Jeannette Quast

Computers en vage normen; Een computermodel voor de behandeling van juridische casus

(Computers and vague norms; A computer model for the treatment of legal cases)

Leiden University, 29 February 1996

Summary

Computers and vague norms
A computer model for the treatment of legal cases

The research reported on in this book is based on two premises:

• the implementation of statutory regulations leaves much to be desired;
• the implementation of statutory regulations can be improved with the use of computer systems.

The research deals with a computer model constructed on the basis of the ÔleidraadÕ knowledge-based system. A knowledge-based system is a computer program that performs a specific cognitive task and differs functionally and structurally from an information system (a conventional computer program containing structured information). leidraad is capable of handling legal cases concerning the concept of commensurate employment in the Dutch Unemployment Act. The model constructed using leidraad is referred to as the qualification model and is intended for multiple re-use. In other words: the qualification model is constructed with the objective of developing knowledge-based systems capable of handling cases concerning various domains of legal expertise.

Chapter 1: Introduction

The research was carried out in the field of artificial intelligence and law and in particular in the area of coincidence between three scientific disciplines: information technology, psychology and jurisprudence. In the Anglo-Saxon literature this multidisciplinary field is referred to as Artificial Intelligence and Legal Reasoning or AI and Law for short. Research into the development of knowledge-based systems in the field of law (legal knowledge-based systems) forms a separate category within this field. We focus on knowledge-based systems that assist with the interpretation of statutory regulations in specific cases. Examples are five systems developed in the Netherlands: juricas, pem, tessec, prolexs AND leidraad.

When developing legal-knowledge-based systems we encounter two significant problems: the knowledge-acquisition problem and the problem of vague norms. The knowledge-acquisition problem is associated with the fact that it is found to be difficult to incorporate into a computer program the knowledge employed by a human expert in performing a cognitive task. Vague norms are legal rules in which reference is made to indeterminate concepts.

In the literature the model-driven approach has been proposed as a solution to the problem of knowledge acquisition. In this approach the knowledge to be incorporated is first analysed as extensively as possible using a generic model, before proceeding with its incorporation into a computer program. A generic model is an abstract representation of collective knowledge concerning the performance of a class of cognitive tasks. The courts are responsible for the specific interpretation of vague concepts. Case law thus results in the development of guidelines for the interpretation of vague concepts and vague norms. The treatment of legal cases that require the interpretation of vague norms is referred to as qualification.

To develop knowledge-based systems for various legal domains, it is desirable to have a generic model of the qualification task with which it is possible to perform model-driven knowledge acquisition. The aim of our research is to contribute to this: we set out to construct a generic model that can be used to develop legal-knowledge-based systems along model-driven lines. In this context we ask:

1. Is it possible to describe knowledge about the application of a specific vague norm in such a way that the description can also be used for the application of other vague norms? We interpret a legal-knowledge-based system as a description of knowledge concerning the application of a specific vague norm and in this context we use an existing system for the development of a generic model of the qualification task. Since much Dutch case law has been implemented in leidraad, we decide to use this system as the basis. As a result, a key focus of our research is represented by the following question (2):
2. Does the way in which leidraad performs the qualification task (with respect to 'commensurate employment') also apply to other legal domains?

We deal with question 1 at the end of this thesis by means of generalisation. From leidraad we use declaration schema's (general frameworks) to generalise to legal knowledge in general.

Chapter 2: Premises

Our formulation of the problem is based on two premises. Firstly that the implementation of statutory regulations leaves much to be desired. Secondly that computerised support for practical implementation can contribute to improvement. We support both premises with information derived from the practical implementation of social security regulations. We give an impression of the enormous scope of the field of practical implementation of social security regulations and discuss the resulting costs. We then demonstrate that the quality of practical implementation, particularly in terms of the accuracy, promptness and comprehensibility of decisions, leaves much to be desired. We show that computerisation of (part of) the practical implementation can contribute to improvement of the quality.

Chapter 3: Methodological framework

We interpret a method for the development of a knowledge-based system as a tool, the form of which must match the way in which people perform this cognitive task. This interpretation dictates the choice of methodological framework: we use kads as method and we ÔplaceÕ the results of our research in the Commonkads library.

kads is a method that focuses on model-driven knowledge-acquisition. When using kads seven models are produced. The models reflect just one or a few properties of the empirical system being modelled. The expertise model in particular is important to our research; this model specifies the problem solving knowledge. In kads the principle that the results of earlier analysis (in the form of models) are used to produce expertise models has been embodied among other things in the form of a library of models. The Commonkads library can accommodate expertise models and model elements that have been created by means of kads and other model-driven methods.

