krickel, Reference Krickel2024). The emphasis is instead on the complex causal properties of the overall cognitive system, and the broader environmental and sociocultural context in which it operates.

In debates about specific cognitive models in neuroscience, it thus remains unclear how strongly metabolic considerations can serve as constraints and how decisively metabolic considerations can settle debates. While metabolic considerations may improve individual models, whether those considerations can come to bear on determining which higher level properties are relevant for some cognitive process like memory storage remains to be seen. Further study of this topic would benefit from discussions of the relationship between mechanistic and non-mechanistic cognitive models and the individuation of cognitive capacities, states, and processes.