When is a causal system explanatory enough to be termed a ‘mechanism’? In their recent Review, Ross and Bassett argue that the term ‘mechanism’ in neuroscience suffers from a lack of clear, consistent definition across different studies and contexts (Ross, L. N. & Bassett, D. S. Causation in neuroscience: keeping mechanism meaningful. Nat. Rev. Neurosci. 25, 81–90; 2024)1. These authors recommend using the term ‘mechanism’ only in its narrow, traditional sense, to refer to detailed, lower-scale causal systems involving specific physical and biological interactions. This usage aligns with historical and common scientific applications, in which mechanisms are expected to provide concrete, detailed causal explanations. For broader causal systems that may include larger-scale or abstract interactions, and do not typically involve detailed, lower-scale, and physical interaction-based explanations, the authors suggest using ‘mechanism’ with caution. Furthermore, the authors strongly recommend against using ‘mechanism’ to describe non-causal systems, such as descriptive or topological models that do not offer causal explanations.

The authors carefully argue against over-reductionism (though they do not use this term) and encourage flexibility and appropriateness when using the term ‘mechanism’ to describe a causal system of any scale. Philosophers have long argued against over-reductionism in the field of cognitive neuroscience2, and the same argument is applicable to the entire field of neuroscience, where neurons might otherwise be reducible to nothing but atomic interactions. Therefore, Ross and Bassett’s argument prompts one key follow-up question: which level is ‘appropriate’ enough to warrant the term ‘mechanism’? Should researchers using functional MRI be required to dig deeper to provide molecular evidence for any causal system that they identify before it is acknowledged as mechanistic?

Here we modify Ross and Bassett’s view and propose a principle of ‘causal prominence’ as a criterion for determining whether a specific level is mechanistically appropriate. Take car engines, for example. Although this complex machinery is mostly made from cogwheels and screws, removing any one particular screw may not have an immediately observable impact on its function, performance, or behavior. As such, a repairperson’s time is best directed towards more macro-scaled parts (such as a spark plug) that are more ‘prominent’, and no less causal, to the behavior of interest (such as ignition). Similarly, the level of appropriateness in neuroscience is largely dependent on the phenomenon of interest. For instance, working memory is underpinned by multiple causal systems, including activities in the dorsolateral prefrontal cortex and parietal cortex, as well as the connectivity between the two3. Although this is true both at the systems level (a relatively macro level) and neuronal level (a relatively micro level), we propose that the systems level warrants the use of the term ‘mechanism’ as it has a bigger and perhaps more discernable causal impact on the phenomenon of interest4. Thus, ‘causal prominence’ can be understood as a combination of causality from philosophy and effect size from statistics. This principle is consistent with Ross and Bassett’s view on micro-scale (still mechanistic) and non-causal (still non-mechanistic) systems, but adds a clearer guideline to help researchers determine the appropriateness of using the term ‘mechanism’ to describe macro, non-reductive, causal systems.

There is a reply to this letter by Ross, L. N. & Bassett, D. S. Nat. Rev. Neuroci. https://doi.org/10.1038/s41583-024-00839-5 (2024).