® InfoJur.ccj.ufsc.br

A Discourse on Law and Artificial Intelligence

Michael Aikenhead, Lord Lloyd Prize winner, 1994-1995, Collingwood College, Durham

Biography:

Michael Aikenhead completed a Bsc (Computer Science) and an LLB (Honours) in 1994, both at the University of Melbourne, Australia. Michael is currently undertaking a Master of Jurisprudence, at the University of Durham, while participating as a member of the University's Centre for Law and Computing. Michael's thesis is an investigation of the use of Artificial Intelligence to implement models of legal analogical reasoning.

Introduction

Attempts to simulate the processes of legal reasoning using artificial intelligence (AI) have a long and stimulating history. To date, attempts have resulted in only very rough simulations; the processes of legal reasoning have only inaccurately been captured.

This inability to capture the richness and complexity inherent in law and legal reasoning has partly been the result of the adoption of outdated jurisprudential models on which to base AI legal reasoning systems.

The recent adoption by researchers in AI and law of more modern jurisprudential theory about the nature of law and legal reasoning, holds the promise of creating truly useful automated legal reasoning systems.

Jurisprudence, AI and Law

Every AI system in law necessarily embodies a theory of law and a theory of legal reasoning.[1] Most AI systems in law have applied one of two dominant legal theoretical paradigms.[2] A majority of these systems are rule based. Legal knowledge is represented in these systems using production rules.[3] Such a representation of legal knowledge implies a positivistic jurisprudential model, a model that views law as relatively certain, certain enough to be concretely expressed as a set of rules.[4] Such a representation of legal knowledge has many shortcomings however. The most discussed of these is the problem of open texture. To regurgitate an example from Hart,[5] in a rule stating;

How do we determine what amounts to a vehicle for the purpose of the rule? Are bicycles prohibited? What of an ambulance attending an emergency? Trying to provide more detailed rules will not solve the problem.[6] Such problems highlight the inability of extreme versions of positivism to fully capture all that is involved in law and legal reasoning.

In contrast to rule based systems, case based reasoning (CBR) systems in law embody a theory of law developed by the American legal realists. This theory of law regards law as composed primarily of experience, as embodied in case law, and not of rules. In the often quoted words of Oliver Wendel Holmes the life of the law has not been logic: it has been experience.[7] However, CBR systems also fail to capture all that is law and all that is involved in legal reasoning. Extreme versions of legal realism deny that the deductive application of rules ever occurs in law. This is jurisprudentially unsupportable.[8]

Thus neither the rule based approach nor the CBR approach to representing legal knowledge is sufficient. Although these approaches can be combined to create hybrid systems for reasons to be discussed below, such systems still fail to accurately reflect the nature of law and legal reasoning.

The Need for a New Jurisprudence - Discourse Theory of Law

As discussed above, the dichotomy between rule based systems and cased based reasoning systems in AI and law research reflects an underlying jurisprudential debate that has raged for the last century. However, this debate between rule positivism and legal realism has ossified and become stale, it is a sterile debate.

As Wasserstrom indicates,[9] of key importance in law is the justification of legal decisions. The debate between formalists and rule sceptics can be viewed as a debate as to what amounts to sufficient justification for legal decisions. The former regard rules as a sufficient basis for decisions while the later regard the multiplicity of human experience as requiring individual treatment of cases. Neither approach however, sufficiently addresses how rules justify decisions or conversely what individual treatment of cases involves. Neither rule positivism nor legal realism is able to fully explain how legal decisions are justified.

Recent jurisprudential theory side-steps the positivist/realist dichotomy and instead focuses on law as a discourse, law as a process of argument that occurs between parties.

The view of law as a discourse is paradoxically both very recent and very old. The framework is old in that it is one discussed by Aristotle.[10] It is recent in that the application of dialectics to law was only revived in the middle of this century by Perelman and Olbrechts-Tyteca,[11] Toulmin[12] and more recently formalised by Alexy.[13]

The theory of law as a discourse is fundamentally different to the positivist/realist dichotomy presented above. Instead of focusing on whether law is based on either rules or cases, the central tenet of discourse theory is that law is a process of argument. Accordingly, the justification necessary for legal decisions arises not from the mere way that law is represented, but through the community's acceptance of conclusions reached through a process of argumentation.

