Flexibility versus Efficiency: the Dual Answer

A fundamental question that always arises in cognitive modeling is how to combine two contradictory constraints: the model should be as flexible as possible (to reflect human flexibility) and at the same time it should be maximally efficient. The decision proposed in this paper is the following. The space to be searched has to be restricted for computational reasons, but this restriction should not be done in advance. It should be dynamic and should reflect the particular situation encountered, i.e. it should reflect the dynamically evolving context. In this way making the computations contextsensitive they will be both flexible and efficient. A multi-agent cognitive architecture is proposed consisting of hybrid (symbolic/connectionist) micro-agents. The behavior of the system at the macro level emerges from the collective behavior of the micro-agents. The symbolic computation performed by the system emerges from the symbolic micro computations performed by the agents, while their “power” or “rate” depends on their connectionist activation levels. The activation distribution over the agents reflects the particular context. This architecture allows for a greater flexibility of the cognitive system while at the same time decreasing the complexity of computation.