A Framework for Scalable Cognition (original) (raw)
It is argued that the prospect of the emergence of a Global Brain as a planetary level Communication Command and Control system capable of demonstrating general intelligence depends on realizing a scalable cognitive process. We have described an agent based framework for scalable cognition by first defining cognition as the combination of two selective processes: Selection for relevance (attention mechanism) and selection for effective action. These selective processes are context sensitive and operate on events that mediate differences in the state of the agent’s environment. The structure of cognitive agents and the structure of the environment co-‐define each other and therefore co-‐emerge. The framework suggests that the up-‐scaling of the cognitive process is realized by the agent’s tendency to form cooperative coalitions. Every such coalition is in fact a super agent constructed from simpler constituent agents operating together in a collective cognitive process. Coalitions are formed and dismantled according to their relevance. The relevance of a coalition or a super agent is derived from: 1. the existence of sufficient triggers from the environment to which they respond effectively according to a context sensitive set of criteria. 2. The extent by which they influence other coalitions and participate in higher level coalitions. 3. A Decay factor that basically represents entropy and the tendency of ordered systems to disintegrate in time. The concept of challenge is introduced into the framework as synonymous with context sensitive items of relevance that are selected by the attention mechanism. These items are analogous to the items ‘brought to consciousness‘ in Baars’ global workspace theory. As attention is spreading among agents, we say that challenges propagate within the population of agents along paths of influence that together form a network of influence. The propagation of challenges is analogous, at least in some aspects, to the monetary flow within a market system. Such flow abstracts the local context sensitive transactions and highlights instead the flow of the currency of attention. This is based on the understanding that the currency of attention ‘buys’ the resources necessary for effective action. Effective action in turn gains influence that further draws attention. This is quite different from the flow of information among agents because we hypothesize that what drives the formation of higher cognitive structures is the spreading of challenges and not only the spreading of information. Of particular interest in this sense is that aspect of propagation we called vertical propagation of challenges. Vertical propagation is associated with the formation of higher cognitive functions and takes place as challenges at a certain level are combined through the interactions of agents to a challenge of a higher level. The final part of the paper explores general considerations and problems of modeling the framework and outlining initial directions for implementation. The approach is modular and divides the frameworkinto a few modeling problems: Modeling agents (the generic agent), modeling challenges and events (modeling the environment), modeling topological aspects and finally modeling coalitions and their formation which is the most critical aspect of the framework. All these are of course subjects to further research. The ultimate test of implementing the framework is the demonstration of general intelligence i.e. the spontaneous discovery of problems in the environment and the emergence of specific problem solving capabilities without the guidance of a designer. This is of course a very hard problem to begin with but this paper makes some conceptual headway in figuring how to get there.
Related papers
Loading Preview
Sorry, preview is currently unavailable. You can download the paper by clicking the button above.