Within discourse theory, neither rules nor cases are in and of themselves adequate to form self-sufficient arguments. This is because rules and cases can themselves be the subject of argument.

That rules can be the subject of argument can be seen in various ways. The much discussed problem of open texture is but one example, what exactly is the scope of a particular rule? At another level, the validity of a rule can be questioned. Further, the very acceptability of rules can be questioned. That cases can be the subject of argument can be illustrated in several ways. The fundamental way in which cases are used in law is through analogical reasoning.[14] However, cases can only be applied analogically when they are similar, but when are two cases similar? Further, only the ratio decidendi of a case is binding, but what is the ratio? Even when two cases are similar, are the reasons supporting the decision in the original case still valid reasons?[15] According to discourse theory, questions such as these are resolved through a process of argument.

However, while rules and cases are not themselves sufficient to form standalone arguments, both rules and cases are elements used in the construction of arguments. According to Alexy, at least one universal norm must be adduced in the justification of a legal argument.[16] Similarly, if a precedent can be cited in favour of or against a decision it should be so cited.[17] Thus while rules and cases can be the subject of arguments, they can also be used in the construction of arguments. Such argument can itself refer to further rules and cases![18]

Unsurprisingly, this theory of law has interesting implications for theories of legal reasoning. As MacCormick states A theory of legal reasoning requires and is required by a theory of law.[19] Instead of implying that legal reasoning is primarily a process of deduction or a process of analogising the theory of law as discourse requires a richer view of the process of legal reasoning.[20]

The task for researchers in AI and law now becomes to model this process of discursive argument which is at the centre of law and legal reasoning.

Artificial Intelligence Models of Law as Discourse

Various attempts have been made by researchers in AI and law to construct computer programmes that emulate the discursive nature of legal argument. There are two aspects of discourse theory that are investigated

While both can be regarded as investigating the nature of argument, the latter is primarily concerned with the procedural flow of argument.

The most obvious attempt to computerise the discursive nature of legal reasoning is through the use of non-monotonic logic. The use of non-monotonic logic in AI and law is a recent occurrence.[21] It is an attempt to provide a formal framework with which to simulate the discursive character of legal argument.[22] Early rule based systems in law were largely constructed using monotonic logics. In monotonic logic, when the antecedents of a rule exist, the consequences of the rule necessarily result. For example, a rule states:

Whenever A, B and C exist the conclusion X automatically follows. The problem with such monotonic logic is that it cannot deal with inconsistency amongst rules and arguments. If two rules leading to opposite conclusions exist, the system concludes that both outcomes are equally likely. For example, if in addition to the above rule, the following rule also existed:

The system would conclude that both X and not(X) will occur!

Discourse theory indicates that law is a system based around opposing arguments. Thus a system that seeks to model law has to be able to deal with opposing arguments. Non-monotonic reasoning systems, which utilise non-monotonic logic, achieve this to a limited extent. In non-monotonic reasoning systems the existence of the antecedents of a rule does not automatically trigger the consequences of the rule; the consequences specified in a rule are only implied by that rule. A rule's conclusion will not occur if the conclusion of another rule is stronger than the conclusion from the first rule. This avoids the problem with monotonic logic which arises when the existence of the antecedents of a rule strictly imply the conclusion of the rule. Researchers in AI and law applying non-monotonic logic claim that non-monotonic legal reasoning systems can emulate the discursive character of law. It is claimed that such systems can reason with inconsistent arguments and can weigh arguments for incompatible conclusions.[23]

Non-monotonic legal reasoning systems such as those by Prakken,[24] Sartor[25] and Loui,[26 ] seek to provide formal specifications for the nature of legal argument. Though differing in their details and definitions, at the heart of such systems are attempts to formalise what amounts to an argument for a conclusion, what amounts to attack on an argument, what amounts to support of an argument and what amounts to defeat of an argument. In the provision of such definitions and in focusing on the character of legal argument prima facie such systems appear to implement jurisprudential theories of legal discourse. They appear able to provide a means by which opposing arguments can be resolved. This claim needs critical assessment however, and will be subject to more detailed investigation shortly.

