Active Interfaces through Software Agents

This paper concerns the development of an interface environment to help users in repetitive tasks in o ce work. The main ideas in the project concern: the development of active interfaces that autonomously perform tasks minimizing the interaction with the user; the use of the agent-oriented paradigm to provide both distributedness and incrementality of the software environment. The paper quickly illustrates the main issues addressed in the project and shows how they are exploited in the development of an active interface for ltering e-mail messages. The architecture of the ltering agent follows a multi-agent implementation. 1 Interfacing Useful Software Tools The spreading of electronic instruments is getting common people more and more in touch with software tools not always easy to be handled with. Particular emphasis can be put on network tools and information sources and on their use for di erent services in both domestic and o ce contexts. Due to the world-wide di usion of computer networks, to the daily increasing number of users connected to them and to the quantity of services o ered, it seems that a lot of our engagements and amusements will be satis ed through the networks while we are comfortably sitting on our chair. This scenario is likely to become true only if interfaces designed to make acceptable and easy using these o ered tools are supplied. In some way, the employment of such tools should be user-transparent. In fact people could not immediately accept the introduction of machines the utility of which is not so clear since they change completely their way of facing (and solving) problems. Also when the users are competent in dealing with sophisticated software mechanisms, the context can change but the problems are similar. In fact, high-skilled people do not disdain an aid that would make their work easy, for example by saving their time. In fact, the full exploitation of services and data available through the Internet takes a lot of time, time that is stolen to their job. A di erent issue concerns the usually large amount of knowledge needed to actually use those tools. Moreover, since the number of di erent versions or releases increases very fast, and it is di cult to follows all the changes especially if the acquisition of any competence is hard. Both in novice and expert cases, and for all the possible intermediate cases, the interfaces with these tools have the great responsibility of making the access easier. If the use of computers must be really enlarged to non-specialist people, their management should be easy and the interface should take care to do something more than what the users strictly request. In fact, since the goal is to disengage the user as far as it is possible from learning the tool, the interface should be active, i.e., it should support the user in each phase of her interaction. It should suggest solutions and have deductive capabilities. In our project we are considering interfaces for useful software tools available on the network. A lot of utilities are accessible on public domain repositories which are useful to solve particular problems but also: (a) they are very powerful but not easily comprehensible. So if from one hand they save time in small tasks, on the other hand they require a lot of time to be learned; (b) most of the utilities have programmer-oriented interfaces and, as a consequence, they are mainly used by "hackers" or specialized programmers. Our work has been focussed on building interfaces that avoid users to learn much about a particular tool. In addressing this kind of problem we have restricted the application context to the development of interfaces for tedious and repetitive o ce tasks, as dealing with e-mail messages or scheduling meetings or looking for useful information on the network. These tasks take advantage from or are motivated by the existence of the network, but the wide use and success of tools solving them are strictly depended on their conjunction with e ective and supportive interfaces. In the project we follows two methodological goals: (a) building active interfaces that carry out some additional role with respect to being simple lters between users and applications. The concept of active interface has been introduced in several research in human-computer interaction (e.g., [5, 7]); (b) employing the agent-oriented paradigm to implement the interfaces in order to make the whole system both easily extensible and tailorable to single users. A number of interesting work is going on similar issues (e.g., [3, 6]). In the rest of the paper we show an example of interface to a software tool that avoids users to learn a speci c language; has a supporting role with respect to the user; follows an agent-oriented implementation that allows for the speci cation of single parts of the whole system. 2 Active Management of E-mail Messages The project we are involved has the aim of incrementally developing a software Interface Agent connecting the user with an intelligent distributed architecture that supports and assists her in dealing with disparate tasks, tackling di erent network utilities and utilizing o ce services. We have built several specialized agents in order to increase the e ciency of the user's daily activities. Task of the Interface Agent is to select the agent(s) that can satisfy the user's needs, to coordinate the work of di erent agents, to make them cooperate in presence of a common goal, to communicate with the external world be it represented either by intelligent or by non-intelligent entities. Interface Agents acting on the behalf of di erent users can interact with each other in order to get help and can cooperate among them to accomplish particular goals. At the present, the architecture consists of: a rst implementation of the Interface Agent, a Mail Agent, a Meeting Agent and an Info Agent. Each agent consists in turn of sub-agents specialized to solve a particular problem. In this paper we illustrate the architecture and the behavior of the Mail Agent in order to show how the concepts above illustrated are translated into an operative framework. The Mail Agent is in charge of handling the incoming e-mail messages according to a user's pro le. With e-mail messages we mean every types of electronic mail such as bulletin-boards, mailing-lists, interest groups, etc. To cope with them speci c tools exist, e.g., packages like Procmail or MailAgent, that allow a user to lter the incoming messages: such tools are powerful but of di cult usage for non-expert programmers. Our Mail Agent hides to the users the use of the Procmail software and adds an active task by automatically suggesting modi cations in the pro le on the basis of a comparison between accepted and rejected messages. Procmail is a software that checks each incoming e-mail message according to a number of user-de ned lters. Filters have to be speci ed using the Procmail command language consisting of a set of operators, as in the following example: MAILDIR=/usr/users/amedeo/mail