Rational Cognition in OSCAR

Stuart Russell [14] describes rational agents as " those that do the right thing ". The problem of designing a rational agent then becomes the problem of figuring out what the right thing is. There are two approaches to the latter problem, depending upon the kind of agent we want to build. On the one hand, anthropomorphic agents are those that can help human beings rather directly in their intellectual endeavors. These endeavors consist of decision making and data processing. An agent that can help humans in these enterprises must make decisions and draw conclusions that are rational by human standards of rationality. Anthropomorphic agents can be contrasted with goal-oriented agents — those that can carry out certain narrowly-defined tasks in the world. Here the objective is to get the job done, and it makes little difference how the agent achieves its design goal. If the design goal of a goal-oriented agent is sufficiently simple, it may be possible to construct a metric the measures how well an agent achieves it. Then the natural way of evaluating an agent architecture is in terms of the expected-value of that metric. An ideally rational goal-oriented agent would be one whose design maximizes that expected-value. The recent work on bounded-optimality ([3], [15], [20], etc.) derives from this approach to evaluating agent architectures. However, this approach will only be applicable in cases in which it is possible to construct a metric of success. If the design goal is sufficiently complex, that will be at least difficult, and perhaps impossible. This paper will focus on anthropomorphic agents. For such agents, it is the individual decisions and conclusions of the agent that we want to be rational. In principle, we could regard an anthropomorphic agent as a special case of a goal-oriented agent, where now the goal is to make rational decisions and draw rational conclusions, but it is doubtful that we can produce a metric that measures the degree to which such an agent is successful in achieving these goals. It is important to realize that even if we could construct such a metric it would not provide an analysis of rationality for such an agent, because the metric would have to measure the degree to which the agent's individual cognitive acts tend to be rational. Thus it must presuppose prior standards of rationality governing individual cognitive acts. In AI it is often supposed that …