In addition to modelling the nature of legal argument, non-monotonic reasoning systems also model the flow of legal arguments, the procedural processes by which legal arguments are made and countered. Such models have been developed by Gordon,[ 27] and Rissland and Skalak.[28] This work formalises when arguments can be asserted, conceded and denied, and also specifies the various norms which govern the overall argumentation process.[29] For example, Gordon's pleadings game is a formalisation of procedural justice and is used to identify substantial arguments and clarify the issues under dispute.[30] In contrast to those systems that seek to model the nature of argument, the goal here is to identify issues, rather than to actually decide the main claim.[31]

The discursive nature of law has also been noted be researchers in CBR.[32] CBR systems such as HYPO[33 ] do not merely return cases relevant to the current set of facts. HYPO uses facts to construct competing arguments for and against a position. The system formulates arguments for a position, identifies likely weaknesses in these arguments, formulates arguments based on these weaknesses for the opposition and finally, generates modified and strengthened arguments for the original position through the use of hypotheticals. In this way the system generates chains of opposing argument. Using these arguments it is concluded which side will win. Other similar systems such as GREBE[34] have been created.

However exciting, and though a great advance on earlier approaches in AI and law, whilst these systems appear to capture the character of law as a discourse they have several shortcomings.

Problems with Computer Models of Law as a Discourse

Computer legal reasoning systems based on discourse theories of law are promising. However, there are problems with such systems. The primary problem, and the one discussed here, arises from the claim that such systems can weigh competing arguments and thus choose between them.[35]

In a system of discourse where conflict between arguments is at the very heart of the system, some method to resolve conflicts between arguments is necessary. If two arguments are in conflict it is necessary to choose between them.[36]

According to legal theory, when arguments conflict they are compared on the basis of the policies, principles and values that underlie them.[37] There is reference to meta-standards. A computer system that seeks to reason with law will have to take account of this.

In current computerised discourse legal reasoning systems, conflicts between arguments are resolved by reference to meta-norms. Examples of standards for comparing arguments given by Prakken are Lex Superior, Lex Specialis and Lex Posterior. The most authoritative, the most specific or the oldest norm will prevail. Sometimes legislation provides specific rules.[38 ] In certain circumstances, the reference to meta-norms has a sound jurisprudential basis.[39] Thus this approach seemingly provides a sound way to resolve conflicts between arguments.

However, while the reference to meta-norms is superficially attractive it has numerous jurisprudential problems. For example:

As Perelman and Olbrechts-Tyteca, and Alexy make clear, such problems are themselves resolved through a process of argument.[40] What amounts to acceptable meta-level argument is determined by discourse theory, just as is what amounts to acceptable argument.[41] Determining which meta-arguments support an argument is also determined by discourse theory.[42] When meta-arguments conflict, this must be resolved through further argument. Present computerised legal discourse reasoning systems only argue with a very limited set of meta-norms. These meta-norms are a priori determined to apply or not to apply to any given rule. Conflicts between meta-arguments are resolved simply by ordering them into a hierarchy of precedence.[43] It is jurisprudentially suspect to assume an a priori hierarchy exists amongst meta-norms.

It may be argued that an a priori hierarchy of meta-norms[44] does exist. This hierarchy however deeply buried, may be discovered, so it is argued, from an examinationof legal texts, particularly statutes and cases. As Hage says, both cases and rules represent different techniques for establishing and weighing reasons.[45] Cases are collections of reasons weighing for and against a particular solution.[46] The problem with this view, apart from it being unclear what happens when meta-norms have not previously been considered in cases, and even assuming that meta-norms could be clearly identified in cases, is that it ignores the problem that the legal system is not a consistent whole. In his theory of legal rights, Ronald Dworkin, while insisting that there is a single right answer to all legal problems, does not go so far as to say that the law is a completely consistent whole. Dworkin argues that since the law has been created by numerous different people in different social contexts, the values that make up the law and that are expressed through cases and statutes, are not consistent.[47] Thus it is impossible, even for Hercules, to use legal texts to construct a consistent hierarchy of values. While Hercules may still be able to reach a right answer, this will not result solely from an examination of legal sources.

A final problem with existing computer implementations of legal discourse theory is that although reference is made to meta-norms, no reference is made to standards of principle, policy or value. It is clear that such standards are applied in law and legal reasoning.[48] Meta-reasoning using principles, policies and values is not implemented in existing frameworks. Nor is it clear how such arguments would be incorporated into existing frameworks. At the very least, the problems relating to the application of meta-norms apply with equal force to reasoning using these standards.

