Computational Substrate Bias

Computational substrate bias refers to the systematic distortion introduced into theoretical frameworks when those frameworks are developed primarily through computational modeling on a specific class of hardware. Because digital computation on von Neumann machines imposes discrete address spaces, finite state, and sequential-or-parallel (but not truly continuous) processing, theories developed and tested through such modeling carry implicit commitments to the discretizable, boundary-stable, and finitely-representable — even when the phenomena being theorized have none of these properties.

The concept is relevant wherever theoretical fields rely heavily on simulation: Systems Theory, Computational Neuroscience, Agent-Based Modelling, Evolutionary Computation, and Artificial General Intelligence research all exhibit substrate bias to varying degrees. A model that cannot be efficiently simulated on available hardware tends to be abandoned in favor of one that can — not because the abandoned model is wrong, but because tractability and correctness are conflated under resource pressure.

Substrate bias is a specific case of Tool Bias in Science, the broader phenomenon by which the instruments available to a discipline shape what that discipline can conceive as a possible result.