This proposal was originally a short paper relating representations of intelligence between three fields: psychology, neuroscience and artificial intelligence (AI). I particularly emphasize the role of modularity in these three areas. To my knowledge, this paper was never published — it was written on commission, but several years ago and I have just done yet another web search to find it. Further,

Learning, like any search, is only tractable for situated, resource-constrained agents if it is tightly focused. Adaptation is only worth the risks inherent in changing a complicated intelligence if it is very likely to improve the agent's performance on its goal tasks. Modularity is one tool for providing the information a learning system needs: it facilitates the use of a specialized representation suitable to a particular learning task, and provides for specialized perception to inform that representation. This paper begins by examining why behavior-based artificial intelligence, a well-known modular theory of intelligent design, has not so-far been used systematically to support such an approach. It then describes a new design methodology, behavior-oriented design (BOD), which does. Examples, drawn from both mobile robotics and models of learning in non-human primates, show the sorts of information such an approach can support, including both explicit and implicit anticipatory representations.

The overall behavior and nature of complex natural systems is in large part determined by the number and variety of the mechanisms involved – and the complexity of their interactions. Embodied natural communication belongs to this class of systems, encompassing many cognitive mechanisms that interact in highly complex ways, both within and between communicating individuals, constituting a heterogeneous, large, densely-coupled system (HeLD). HeLDs call for finer model granularity than other types of systems, lest we risk them to be not only incomplete but likely incorrect. Consequently, models of communication must encompass a large subset of the functions and couplings that make up the real system, calling for a powerful methodology for integrating information from multiple fields and for producing runnable models. In this paper I propose such an approach, abstract module hierarchies, that leverages the benefits of modular construction without forcing modularity on the phenomena being modeled.

This paper describes a research program designed to use agent models to understand how primates incorporate new, learned behavior safely into their established, reliable behavior repertoires. We suggest that some of the findings in the primate learning literature that we currently find surprising may in fact reflect evolved solutions to the problem of safe learning in intelligent agents. We sketch an approach to trying to model this learning, with the expectation that our experience will also lead to insights into idioms and strategies for incorporating safe learning into artificial agents.

There is strong evidence in the literature for at least three di#erent representations underlying transitive choice in various species of animals. This paper focuses primarily on understanding one of the most neglected: the production-rule model of Harris and McGonigle (1994). The production-rule model has been to date the best model at accounting for the performance of squirrel monkeys (Saimiri sciureus) and human children under 7 trained on the transitive inference (TI) task (e.g. given A > B and B > C, then A > C) when presented with three items from the training set. This paper presents a new neurologically-plausible version of this representation, the two-tier model, which explains how this sort transitive performance is learned. This new model perfectly replicates the positive aspects of the productionrule model while accounting for more of the data, particularly subject's failures to learn transitive inference. This paper also discusses how the two-tier model fits with other transitive inference models, and characterises how to recognise which TI representation underlies which sorts of TI performance. Of the three representations discussed, we suggest that the two-tier model may be the most relevant for understanding general-purpose primate task learning, and that it may even provide the scaffolding for the human acquisition of concrete operational thought.

Present approaches for computing do not have the performance, flexibility and reliability of neural information processing systems. In order to overcome this, conventional computing systems could benefit from various characteristics of the brain such as modular organisation, robustness, timing and synchronisation, and learning and memory storage in the central nervous system. This overview incorporates some of the key research issues in the field of biologically inspired computing systems.

This dissertation presents Behaviour-Based Learning (BBL), a methodology, for developing rapidly adapting behaviours in Behaviour-Based (BB) robots. BBL deals with a set of current issues related to learning in robots, in particular: speed of adaptation, the use of domain knowledge, problem restriction, robustness and integration of programmability and adaptivity.

In this paper we focus on how instructions for actions can be modelled in a selforganising memory. Our approach draws from the concepts of regional distributed modularity and self-organisation. We describe a self-organising model that clusters action representations into different locations dependent on the body part they are related to. In the first case study we consider semantic representations of action verb meaning and then extend this concept significantly in a second case study by using actual sensor readings from our MIRA robot. Furthermore, we outline a modular model for a selforganising robot action control system using language for instruction. Our approach for robot control using language incorporates some evidence related to the architectural and processing characteristics of the brain (Wermter et al. 2001b). This paper focuses on the neurocognitive clustering of actions and regional modularity for language areas in the brain. In particular, we describe a self-organising network that realises action clustering (Pulvermüller 2003). 1.

variation in the degree of social tolerance by dominant individuals of subordinate ones, particularly in the context of resource acquisition, and variation in the degree to which relationships damaged by aggression are repaired via reconciliation (de Waal & Luttrell, 1989). Although it appears that this co-variation in conflict management mechanisms varies in predictable ways across species, it does not appear that the co-variation can be explained by ecological factors. Rather, the variation seems to be emergent from patterns of social interaction among individuals, and self-reinforced through social learning. The importance of social learning on styles of interaction was made clear by the results of a cross-fostering study of two macaque species the individuals of which have drastically different proclivities for aggression and reconciliation (de Waal & Johanowicz, 1993). In this study, juvenile rhesus macaques, which typically live in social systems characterized by high levels of

Introduction This is a short paper relating representations of intelligence between three fields: psychology, neuroscience and artificial intelligence (AI). I particularly emphasize the role of modularity in these three areas. Because space is limited, I will assume a general familiarity with modular AI architectures, and concentrate on relating them to natural intelligence . 2 Modularity in Psychology I will begin with an incredibly simple definition of modularity from the psychological literature, due to Flombaum et al. (2002): "Modularity is the thesis that the mind contains independent input systems that, when engaged, are restricted in the types of information that they can consult." This definition is useful for two reasons. First, it introduces a very clean criteria for modularity: that some part of the mind does not have access to some other part of the mind. Given this simple criteria, anyone who accepts the idea of implicit knowledge or unconscious action has already ac