Robotics AI model AICON predicts human planning failures in clinical test

A team led by Michael Migacev and Vito Mengers from the Robotics and Biology Laboratory at TU Berlin applied AICON—a reactive gradient-descent framework originally designed for robotic manipulation—to the Tower of London test, a classic cognitive assessment used to diagnose planning deficits in Parkinson's disease, mild cognitive impairment, and stroke. Remarkably, AICON contains no lookahead planning mechanism and was never trained on human behavior, yet it predicted the fine-grained difficulty ordering of 24 test problems more accurately than traditional structural task parameters. In a leave-two-out generalization test, the model successfully predicted difficulty for held-out problems, demonstrating robustness.

Crucially, the study found a dissociation: AICON outperformed a standard planning baseline only for groups with reduced planning capacity, while the baseline better captured healthy controls. This pattern was anticipated by the original AICON paper, which noted that the model's failure modes resemble those of Parkinson's patients who struggle with goal hierarchies but not move counts. The findings suggest that as planning capacity decreases—whether due to neurological conditions or cognitive load—human behavior shifts from deliberate planning toward the reactive, gradient-driven mode that AICON models. This broadens AICON's relevance beyond robotics into neuroscience and cognitive psychology, hinting that the core abstraction of reactive dynamics may reflect fundamental principles of biological information processing.