Geof Staniford

PLAID: proactive legal assistance

This paper describes PLAID, a system designed to support the preparation of a brief concerninga legal question.The s@em provides a multi agent framework in which ideas from argument basedexplanation,dialoguegames,documentmodellingandconcep tual retriewd are integrated into a tool capable of drawing on a variety of sourcesto produce a coherentargument for a particular point of view. A prototype, illustrating the functionality and componentsof the systemand implementedin PROLOG, is described.

A multi-agent legal argument generator

One of the most significant series of experiments in AI and law is the investigation of case based reasoning carried out in the HYPO, CABERET and CATO projects. It is important to understand what has been achieved in these experiments, and this requires that the techniques be applied to a variety of domains. The techniques are not, however, straightforward to apply. In order to provide an experimental environment for investigating the techniques further, we have re-implemented the central argument generation process as a set of agents written in Java using IBMs Aglet agent building framework. By using this environment it will be possible to explore the general applicability of the techniques. In this paper we give a summary of an algorithmic description of argument building in HYPO and CATO, and describe our implementation of the algorithm.

Simulating animal societies with adaptive communicating agents

Animal societies are an excellent source of ideas for studying distributed and emergent computation. These complex systems can be simulated in a number of ways and there is a range of models upon which such simulations can be based. These include both stochastic systems in which large numbers of very simple agents interact and social behaviours emerge and also logic-based systems in which the numbers of interacting units are reduced but the functionality of each individual is increased. This chapter presents an approach for investigating logic-based systems based on the formal specification of communicating agents in a distributed system. The foundations for this model specification methodology are presented and applied to the specific example of honeybee colony defence. The chapter concludes with a discussion of how and where the approach can be applied.

Linearising Hypertext through Target Graph Specifications

Linearisation is a special case of the navigation problem for hypertext whereby it is required that an orthodox linear text be produced from a hypertext network. Complications arise in solving this from the intricacy of the linkage structure that will be present in a typical hyperdocument. In addition, what constitutes an acceptable linearisation may depend heavily on the intended audience for the final linear text. Thus some readers may want only a precis of the hypertext content whereas others may wish to see an almost complete exposition of the textual material. For example in many cases it is normal practice to include a management summary and a full text. A variably linearised hypertext could in principle address both these and intermediate needs. In this paper we describe an approach to the development of hypertext linearisation algorithms that is capable of dealing with such problems. The approach insists that a specification of the target document structure be given as input together with the hypertext to be linearised. The intention of this specification is to prescribe the relationships between textual nodes in the hypertext to be included in the actual linear version.

Cooperative dialogues with the support of autonomous agents

An architectural specification of independent communicating agents is described. Such agents are designed for use in an environment which is most suitably represented as a graph or network and an example of one such environment is given based on the structured document graph paradigm. The specifications extend agent architectures in a natural way so that the complexities found in distributed systems may be addressed. Such an approach has a number of advantages. It provides a general representation paradigm that can be used as a basis in a wide range of intelligent distributed systems, and is easily customized to cope with problems requiring graphs with differing structures and with different node and edge properties and attributes. A suitable platform to build agents and environments and an example of cooperative communication using dialogue analysis are presented.<<ETX>>

Guardian agents to support student group projects

The paper describes the design of a set of agents intended to assist students undertaking group project work in a distributed environment. Increasing reliance on distance learning causes particular problems, as conventional monitoring techniques are often ineffective when group interactions may take a variety of different forms. Our objective is to monitor group activity and to provide assistance when the group appears to be at risk of failing to meet its objectives and then to offer advice on how the group can return to plan. If there is a serious risk of the group failing, appropriate members of the teaching staff are informed, together with the necessary logs to identify the problems. The initial design and construction of the agents is based on the ABW-Zeus methodology. Since all code generated is in Java, additional functionality can be added readily either by using class libraries or by adding Java code. This allows guardian agents to link effectively with existing Web based student learning environments.

Linearisation Schemata for Hypertext

Hypertext systems typically contain massive volumes of textual information. This information is structured as a labelled directed graph, so that progression from one item of text to another is via labelled edges in the hypertext network. A user, faced with this quantity of material, must cull the information required by carefully navigating through the nodes of the network. Even experienced hypertext users can become disorientated when navigating a complex network structure. A formal aid to hypertext navigation (with particular reference to the problem of linearisation) was presented in [1]. The mechanism described, however, has the drawback that the linkage structure on which it is built must be already present in the hypertext. In this paper we describe a formal extension to the earlier work which provides greater navigational flexibility through a higher level approach to user profile specifications.