The Next Step in the Computerisation of Legal Discourse

With these problems, one might conclude that current systems are of no value, or worse, that discourse theory should be abandoned as too complex and unhelpful. These would be hasty and unfortunate conclusions.

For computerised discourse reasoning systems to accurately reflect the processes of legal reasoning they must not only justify the application of arguments through the use of meta-standards, but also reason about the application of those meta-standards through a process of argument. This is acknowledged by researchers themselves.[49] Systems which predetermine how a meta-norm supports a norm, and how meta-norms interact, and which ignore standards of principle, policy and value, miss the qualitative value of discourse theory. What is required is a system that fully uses discourse theory to resolve arguments.

This requires a system that can truly weigh arguments. In such a system arguments about arguments would occur just as for arguments about other issues. Arguments, conflicts between arguments, arguments about conflicts between arguments etc would all be resolved through the process of argument. Arguments are resolved by further argument and discourse theory applies at every stage and at every level of argument. When Prakken asks What is the best way of reasoning about priorities? he merely concludes that this is an interesting topic for future research.[50] Gordon is more consistent in his conclusions. He states that a discussion about backing can be held just as for any other claim.[51] Any accurate implementation of discourse theory will thus need to allow for meta-level arguments on the same basis as for arguments themselves.[52]

The self-referential nature of this aspect of discourse theory may seem worrying. Although beyond the scope of this paper it is interesting to consider how arguments would ever terminate. Restricting the set of specific moves by which players in a discourse are allowed to make assertions[53] may itself provide a procedure leading to termination of arguments. While this will prevent certain arguments from being made it is uncertain if it will prevent argument from continuing indefinitely. Interestingly, Alexy states that some rules of discourse simply have to be accepted for the process of justification to be possible at all.[54] Discourse theory then cannot explain everything. Perhaps argument terminates when it reaches a level where things simply have to be accepted. A difficulty with this approach is of course that, to be consistent, it must be conceded that exactly what it is necessary to simply accept as a prerequisite to a discourse should itself be open to argument.

In this context the CBR system, GREBE[55] is particularly interesting. Branting's system addresses a major problem that arises in case based reasoning; of determining when two cases are sufficiently similar to allow analogical reasoning to occur. For two cases to be similar, their constituent elements must be sufficiently similar. How is the existence of this similarity determined? Branting's innovation is to allow GREBE to recursively use precedents to generate arguments as to why elements of cases are similar. Once the elements of cases can be said to be similar, the cases themselves can be regarded as similar. The system thus generates an argument as to why cases are similar by recursively generating an argument as to why constituent elements of cases are similar.
While GREBE only operates in the domain of CBR and is not a complete argumentation system it does illustrate how useful argumentation processes can be self-referential. What is needed is an extension of this approach; applying the full theory of legal discourse. In such an approach, arguments would recursively be generated as to how and why rules or cases should apply. A rule or case may be argued to support a certain conclusion, if it is questioned why this rule or case is relevant, argument about relevancy can occur. Such argument may proceed through reference to further rules and cases. Such a system would recursively resolve arguments and conflicts between arguments. All the basic elements of such a system exist in, for example, the systems of Prakken, Sartor and Loui. What is needed is an extension of these systems to allow arguments about meta-arguments.

Conclusion

The adoption of discourse theory by researchers in law and AI is a welcome step and promises release from the stale confines characteristic of the positivist/realist jurisprudential debate. The creation of automated legal reasoning systems based on discourse theory is an exciting development which promises much.

However, before these promises can be delivered, before computer systems that reason like lawyers can be constructed, the full implications of discourse theory must be faced. Only when research moves beyond mere partial implementation of discourse theory will we possibly reach a state of affairs where we have computer implementations of law and legal reasoning rich enough that jurisprudence itself can learn something from the application of artificial intelligence to law.

Bibliography

Aikenhead M, The Use and Abuse of Neural Networks in Law, 12 Santa Clara Computer and High Technology Law Journal 31.

Alexy Robert, (1989) A Theory of Legal Argumentation Clarendon Press.

