Redesigning the Agents’ Decision Machinery

Abstract

Abstract. In a multi-agent system, agents must decide what to do and by what order. Autonomy is a key notion in such a system, since it is mainly the autonomy of the agents that makes the environment unpredictable and complex. From a user standpoint, autonomy is equally important as an ingredient that has to be used with parsimony: too much leads to agents fulfilling their own goals instead of those of the user, too little renders agents that are too dependent upon user commands and choices for their execution. Autonomy has a role in deciding which are the new goals of the agent, and it has another in choosing which of the agent's goals are going to be addressed next. We have proposed the BVG (Beliefs, Values, Goals) architecture with the idea of making decisions using multiple evaluations of a situation, taking the notion of value as central in the motivational mechanisms in the agent's mind. The agent will consider the several evaluations and decide in accordance with its goals in a rational fashion. In this paper we extend this architecture in three different directions: we consider the source of agent's goals, we enhance the decisional mechanisms to consider a wider range of situations, and we introduce emotion as a meta-level control mechanism of the decision processes.

On How to Conduct Experimental Research with Self-Motivated Agents

“…In [1,3,2], we have defended a model of choice that depends on the idea of multi-dimensional evaluation of a choice situation to accomplish an enhanced adaptivity to a dynamic and complex environment. These dimensions (which we have called values), are used to select which goals to pursue (through the information of mechanisms for goal adoption), and also which sub-goals are to be preferred and selected for execution.…”

Improving Choice Mechanisms within the BVG Architecture

“…In the model proposed in [3], emotions have a determinant role in the control of the choice process (see fig. 1).…”

“…We sustain that the multiple values approach represents a step towards autonomy. Here we will be more concerned with executive autonomy [6], in [3] we further addressed "cognitive mediation" issues, by proposing a mechanism for value-based goal acquisition, thus assuring goals autonomy. Autonomy should pay no tribute to a super-imposed rationality: autonomy means exactly the liberty to adopt one's own rationality.…”

“…Decision if often time-dependent, and an additive form for the choice function allows for the process to be interrupted (or even pre-emptive) and some anytime character to be achieved (cf. [3]). On the other hand, mechanisms must be included to avoid ties between alternatives (for instance, in the case of major absence of evaluative information).…”

“…The agent adopts and selects for execution the first solution that meets all of the aspiration levels. In a similar line of reasoning, we have addressed the issue of choice, as one of the central components in the agent's decision machinery [1,2,3]. We have proposed the use of multiple values to assess a decision situation.…”