Chapter 4: leidraad: commensurate employment

We start with an overview of leidraad and of the implementation of knowledge concerning the concept of commensurate employment as used in the Dutch Unemployment Act. We describe the system on the basis of existing rationalisation of legal decisions. From this we derive the functional specifications for leidraad. We explain the way in which knowledge is presented in the system with leitaal and we illustrate the operation of the system with examples.

leidraad is then defined in terms of kads. As part of this definition, in addition to the expertise model, we also construct parts of the task model and the cooperation model. Particularly at the inference level of the expertise model, the definition in kads is not entirely successful. This can be explained as follows. Firstly, we are defining an implemented system in terms of the products of a method for system development. This reverse approach means that we cannot expect complete one-to-one correspondence. Secondly we are defining a first-generation knowledge-based system as a second-generation knowledge-based system.

Chapter 5: leidraad: generic model

In this chapter we show the system from a different perspective: domain specific knowledge concerning commensurate employment is omitted and the emphasis is placed on generic knowledge in leidraad. From this definition we construct a generic model, referred to as the qualification model. The model is constructed by making a conceptual specification of the generic elements of the knowledge represented in leidraad. Since the model is intended for multiple re-use, we then `place' it in a library of generic knowledge models: the Commonkads library. On this basis we conclude that the qualification model can be conceived as a sub-type of the assessment model present in the library.

Chapter 6: Evaluation of the qualification model

In this chapter we investigate the extent of our success in constructing a generic model that can be used for developing legal-knowledge-based systems along model-driven lines. In particular we examine whether the qualification model is suitable for this purpose. The qualification model is examined from two perspectives: from the perspective of the knowledge engineer and from the perspective of the domain expert. If we evaluate the qualification model from the perspective of the knowledge engineer, we investigate whether the model is compatible with the method used by the knowledge engineer in developing a knowledge-based system. Since the model is compatible in two respects (global structure of kads models and inference structure of the assessment model) with an acknowledged method of knowledge-based-system development, we conclude that the qualification model is usable from the perspective of a knowledge engineer.

To answer the question whether the qualification model is also usable from the perspective of a legal-domain expert, we examine whether, in addition to modelling the domain of commensurate employment, it is also possible to model other legal domains with the qualification model. To this end we investigate whether the empty domain level of the qualification model can be filled with knowledge derived from empirical descriptions of the application of vague norms in five legal domains: the establishment of entitlement to unemployment benefit, unemployment-act sanction decisions, compensation on termination of contract of employment, extension of confinement to a state mental hospital by a judicial order, and damages (pretio doloris) in the case of an accident.

We observe that the domain scheme (the structure of the domain layer) contains primitives that are significant in combination. The scheme also contains mutually redundant primitives. These two characteristics of the domain scheme constitute reasons for translating the domain scheme into a declaration scheme that includes significant combinations of domain primitives are included and in which no mutually redundant primitives occur. The declaration scheme is intended as a means for ÔfittingÕ legal knowledge-based systems relatively quickly (i.e. without implementation). We evaluated the qualification model by searching in the five different domains for examples of the nine domain primitives in the declaration scheme. After entering each set of examples we asked two questions:

• are there unused primitives?
• are there missing primitives?

Since both questions could almost always be answered in the negative, we conclude that we have demonstrated that the qualification model has a generic legal character and that the model is usable for a domain expert, since it can be used to model several legal domains.

Chapter 7: Conclusions

With regard to the problem examined in our research, we conclude that the manner in which leidraad performs the qualification task (in the field of commensurate employment) can also be applied to other legal domains. This affirmative answer leads to a similarly affirmative answer to the question whether it is possible to define knowledge concerning the application of a particular vague norm in such a way that this definition can also be used in the application of other vague norms.

With regard to the objective of our research, we conclude that we have succeeded in constructing a generic model, that is demonstrably capable of use for developing legal-knowledge-based systems along model-driven lines.

In the concluding observations we argue that, provided that a number of conditions are met, legal practitioners can benefit from generic models such as the qualification model. In the longer term, society can also benefit from models such as the qualification model, since such models can have an effect on the judicial process. In addition, we advocate a criterion for determining the degree of vagueness of norms with which suitable domains can be selected. We contend that the qualification model can also be used for developing knowledge-based systems to perform tasks other than the qualification task, such as drafting legislation. The usefulness of the qualification model is not limited to the legal domain: knowledge concerning decisions in which the arguments for and against can be weighed against each other, can also be modelled with the qualification model. On the basis of our research we contend that the specific nature of Ôlegal reasoningÕ is associated with the constraints on the reasoning process. Finally we conclude that the qualification model removes one of the remaining obstacles to justice dispensing computers.