Ashley Kevin D, (1990) Modelling Legal Argument: reasoning with cases and hypotheticals MIT Press.

Branting L. Karl, (1991) Reasoning with Portions of Precedents p.145, in The Third International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

Dworkin Richard, (1977) Taking Rights Seriously, Gerald Duckworth & Co.

Gordon Thomas F, (1993) The Pleadings Game - Formalizing Procedural Justice p.10, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

Hage Jaap, (1993) Monological Reason-Based Logic: A Low Level Integration of Rule-based Reasoning And Case-Based Reasoning p.30, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

Hart H.L.A., (1961) The Concept of Law, Oxford University Press.

Levi E.H., (1949) An Introduction to Legal Reasoning, University of Chicago Press.

Loui Ronald P., Norma Jeff, Olson Jon and Merrill Andrew, (1993) A Design for Reasoning with Policies, Precedents, and Rationales p.202, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

MacCormick N., (1978) Legal Reasoning and Legal Theory.

Perelman Chaim and Olbrechts-Tyteca L., (1969) The New Rhetoric: A Treatise on Argumentation, University of Notre Dame Press.

Prakken Henry, (1993) A Logical Framework for Meodelling Legal Argument p.1, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

Routen Tom, (1989) Hierachically Organised Formalisms p.242, in The Second International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

Sartor Giovanni, (1993) A Simple Computational Model for Nonmonotonic and Adversarial Legal Reasoning p.192, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

Skalak David B. and Rissland Edwina L., (1991) Argument Moves in a Rule-Guided Domain p.1, in The Third International Conference on Artificial Intelligence and Law: Proceedings of the Conference, ACM Press.

Stone Julius, (1964) Legal System and Lawyer's Reasoning, Stevens & Sons.

Susskind Richard E., (1986) Expert systems in law: A jurisprudential approach to artificial intelligence and legal reasoning, 49 Modern Law Review 168.

Toulmin S., (1958), The Uses of Argument.

Wasserstrom R.A., (1961) The Judicial Decision, Stanford University

Zeleznikow John and Hunter Dan, (1994), Building intelligent legal information systems: representation and reasoning in law, Kluwer.

Footnotes

1. Susskind Richard E., Expert systems in law: A jurisprudential approach to artificial intelligence and legal reasoning (1986) 49 Modern Law Review 168.

2. Approaches in AI and law such as the use of neural networks will not be considered here. For an overview of the use of neural networks in law see: Aikenhead M; The Use and Abuse of Neural Networks in Law, 12 Santa Clara Computer and High Technology Law Journal 31.

3. See: Zeleznikow John and Hunter Dan, Building intelligent legal information systems: representation and reasoning in law (1994) Kluwer for an overview of rule based systems.

4. This is an extreme version of rule positivism though.

5 . Hart H.L.A., The Concept of Law (1961) Oxford University Press.

6 . Zeleznikow, above n 3.

7. Cited in Zeleznikow, above n 3, 62.

8. MacCormick N., Legal Reasoning and Legal Theory (1978), ch 2; Stone Julius, Legal System and Lawyer's Reasoning (1964) Stevens & Sons, chs 6, 7.

9. Wasserstrom R.A., The Judicial Decision (1961) Stanford University Press, 27.

10. See Perelman Chaim and Olbrechts-Tyteca L., The New Rhetoric: A Treatise on Argumentation (1969) University of Notre Dame Press.

11. Ibid.

12. Toulmin S., The Uses of Argument (1958).

13. Alexy Robert, A Theory of Legal Argumentation (1989) Clarendon Press.

14. Levi E.H., An Introduction to Legal Reasoning (1949) University of Chicago Press.

15. Within Alexy's theory, the ability to question all these results from R.2.2(a), see Alexy, above n 13.

16. Alexy, above n 13 Rule J.2.1. Refer to Alexy as to why this is necessary and what amounts to a universal norm.

17. Alexy, above n 13, Rule J.13.

18. In this way the process of discourse can be seen as a self-referential one. But this does not mean that it is circular. A circular argument is one that repeatedly covers the same ground. As will be familiar to computer programmers, a self referential process can contain its own termination conditions.

19 . MacCormick, above n 8, 229.

