This page would look a lot spiffier if you downloaded a standards compliant web browser.

A Framework for Interaction and Task Decomposition for Objects Emulating Agency Behavior
    Carnegie Mellon University School of Architecture PhD dissertation 2000
    abstract:
    Computational systems for decision support are typically stand-alone tools. These are often designed assistance with respect to a single aspect of the decision making. In a design process, where decision integral to the activity, designers use such computational tools to generate alternative solutions, simulate the behavior of the artifact being designed, and to produce design documents. Stand-alone provide design assistance, but not without pitfalls:

    • Each tool requires designers to commit to a schema of representation.
    In order to examine various aspects of a design, the same artifact would be represented
    according to the schema of the tool employed.
    • Interdependency among the various design aspects is rarely examined.
    A lack of a unified representation leaves such examination to the judgement of the designer.
    • Designers are often required to provide vast amounts of information even for the smallest
    • Designers seldom have access to the mechanism by means of which a tool internally decomposes design problem. Designers are, thus, deprived of opportunities to make decisions that may, incrementally, impact on the evaluation process.

    In an attempt to improve the efficiency of these tools, research groups undertook the task of developing for comprehensive design environments where multiple design tools share the same representation Designers use modeling and generative tools to produce a model of the artifact being designed, where simulate, according to some domain expertise, the behavior of the model within the same shared Such design environments are often described as multi-agent decision making environments. Agents designer(s) and/or the computational applications each of which encompasses a specific domain Agents interact and execute tasks to manipulate the design objects until the collective state of these deemed acceptable by the designer(s).

    In this thesis I introduce an enhancement to the design of computational assistant tools, mainly multi-agent design environments that use shared representation schemes. I propose an expansion agency to include design objects in which agents may interact with other agents in the execution related to the objects. I call this the objects as agents approach, where objects are selectively activated participate in design decision making sessions to execute tasks regarding their immediate design sense, an object-agent is a design object that is activated to perform tasks. I provide a framework object agent-based design environment, in which domain applications are global problem solving nodes, are local coordination and management nodes, and, collectively, the designer(s) act as coordinator judge. In this sense, the designer orchestrates this fine grained agent environment through incremental interactions until the model arrives at an acceptable design state. Within this framework, I address pertaining to the notion of agency in design such as autonomy, planning and interaction. Vital to such an object-agent-based design environment is the ability of an object-agent to manage assigned self initiated tasks. Managing tasks relies on the ability to decompose and delegate such tasks. I provide decomposition/aggregation mechanism to enable object-agents to manage their tasks. Such a mechanism provides the designer with access to a wealth of local decision making information.