Representational Realism is not the Default Position of Cognitive Neuroscience

Abstract Mental representations figure prominently in cognitive neuroscience, and this prominence is often taken to imply an underlying commitment to robust representational realism: the view that representations are real neural entities bearing determinate content grounded in their physical properties. The present paper challenges this assumption on two fronts. First, I argue that robust realist approaches face principled limitations, such as overgeneralization, and struggle to accommodate the diversity and context-sensitivity of neuroscientific data. Second, I claim that robust realism is neither required nor typically presupposed in scientific practice, as researchers rarely adopt such strong ontological commitments. Moreover, I present an alternative position, representational pragmatism, which treats representations not as discrete entities but as idealized epistemic artefacts, namely modelling tools that enable intelligibility, guide experimentation, and support prediction. I contend that this approach is more consistent with both neuroscientific practice and empirical data. To illustrate these points, I articulate a set of widely accepted constraints on representational attribution and examine varitel semantics, a leading contemporary robust realist framework. I argue that it fails to meet these desiderata and cannot reconcile its commitments with empirical findings. This, I maintain, exemplifies a broader difficulty of robust realism: the fallacy of misplaced concreteness, whereby methodological abstractions are mistaken for literal features of neural architecture. Finally, I defend representational pragmatism as a more viable and scientifically grounded stance that also fares better on representational desiderata. I conclude that if there is a “default scientific position”, it appears more pragmatist than robust realist.