It is time to integrate models across disciplines: a commentary on Krüger et al. (2022)

In their target article, Krüger et al. (2022) provide a comprehensive review of the many motor imagery facts that have been gathered in recent years. One point that strikes us is a lack of convergence between different theoretical frameworks. As is currently the case in many areas of cognitive (neuro)science and psychology, what we have is a list of findings that we can sort according to essentially atheoretical criteria, such as “representations” and “experience and expertise.” Yet, some reliable phenomena are beginning to emerge from these collections. For example, the shared assumption across many accounts in sensorimotor control that sensory predictions play a pivotal role in the planning and control of movements may encourage new efforts to mediate between models, approaches, or frameworks that are currently separate from each other. Such thoughts appear in the review, e.g., in brief comments on the (supposed) incompatibility between ideomotor theory and forward models but are not developed further.

Theories and frameworks usually emerge in circumscribed fields and focus on issues that are relevant to the respective community. Accordingly, they often lack the ability to directly address questions currently asked in other fields. Consequently, theories are often mentioned but not really adopted and extended by other fields. We would like to acknowledge two recent examples––originating from the field of motor learning––that have attempted to integrate findings from different domains into a unified account. Of course, our selection reflects a subjective choice and is not meant as an exhaustive list. First, phenomena from different learning paradigms (classic conditioning, episodic memory, instrumental learning, and motor adaptation) have been connected within an overarching formal Bayesian model which puts the inference of context at the heart of learning (Heald et al., 2023). Second, some puzzling findings in using cues for context switching in motor adaptation (Gippert et al., 2023; Howard et al., 2020) have been explained by falling back on general characteristics of associative learning (Avraham et al., 2022). Independent of whether these works turn out to hold against further experimental scrutiny, they demonstrate how reaching beyond the concepts of a circumscribed field can stimulate new and integrative theorizing.

We envision that motor imagery, too, could serve as a hook for such integration across theories and fields. In other words, the motor imagery phenomena summarized by Krüger et al. (2022) could trigger more dedicated theory development.

For instance, as pointed out by the authors, ideomotor theory accounts only for goal, but not movement-related sensory information. The theory is based on the idea that the prediction of sensory goal states triggers the associated movements that realize them. This stands in contrast to the sensory predictions output by a forward model, which also encompass the sensations experienced during movement execution to achieve that goal state. It seems straightforward to expand the concept of goal state representations to integrate movement-related sensory information into this associative conglomerate. Notably, a debate about the relationship between goal states and movement detail exists also in sensorimotor neuroscience, cast as a (potential) dichotomy of action selection vs. action specification (Cisek & Kalaska, 2010). There is no consensus about whether multiple possible actions are present in brain activity only with respect to their attainable goal or whether also the respective underlying, fully specified movement plans are available for each potential action (Cisek & Kalaska, 2005; Dekleva et al., 2018). Could body-specific influences on motor imagery phenomena––which suggest the existence of fine-grained, movement-related details in thought (e.g., de Lange et al., 2006; Willems et al., 2009)––help sharpen theoretical concepts and stimulate attempts to bridge between these theories?

Similarly, motor imagery research might help explaining unresolved issues in the influential optimal feedback control (OFC) theory, a computational motor control framework (Scott, 2016; Shadmehr & Krakauer, 2008; Todorov & Jordan, 2002). The OFC theory emphasizes inherent links between perception and action and posits that the brain continually estimates the state of body and world by combining sensory feedback from different modalities and predictions about the sensory consequences of the motor command (Miall & Wolpert, 1996; Wolpert et al., 1995). One important aspect of OFC is that responses to potential movement perturbations are set prospectively, with varying settings across the time of the movement (Diedrichsen et al., 2010; Franklin & Wolpert, 2011). Yet, these models lack explicit integration of body-related processing (Martel et al., 2016; Medendorp & Heed, 2019). Could one exploit the fact that motor imagery removes real feedback and leaves only internally computed predictions to tease apart some of the concepts that are currently underspecified in OFC? Moreover, OFC models do not account for asynchronous processing delays across different modalities in the context of rapid movement corrections (Cluff et al., 2015; Oostwoud Wijdenes & Medendorp, 2017). For example, some authors have postulated that multisensory integration in online control uses both sensory variability and temporal delays (Crevecoeur et al., 2016); others, in contrast, have proposed that sensory feedback is initially processed in modality-specific channels (Keyser et al., 2023). From a very different but just as relevant angle, theories of body models emphasize the critical role of multisensory integration, as is well illustrated by the famous rubber hand and Pinocchio nose illusions (Botvinick & Cohen, 1998; Lackner, 1988), among many others. Might the constraints that human imagery abilities place on its implementation help furnish new ideas about how to link the currently disjunct areas of OFC, body models, and multisensory integration?

Finally, motor imagery research might advance theoretic accounts on the neurophysiological basis of sensorimotor control. Research on the neural dynamics underlying the control of movement based on multielectrode array recordings in nonhuman primates has fostered the idea that primary motor cortex can be adequately characterized from a dynamical systems perspective (Shenoy et al., 2013). Accordingly, preparatory activity involves setting an initial neural state specific to the upcoming movement, from which the movement will then mechanistically evolve. In support of this view, planning, but not executing, different follow-through movements allows simultaneous learning of opposing perturbations over the same movement (Sheahan et al. 2016). Importantly, merely imagining such follow-throughs also resulted in substantial learning (Sheahan et al. 2018), suggesting that motor imagery evokes different neural states for the same physical states, from which differentiated motor learning can emerge.

In summary, the rich, multisensory cognitive content elicited by motor imagery poses multiple challenges to existing theories, clearly indicating that we have not yet understood how body representation and control really work. At the same time, these challenges could foster new efforts to extend and bridge between currently isolated theories and, thus, foster true scientific advance.