A real‐world rational agent: unifying old and new AI

Abstract

Explanations of cognitive processes provided by traditional artificial intelligence were based on the notion of the knowledge level. This perspective has been challenged by new AI that proposes an approach based on embodied systems that interact with the real‐world. We demonstrate that these two views can be unified. Our argument is based on the assumption that knowledge level explanations can be defined in the context of Bayesian theory while the goals of new AI are captured by using a well established robot based model of learning and problem solving, called Distributed Adaptive Control (DAC). In our analysis we consider random foraging and we prove that minor modifications of the DAC architecture renders a model that is equivalent to a Bayesian analysis of this task. Subsequently, we compare this enhanced, “rational,” model to its “non‐rational” predecessor and a further control condition using both simulated and real robots, in a variety of environments. Our results show that the changes made to the DAC architecture, in order to unify the perspectives of old and new AI, also lead to a significant improvement in random foraging.

Reactive navigation through multiscroll systems: from theory to real-time implementation

“…Drawing inspiration from perceptual mechanisms of biological systems, machine perception researchers are starting to develop new perception schemes for roving robots. For example, Verschure and co-workers developed a perceptual scheme (Distributed Adaptive Control, DAC5) as a neural model for classical and operant conditioning Verschure and Althaus 2003). In DAC5 three tightly connected control layers are introduced: the reactive layer, the adaptive layer and the contextual layer.…”

Chained learning architectures in a simple closed-loop behavioural context

“…Specifically we are confronted with the question of how animals arrive at useful, reproducible and, hence, stable behavioural patterns, while they are at the same time able to learn something new. Recently Verschure suggested that such systems should contain several layers for control and learning: at the bottom a reactive layer performs pure reflex-based control, an adaptive layer above performs predictive learning much in the sense of classical or operant conditioning, and finally a top contextual layer carries out higher-level adaptation (the DAC architecture, Verschure and Althaus 2003). In our study we are concerned with the first two layers only.…”

Factors influencing the Development and Adoption of Knowledge Based Decision Support Systems for Small, Owner-Operated Rural Business

“…It is also important that the system have an explanation facility that can show the end user how it arrived at an answer. This explanation facility can be at the knowledge level as described by Verschure and Althaus (2003) and is typical of what is used in traditional AI applications. The knowledge level describes knowledge, goals and actions according to the principle of rationality.…”

“…The knowledge level describes knowledge, goals and actions according to the principle of rationality. Verschure and Althaus (2003) suggest that there are several problems with this, most of which can be brought back to the dependence of solutions on the a priori specification of rules and representations. This does not appear to pose a problem in this case, because traditional approaches to development relating to problem solving and planning as described by Newell (1990), cited in Verschure and Althaus (2003) as opposed to the new AI described by the same authors as 'aimed to solve problems in the real world, incorporating most biologically motivated principles in its solutions' are used in this domain of KBDSS development.…”