The influence of software agent reliability and failure consequences on reliance, use, and performance

摘要

Software agents are one of the more recent attempts to support computer users. The term software agent has been used to refer to various software entities, with different levels of functionality. Typically, these agents perform tasks for users, within a computing environment, with little or no user intervention. While this level of autonomy can be beneficial, independent functionality can be problematic as well. These difficulties have been documented for other forms of autonomous assistance such as automation technology and other forms of computer-based aiding.;Some of the potential problems cited for these aiding technologies result from the reduction of user control and the modification of user responsibility. As with other autonomous systems, software agents alter the user's role. In the case of software agents, such as interface agents, the user's role changes from worker to manager, or to one who delegates tasks to the agent. Certain expectations must be fulfilled in order for delegation to occur. Two factors that guide delegation decisions are the delegate's reliability and the consequences of delegate failure. As with any complex software, perfect performance cannot be expected. Therefore, developers of agent-assisted systems must carefully consider the types of applications for which agency is best suited, as well as appropriate agent capabilities or functionality.;This research examined the influence of agent reliability (performance) and failure consequences on whether users delegate tasks to an interface agent instead of performing the tasks themselves. The results of usage decisions on task performance were also studied. The goal was to identify and measure the effects of agent reliability and the consequences of error on agent use as demonstrated by task delegation, continued use, and (overall) task performance.;Research results suggest that when agent error is the primary source of failure consequences, the frequency of failure influences the usage decision. In instances where failure consequences are experienced prior to agent failure, these consequences influence the agent use decision. Relatively small differences in agent functionality (as well as failure consequences) are noticeable and influence perceptions of agent reliability. However, if agent support is considered beneficial, some agent failure is tolerated.