Participant: Augustus Bacigalupi
Affiliation: Institute for Augmenting Minds
Format: Presentation and Conversation
Themes: paradigm, praxis

Many theories of mind have been proposed, yet they often ignore the physical implementation of our only proof of concept for cognition, namely embodied animal brains. Any theory and implementation of embodied physical cognition must account for the physical constraints that an environment imposes on evolving and learning agents in practice. Reflection upon this praxis re-informs theory in a way that challenges the current paradigm of information, namely discrete and independent binary states. Although this paradigm has enabled scientific success for over half a century, it is the main reason complex systems such as cognition remain mysterious. This paper is inspired by classic information theory, but goes further to account for the distinct and inter-dependent nature of information as it exists in embodied cognitive agents.

Refinement is a concept defined to empirically test for autonomous adaptive behavior. Refinement provides the theoretical basis for understanding, creating, and testing synthetic autonomous agents, homologues to simple biological organisms. Refinement is defined by both internal complexity and the capacity to do work. In order to internally represent their environment, all sentient agents must structurally embody complexity. This internal structural bias not only represents many distinct patterns, it also embodies the inter-correlations between patterns. Internal complexity is a necessary pre-condition for increasing an agent’s chances of survival via adaptation. However, complexity doesn’t measure how the system dynamically achieves successive states of increased complexity. For that, a path function is required.

This path function, able to measure the dynamics from state to state, is the time integral of power, i.e. work. Arbitrary degrees of internal complexity are not useful to the adaptive agent if they do not increase the agent’s autonomous capacity to do increasingly useful and complex work in its environment. Complexity, although necessary, does not ensure that an agent’s behaviors are relevant to its environment. Adaptive work potential measures this relevance. When complexity and work increase, Refinement increases denoting increased chances for persistence.