Internal models—computational representations supposedly stored in the nervous system—dominate current theories of how animals control movement. This perspective argues that such representational approaches fundamentally mischaracterize biological systems for six reasons: they require an infinite regress of internal interpreters; decades of research have failed to find neural evidence for them; biological movement is nonlinear and multiscale, not reducible to conventional computations; they implicitly rely on Cartesian dualism by separating a controller from what is controlled; the framework is circular and unfalsifiable; and alternative ecological dynamics and self-organization explain adaptive behavior without representations. Sensorimotor control instead emerges from dynamic coupling between organism and environment across multiple scales.
David Marr's influential tri-level framework for cognitive neuroscience, which separates computational goals, algorithmic implementation, and physical realization, is no longer tenable. The framework assumes cognitive goals can be specified in advance, neural processes implement algorithms, and physical substrates are interchangeable. Evidence from slime molds, planarian flatworms, bioelectric networks, and degenerate neural circuits shows these assumptions are false. The framework has survived by expanding its terminology rather than revising its theory, a pattern resembling unfalsifiability. Tinbergen's four questions reveal that developmental and evolutionary explanations were excluded. The authors argue for treating physical dynamics, morphology, and information as co-constitutive of adaptive behavior, questioning whether such behavior should be called computation.