AI fails catastrophically while brains fail gracefully because of a fundamental difference in mathematical 'conditioning.'
April 14, 2026
Original Paper
The Stability Asymmetry across Biological Cognition & Artificial Neural Inference
SSRN · 6454438
The Takeaway
It posits that biological cognition is a 'well-posed convolution,' whereas AI is an 'ill-posed deconvolution.' This explains why tiny perturbations break neural networks while human brains remain functional even after physical lesions.
From the abstract
Brains degrade gradually under progressive lesion and may exhibit paradoxical lucidity, whereas artificial neural networks can fail catastrophically from imperceptible perturbations. We hypothesize that this divergence reflects the stability asymmetry between convolution and deconvolution: convolution is well-posed, while deconvolution becomes ill-posed when the forward filter attenuates signal below noise. We formulate an analogous asymmetry for nonlinear networks via Jacobian conditioning and