The illusion of internal models in biological movement | European Journal of Applied Physiology | Springer Nature Link
The concept of internal models dominates contemporary theories of sensorimotor control, with researchers across neurosciences, specifically motor control, routinely explaining observed behaviors through computational representations that supposedly exist within the nervous system. In this perspective, I present a critical examination of internal model frameworks in sensorimotor control. I argue that representational approaches mischaracterize biological systems for several fundamental reasons: (1) Internal models require homuncular interpreters, creating infinite regress problems; (2) The purported neural implementations of internal models remain empirically elusive despite decades of research; (3) Biological movement systems exhibit multiscale, nonlinear, and non-Gaussian dynamics that fundamentally defy reduction to conventional computational representations; (4) Internal model frameworks implicitly depend on Cartesian dualism through their separation of the “controller” and “controlled;” (5) The framework is methodologically circular and largely unfalsifiable as virtually any behavior can be retroactively modeled as implementing some internal representation; and (6) Alternative frameworks based on ecological dynamics and self-organization can account for adaptive behavior without invoking representational assumptions. Instead of representational models, I propose that sensorimotor control emerges from the dynamic coupling between the organism and the environment across multiple spatial and temporal scales. By moving beyond the internal model paradigm, sensorimotor neuroscience can develop more powerful explanatory frameworks that better capture the emergent, context-sensitive properties of biological movement without invoking physiologically intractable computational metaphors.
The ubiquity of internal model thinking The seduction of internal models lies not in what they explain, but in how they flatter our engineering instincts—projecting computational order onto biological chaos and mistaking our metaphors for mechanisms. Sensorimotor control theory is deeply permeated with the concept of internal models. When a behavior is explained as utilizing neural representations of bodily dynamics, environmental properties, or the causal structure of the world, the so-called “internal model” is implicitly or explicitly invoked. This approach has become so ubiquitous that…
saved by
related reading
- Internal Models in Biological Controlpmc.ncbi.nlm.nih.gov
- Mechanistic Interpretability: A Challenge Common to Both Artificial and Biological Intelligencekempnerinstitute.harvard.edu
- The Case For World Models, Part I: The Neuroscientific Reasonwheremachinesthink.substack.com
- Computational and systems neuroscience: The next 20 yearspmc.ncbi.nlm.nih.gov
- Trading places: What happens when neuroscience turns into machine learning, and machine learning turns into neuroscience?thetransmitter.org
- Bridging the gap between theories of sensory cue integration and the physiology of multisensory neuronspmc.ncbi.nlm.nih.gov
- When Models Manipulate Manifolds: The Geometry of a Counting Tasktransformer-circuits.pub
- The World Inside Neural Networksgoodfire.ai
- Book Review: Surfing Uncertaintyslatestarcodex.com
- pdfopenreview.net
- Practically-A-Book Review: Byrnes on Tranceastralcodexten.com
- A Model of Biological Motion Perception from Configural Form Cuespmc.ncbi.nlm.nih.gov