Theoretical strategies for an embodied cognitive neuroscience: Mechanistic explanations of brain-body-environment systems

ABSTRACT

Cognitive neuroscience seeks to explain mind, brain, and behavior. But how do we generate explanations? In this integrative theoretical paper, we review the commitments of the ‘New Mechanist’ movement within the philosophy of science, focusing specifically on the role of mechanistic models in scientific explanation. We highlight how this approach differs from other explanatory approaches within the field, showing its unique contributions to the efforts of scientific explanation. We then argue that the commitments of the Embodied Cognition framework converge with the commitments of the New Mechanist movement in a way that provides a necessary explanatory strategy available to cognitive neuroscience. We then discuss a number of consequences of this convergence, including issues related to the inadequacy of statistical prediction, neuroscientific reduction, the autonomy of psychology from neuroscience, and psychological and neuroscientific ontology. We hope that our integrative thesis provides researchers with a theoretical strategy for an embodied cognitive neuroscience.

KEYWORDS:

• Embodied cognition
• new mechanists
• cognitive neuroscientific theory
• scientific explanation

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Our target is ‘mainstream’ cognitive neuroscience – investigations that primarily use neuroimaging techniques in the study of behavior (e.g., see Cooper and Shallice, 2010). Our arguments apply whenever the effort is to understand the relationship between the brain and mind/behavior. Whether our arguments apply at the boundaries of these efforts (e.g., investigating artificial neural ensembles developed in vitro, computational modeling) is beyond the scope of the present argument.

² The New Mechanist literature continues to develop. Here we provide a summary necessary for our discussion of embodiment.

³ There are many computational models that have an architecture and can be said to be made of parts that interact, for instance deep convolutional neural networks or other connectionist architectures in computational neuroscience. However, many computational models are designed to make predictions in restricted domains (e.g., recognition memory tasks, object perception) and while they are often biologically inspired (nodes are like neurons) they are not designed to map to the parts of a (neural) system and therefore they remain most useful for making predictions (e.g., classifying images) and/or functionalist descriptions, not as mechanistic models. While the term computational model applies to a wide range of modeling approaches, only a subset of those can be said to be mechanistic in the sense highlighted here (see Pulvermüller, 2023, for elaboration).

⁴ Under the umbrella term 4e cognition, researchers also emphasize how we are embedded in environments, enacting actions in the service of maintaining well-being, and use the environment to extend our behavioral capacities (see Newen, De Bruin, and Gallagher, 2018 for an overview). Further, some authors focus on the role of grounding and situatedness in cognition (Robbins & Aydede, 2009). Importantly, it can be suggested that cognition began as a subfield in psychology with the publication of the first textbook by Neisser (1967); thus, if we want to characterize Embodied Cognition, we can look to the first textbooks and handbooks with the term ‘Embodied Cognition’ in the title (e.g., Chemero, 2009; Coello & Fischer, 2015; Fincher-Kiefer, 2019; Fischer & Coello, 2016; Robinson & Thomas, 2021; L. Shapiro, 2010, 2014), or to special issue treatments (Borghi & Pecher, 2011; Kiverstein & Clark, 2009; Mahon & Hickok, 2016; Pecher & Zwaan, 2005). There is ongoing philosophical discussion about category membership of these different approaches. However, because of the significant impact it has had on psychology and neuroscience specifically, here we will restrict our focus to the use of the term Embodied Cognition and its relevance for current practices within cognitive neuroscience. See Gallagher (2023) for a thorough discussion of relevant similarities and differences between these various perspectives.

⁵ We recognize that the respective theorists in this section may not appreciate our categorization of their work as being mechanistic or embodied and may bristle at our taxonomy; we also recognize that there are ongoing debates about the compatibility of different modeling approaches (e.g., whether dynamical systems are compatible with mechanistic approaches). We apologize if we do them disservice and don’t aim to solve these debates here. Rather, these authors have engaged with the literature on embodied cognition (even if to distance themselves from it) and we argue that the commitments of these models are consistent with the specific characterization of the embodied cognitive neuroscience that we provide here.

⁶ One could argue that in one way or another they all model ‘categorization’ behaviors. However, categorization might be synonymous with having a psychology (e.g., Millikan, 2017) and is often argued as the most fundamental of all behaviors (e.g., Edelman, 1989).

⁷ Note that some classic arguments for embodiment rely on appeals to substrate-independence (e.g., Clark & Chalmers, 1998), specifically, the principle of parity. However, we are suggesting that a purely functionalist argument from classic cognitive science to embodied cognitive science is not the best argument; rather, the best arguments are the mechanistic ones. That is, under some of the models presented here (e.g., active inference) we can understand artifacts as entering sensorimotor loops that play a mechanistic – and not simply a functional – roles in behavior (see Clark, 2023).

⁸ To be clear, we are not suggesting that folk psychological ontology is useless, and indeed if prediction was the exclusive goal of psychological and neural science, then there is much value in folk psychological concepts (Sterelny, 2003). Even non-specialists can predict behavior better than chance based on folk concepts (see Hoogeveen et al., 2020). Further, our suggestion does not detract from the usefulness of folk concepts as applied tools (e.g., using an attention task as a way to predict future performance in a specific occupational setting). Our point is simply that such empirical investigations do not help us explain anything (see L. Shapiro, 2019).

Additional information

Funding

Heath Matheson is supported by funding provided by the Discovery Grants program of the Natural Sciences And Engineering Research Council of Canada (NSERC).