Updated January 2022. Check out our other EDs here
Bradly Alicea, Morgan Hough, Amanda Nelson, and Jesse Parent
A classic idea from the cybernetics literature is the good regulator approach first formulated by Conant and Ashby in 1970, who proposed a formalism for good regulation. The Every Good Regulator Theorem (EGRT) provides a unique perspective on intelligent autonomous learning systems reliant on a type of compressed global representation (world model). We will discuss the concepts of modeling and regulation in the original EGRT, requiring a revisitation of the historical and technical underpinnings relevant to regulating a system with communication, equilibrium, and feedback. A homeomorphic mapping between controller and the controlled system (or model) provides a reduced representation that preserves useful variety for all possible outcomes of a system. Several toy models challenge the notion of tightly-coupled good regulation, and demonstrate how diverse models of physical systems can address the challenges of far-from-equilibrium and out-of-distribution phenomena. Of particular interest are learning systems that utilize physical phenomena such as diffusion, criticality, phase transitions, rotational forces, and bifurcation. The EGRT is then connected to a sampling of approaches and trends in machine learning (ML), deep learning (DL), and reinforcement learning (RL). We aim to recast the EGRT as a modern tool for ML and RL architectures by considering the role of good regulation and complexity in understanding the performance of intelligent systems.
Bradly Alicea
Naturalistic cognition in human performance is defined by dynamical responses to stimuli. Allostasis Machines (AMs) are characterized by an internal model and corresponding output trajectory characterizing a generalized response to stresses and sudden changes. The effects of the environment on the internal model are collectively known as perturbations, with a generalized response analogous to allostatic load. AMs consist of a sensory input, an internal model, a source of environmental perturbation, and an dynamical output that represents the response to perturbation over time. These dynamical output trajectories characterize this response either by recovering from perturbation (well-matched, ergodic), or drifting to a new stable state (accommodative, non-ergodic). We construct a quantitative model of AMs and consider their behaviors in a variety of scenarios, including isolated, serial, and new state perturbations. Control-theoretic strategies and multi-scale information processing can also be employed to provide AM models with more sophisticated feedback and control mechanisms. Understanding the difference between well-matched responses (stably matching environmental states) and allostatic drift (hysteretic responses to perturbation) clarifies how nonlinear responses produce continuous stability.
Robert Stone and Bradly Alicea
In this paper, we will attempt to bring the concept of the Every Good Regulator Theorem (EGRT) into the modern scientific discourse. The EGRT, first proposed 50 years ago by Conantand Ashby, will be connected to a broader set of concepts that are of interest to a modern, interdisciplinary set of issues. We will also engage in a discussion on how the EGRT relates tomodels of game theory and the mind in ways that recast so-called “good” regulation as a form ofcognition. Yet we also turn the very notion of cognition on its head by relating thisnon-purposeful notion of generalized cognition to universal properties of regulation. In doing so, we propose specific models for very simple regulation (zeroth-order switching) and very complex regulation (cybernetic convolution architecture). To conclude, we reconsider the role ofmodeling itself in regulating a complex system, which may encourage people to reconsider therelevance of the EGRT and associated cybernetic approaches.
Robert Stone, Tom Portegys, George Mihkailovsky, and Bradly Alicea
The construction of an embryo from a single cell precursor is a highly complex process. Evolutionary emergence of the first embryos is even more complex, and involves both a transition to multicellularity along with the establishment of developmental mechanisms. We propose that embryogenesis relies on a community of cells conforming to a regulatory model of emergent multicellularity. This model draws together multiple threads in the scientific literature, from complexity theory to cybernetics, and from thermodynamic entropy to artificial life. All of these strands come together to inform a model of goal-oriented regulation for emergent structures in early life. This is an important step in the evolution of early life, as well as the emergence of complex life in the earliest habitats. Our model, called the cybernetic embryo, allows for a systems-level view of the embryogenetic process.