Machine Cognition, Control and Embodiment on Landscapes of Fog, Friction and Selection

Abstract

Real-world cognitive structures — embodied biological, machine or composite entities — are inherently unstable by virtue of the “topological information” imposed upon them by external circumstance, adversarial intent, and other persistent “selection pressures”. Consequently, under the Data Rate Theorem (DRT), they must be constantly controlled by embedding regulators. For example, blood pressure and the stream of consciousness require persistent delicate regulation in higher organisms. Here, using the Rate Distortion Theorem of information theory, we derive a form of the DRT of control theory that characterizes such instability for adiabatically stationary nonergodic systems and uncover novel forms of cognitive dynamics under stochastic challenge. These range from aperiodic stochastic amplification to Yerkes–Dodson signal transduction and outright system collapse. The analysis, deliberately closely adapted from recent purely biological studies, leads toward new statistical tools for data analysis, uncovering groupoid symmetry-breaking phase transition analogs to Fisher Zeros in physical systems that may be important for studies of machine intelligence under real-world, hence embodied, interaction. The challenges facing construction, operation, and stabilization of high-order “workspace” or “multiple-workspace” machine cognition, perhaps backed by rapid pattern-matching “emotional” AI, whether explicitly recognized as conscious or not, will require parallel construction of new analytic machinery. This work provides one example, solidly based on the asymptotic limit theorems of information and control theories.