Personal software assistants that help users with tasks like finding information, scheduling calendars, or managing work-flow will require significant customization to each individual user. For example, an assistant that helps schedule a particular user’s calendar will have to know that user’s scheduling preferences. This paper explores the potential of machine learning methods to automatically create and maintain such customized knowledge for personal software assistants. We describe the design of one particular learning assistant: a calendar manager, called CAP (Calendar APprentice), that learns user scheduling preferences from experience. Results are summarized from approximately five user-years of experience, during which CAP has learned an evolving set of several thousand rules that characterize the scheduling preferences of its users. Based on this experience, we suggest that machine learning methods may play an important role in future personal software assistants.

Agents are becoming widespread in a variety of computer systems and domains, but often appear to have little in common with each other. In this paper we look at different agent systems and identify what a generic agent should be composed of. We also identify the characteristics of a task that make it worthy of an agentbased approach. We then discuss the implications for the interaction of using agents, that is, the notion of a balanced interface, and briefly look at how an agent-based approach assists in two very different application domains.

. To be highly useful, office automation systems require customization to individual users, work environments, and tasks. We consider the question of whether office automation systems can be developed that allow users who are not skilled programmers to easily "program", or customize, these tools themselves. In other words, can we deploy very general, non-customized programs that can be extended and adapted during usage? Besides the well-known knowledge acquisition and automated learning methods we propose an approach called dialog-based learning (DBL), that allows the user to teach the system directly while performing the task. It acquires knowledge through a dialog in which the user both illustrates the procedure using a grounded example, and provides instructions about how to perform the task in the general case. We focus discussion on CAP II (a program that schedules meetings by negotiation via email), and on RAP (a program that makes room reservation via email). 1 Introduction We ...

The calendar apprentice CAP II, a learning agent that supports his user in scheduling meetings.

This report attempts to show that there is a fundamental need for the creation of a software medium that aids interaction of human and software agents by actively supporting various communication styles (including observation and manipulation). This need is identified by examining trends in the humancomputer interaction and software agent literature. The report then goes on to present a novel specification for such an operating medium and two currently fictional agent-based system scenarios that could be implemented in it. The intention of the proposed thesis work is to create a prototype of this medium and to implement the scenarios using it to support the necessary communication. The object of this is to demonstrate: firstly, that the operating medium actively supports the communication styles required, and secondly, that the agent-based interaction implemented via the medium is a useful way of supporting the tasks described in the scenarios, and other similar ones.

Programming by Demonstration (PbD) is a new programming method in which new functionalities are coded by simply showing them to a PbD system. The prospects of possibly successful applications in the domains of text and graphics editors, user interfaces, personal assistants, or robot programming are explored by an increasing number of researchers. The goal is twofold: First, the programming of a complex task can be simplified using PbD techniques. Thus, programming time can be reduced and complex systems also become programmable by non-experts. Second, the PbD systems enable machines or programs to adapt to their users and the work environment (auto-customization). Providing an integrated PbD component, a system can be deployed with only a few limited features, but acquire enhanced functionality later. By using two examples from the graphics domain (ProDeGE+: Programming by Demonstration Graphics Editor) and the robotics domain (RPD: Robot Programming by Demonstration), the general prin...