20 . It can now be appreciated why hybrid systems are not accurate representations of law. They misrepresent the process of legal reasoning. While seemingly reaching a compromise between positivist and realist models of legal reasoning, such systems adopt a two stage process model of legal reasoning. First, if the question is clear then rules are applied. However, if there is uncertainty over the application of a rule then the second stage is invoked and cases are used in an attempt to resolve this uncertainty. However, discourse theory indicates that argument does not occur in such neat two tiered stages. Both rules and cases can be used at any stage of an argument. It is not valid therefore to a priori apply a two stage reasoning process under which rules are privileged.

21. Prakken Henry, A Logical Framework for Meodelling Legal Argument p.1, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1993) ACM Press, 1.

22 . Ibid.

23. Ibid.

24. Ibid.

25. Sartor Giovanni, A Simple Computational Model for Nonmonotonic and Adversarial Legal Reasoning p.192, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1993) ACM Press.

26. Loui Ronald P., Norma Jeff, Olson Jon and Merrill Andrew, A Design for Reasoning with Policies, Precedents, and Rationales p.202, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1993) ACM Press.

27. Gordon Thomas F., The Pleadings Game - Formalizing Procedural Justice p.10, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1993) ACM Press.

28 . Skalak David B. and Rissland Edwina L., Argument Moves in a Rule-Guided Domain p.1, in The Third International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1991) ACM Press.

29. Such as the prevention of self contradiction.

30. Gordon, above n 27. This is based upon the work performed by Alexy towards formulating dialectics, ibid 10.

31. Ibid 18. In this sense, these two formulations of the different aspects of dialectic are complementary. The latter provides a formalisation of the processes by which arguments are made, while the former provides means for evaluating and choosing between those arguments made during the course of the game. In fact the separation is not one that bears close scrutiny. The 'moves' that are allowed during argument inevitable influence the character of arguments themselves.

32. The current split in investigations on legal discourse, between investigations using non-monotonic logic and those based on CBR is a hangover from the positivist/realist debate. That both approaches see the need to model law as a discourse is acknowledgement of its importance.

33. Ashley Kevin D., Modelling Legal Argument: reasoning with cases and hypotheticals (1990) MIT Press.

34. Branting L. Karl, Reasoning with Portions of Precedents p.145, in The Third International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1991) ACM Press.

35. Prakken, above n 21, 1.

36. Other than merely concluding that the situation is uncertain.

37. That such argument occurs is not controversial; how it occurs and its validity is more so. See: Levi, above n 14, MacCormick, above n 8 and Dworkin Richard, Taking Rights Seriously (1977) Gerald Duckworth & Co. Although there are big differences in the approaches taken by these theorists, they are all in agreement that values, principles and policies play a part legal reasoning.

38. Prakken, above n 21, 4; Routen Tom, Hierachically Organised Formalisms p.242, in The Second International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1989) ACM Press, adopts the idea of hierarchies of rules in legislation in order to better implement AI models of legislation.

39. All law students will be familiar with the canons of statutory interpretation.

40. Alexy, above n 13.

41. Ibid.

42. Ibid.

43 . Prakken, above n 21, 5-6; Sartor, above n 25, 193. While Prakken's system provides a general reasoning framework in that it is not tied to any particular standard of argument defeat, in a concrete situation involving two competing arguments the system does require specific standards to be specified; Prakken, above n 21, 3.

44. An indeed principles, policies and values.

45. Hage Jaap, Monological Reason-Based Logic: A Low Level Integration of Rule-based Reasoning And Case-Based Reasoning p.30, in The Fourth International Conference on Artificial Intelligence and Law: Proceedings of the Conference (1993) ACM Press, 31.

46. Ibid 35.

47 . Dworkin, above n 37, 119.

48. Refer to the sources cited in n 37 above.

49. Hage, above n 45. 31.

50 . Prakken, above n 21. 7.

51. Gordon, above n 27, 15.

52. The imposition of a predetermined hierarchy of values, and the inflexibility associated with it is perhaps the reason for Hage's claim that weighing reasons does not always lead to a conclusion and that it leads to a new equivalent of hard cases; Hage, above n 45, 36.

53. Gordon, above n 27, 13.

54. Alexy above n 13, 202.

55 . Branting, above n 34.
Published in the Law Technology Journal: Vol 5, No 1

Original publication date: June 1996
Web publication date: February 1997