The Understanding Capacity and Information Dynamics in the Human Brain.

This article proposes a theory of neuronal processes underlying cognition, focusing on the mechanisms of understanding in the human brain. Understanding is a product of mental modeling. The paper argues that mental modeling is a form of information production inside the neuronal system extending the reach of human cognition "beyond the information given" (Bruner, J.S., Beyond the Information Given, 1973). Mental modeling enables forms of learning and prediction (learning with understanding and prediction via explanation) that are unique to humans, allowing robust performance under unfamiliar conditions having no precedents in the past history. The proposed theory centers on the notions of self-organization and emergent properties of collective behavior in the neuronal substrate. The theory motivates new approaches in the design of intelligent artifacts (machine understanding) that are complementary to those underlying the technology of machine learning.

Situational Understanding in the Human and the Machine.

The Air Force research programs envision developing AI technologies that will ensure battlespace dominance, by radical increases in the speed of battlespace understanding and decision-making. In the last half century, advances in AI have been concentrated in the area of machine learning. Recent experimental findings and insights in systems neuroscience, the biophysics of cognition, and other disciplines provide converging results that set the stage for technologies of machine understanding and machine-augmented Situational Understanding. This paper will review some of the key ideas and results in the literature, and outline new suggestions. We define situational understanding and the distinctions between understanding and awareness, consider examples of how understanding-or lack of it-manifest in performance, and review hypotheses concerning the underlying neuronal mechanisms. Suggestions for further R&D are motivated by these hypotheses and are centered on the notions of Active Inference and Virtual Associative Networks.

Generative models, linguistic communication and active inference.

This paper presents a biologically plausible generative model and inference scheme that is capable of simulating communication between synthetic subjects who talk to each other. Building on active inference formulations of dyadic interactions, we simulate linguistic exchange to explore generative models that support dialogues. These models employ high-order interactions among abstract (discrete) states in deep (hierarchical) models. The sequential nature of language processing mandates generative models with a particular factorial structure-necessary to accommodate the rich combinatorics of language. We illustrate linguistic communication by simulating a synthetic subject who can play the 'Twenty Questions' game. In this game, synthetic subjects take the role of the questioner or answerer, using the same generative model. This simulation setup is used to illustrate some key architectural points and demonstrate that many behavioural and neurophysiological correlates of linguistic communication emerge under variational (marginal) message passing, given the right kind of generative model. For example, we show that theta-gamma coupling is an emergent property of belief updating, when listening to another.

Life and Understanding: The Origins of "Understanding" in Self-Organizing Nervous Systems.

This article is motivated by a formulation of biotic self-organization in Friston (2013), where the emergence of "life" in coupled material entities (e.g., macromolecules) was predicated on bounded subsets that maintain a degree of statistical independence from the rest of the network. Boundary elements in such systems constitute a Markov blanket; separating the internal states of a system from its surrounding states. In this article, we ask whether Markov blankets operate in the nervous system and underlie the development of intelligence, enabling a progression from the ability to sense the environment to the ability to understand it. Markov blankets have been previously hypothesized to form in neuronal networks as a result of phase transitions that cause network subsets to fold into bounded assemblies, or packets (Yufik and Sheridan, 1997; Yufik, 1998a). The ensuing neuronal packets hypothesis builds on the notion of neuronal assemblies (Hebb, 1949, 1980), treating such assemblies as flexible but stable biophysical structures capable of withstanding entropic erosion. In other words, structures that maintain their integrity under changing conditions. In this treatment, neuronal packets give rise to perception of "objects"; i.e., quasi-stable (stimulus bound) feature groupings that are conserved over multiple presentations (e.g., the experience of perceiving "apple" can be interrupted and resumed many times). Monitoring the variations in such groups enables the apprehension of behavior; i.e., attributing to objects the ability to undergo changes without loss of self-identity. Ultimately, "understanding" involves self-directed composition and manipulation of the ensuing "mental models" that are constituted by neuronal packets, whose dynamics capture relationships among objects: that is, dependencies in the behavior of objects under varying conditions. For example, movement is known to involve rotation of population vectors in the motor cortex (Georgopoulos et al., 1988, 1993). The neuronal packet hypothesis associates "understanding" with the ability to detect and generate coordinated rotation of population vectors-in neuronal packets-in associative cortex and other regions in the brain. The ability to coordinate vector representations in this way is assumed to have developed in conjunction with the ability to postpone overt motor expression of implicit movement, thus creating a mechanism for prediction and behavioral optimization via mental modeling that is unique to higher species. This article advances the notion that Markov blankets-necessary for the emergence of life-have been subsequently exploited by evolution and thus ground the ways that living organisms adapt to their environment, culminating in their ability to understand it.