Designing agents for systems with adjustable autonomy
2001 (English)Doctoral thesis, monograph (Other academic)
Agents are an artificial intelligence technique of encapsulating a piece of pro-active, autonomous, intelligent software in a module that senses and acts in its environment. As the technology underlying sophisticated multi-agent systems improves, such systems are being deployed in ever more complex domains and are being given ever more responsibility for more critical tasks. However, multi-agent technology brings with it not only the potential for better, more efficient systems requiring less human involvement but also the potential to cause harm to the system's human users. One way of mitigating the potential harm an intelligent multi-agent system can do is via the use of adjustable autonomy. Adjustable autonomy is the idea of dynamically changing the autonomy of agents in a multi-agent system depending on the circumstances. Decision making control is transferred from agents to users when the potential for costly agent errors is large.
We believe that the design of the agents in a multi-agent system impacts the difficulty with which the system's adjustable autonomy mechanisms are implemented. Some features of an agent will make the implementation of adjustable autonomy easier, while others will make it more difficult. The central contribution of this thesis is a set of guidelines for the design of agents which, if followed, lead to agents which make adjustable autonomy straightforward to implement. In addition, the guidelines lead to agents from which it is straightforward to extract useful information and whose autonomy may be changed in a straightforward manner.
The usefulness of the guidelines is shown in the design of the agents for two systems with adjustable autonomy. The first system is EASE, which is used for creating intelligent actors for interactive simulation environments. The second system is the E-Elves which is a multiagent system streamlining the everyday coordination tasks of a human organisation. An evaluation of the two systems demonstrates that following the guidelines leads to agents that make effective adjustable autonomy mechanisms easier to implement.
Place, publisher, year, edition, pages
Linköping: Linköpings universitet , 2001. , 276 p.
Linköping Studies in Science and Technology. Dissertations, ISSN 0345-7524 ; 724
IdentifiersURN: urn:nbn:se:liu:diva-35601Local ID: 27906ISBN: 91-7373-207-9OAI: oai:DiVA.org:liu-35601DiVA: diva2:256449
2001-12-14, Visionen, Hus B, Linköpings universitet, Linköping, 13:15 (Swedish)