Broad agents

Abstract

The Oz project at Carnegie Mellon is developing technology for dramatic virtual worlds. One requirement of such worlds is the presence of broad, though perhaps shallow, agents. To support our needs, we are developing an agent architecture that provides goals and goal directed reactive behavior, emotional state and its effects on behavior, some natural language abilities (especially pragmatics based language generation), and some memory and inference abilities. We are limiting each of these capacities whenever necessary to allow us to build a broadly capable, integrated agent.In attempting to construct a broad agent, constraints seem to arise between components of the architecture. In this brief note, we discuss some of these constraints.

What Sort of Architecture is Required for a Human-Like Agent?

“…a set of requirements) will always decompose into a collection of distinct capabilities which can be served by distinct components of a design, or whether there is always so much intricate "cross-talk" between requirements and between elements of designs that clean, intelligible, modular solutions will turn out to be impossible, except in relatively trivial cases. 1 Even if designs are unintelligible at one level of description, there may be higher level descriptions of important features which can be discovered if only we develop the right sets of concepts. Cohen and Stewart [3] suggest that this emergence of higher level order is a feature of all complex systems, including biological systems.…”

“…much of animal behaviour serves the needs of a community, or a gene-pool, rather than the individual.) This focus on the problem of combining a large number of diverse kinds of functionality, each of which may not (at first) be specified or modelled in much depth, has been dubbed the "broad and shallow" approach by the OZ group at Carnegie Mellon University [1].…”

SIM_AGENT: A toolkit for exploring agent designs

“…[9,6,17]) though these projects are still at a relatively early stage, and those involving complete physical robots include only a small subset of the motive management processes that interest us. Bates et al [3,2] attempt to implement 'broad and shallow' systems based on rules of a type that might emerge from the architecture (Fig. 1) that we are considering.…”

Linking perception and action through motivation and affect

“…There is a growing consensus among theorists and designers of complete intelligent systems (Minsky 1987, Franklin 2001, Sloman 2001, Minsky 2006) that synthetic minds, to be complete and believable, require a computational equivalent to emotion to complement their behavioural and cognitive capabilities. This need not be a deep model as the thesis behind the work on the OZ project (broad and shallow) demonstrates (Bates et al 1991, Reilly andBates 1993). This requirement has been highlighted by earlier prominent researchers (Simon 1979, Norman 1980 in their discussions on the nature of cognition in biological systems.…”