LEGAL REASONING, LEGAL THEORY, AND ARTIFICIAL INTELLIGENCE

J.C. Smith and Daphne Gelbart

Introduction

During a lecture on the doctrine of Rylands v Fletcher, [(1868), L.R. 3H.L. 330], a law professor decides to test the student's understanding of this particular cause of action. She canvasses their opinions on whether or not an occupier of land who is carrying out genetic engineering research would be strictly liable for damages when, without any person being at fault, a genetically modified bacteria reaches or is carried on to an adjacent piece of land, with the result that all of the cattle belonging to the occupier are put to death by the government in order to prevent the further spread of disease. The first student has superficially listened to the lecture and remembers that the legal test is whether or not the use of the land was "non-natural", and concludes that nothing could be more non-natural than the modification of the genetic structure of bacteria. The second student has paid sufficient attention to remember the professor discussing a case where an occupier running a laboratory for testing cattle disease was held strictly liable when a foot and mouth virus escaped. The third student, has been day dreaming during the lecture, but thinks to herself that it would be unfair for the adjoining land owner to suffer a loss as the result of the actions of others who were engaged in making a profit for themselves.

All three students reach the same conclusion, but each does so by reasoning from a different kind of normative standard or premise. The first student knows the appropriate doctrinal test and applies it to the facts. We might call this "reasoning from rules". The second student remembers a similar case and cites it as the justification for her conclusion. "Precedential reasoning" or "reasoning from precedents" might be a convenient way of referring to this type of legal method. The third student analyzes the problem in terms of the purposes of the law. In the framework of legal arguments and judgments such premises are generally referred to as policies, so we might call this "purposive reasoning" or "reasoning from policy".

The legal profession uses all three kinds of premises or normative standards in conjunction with one another. Judges justify their decisions in particular cases by the use of legal doctrines. When a precedent is cited as an authority for a legal conclusion, the doctrinal justification of that particular precedent is implicit, since when citing the precedent one is presumed to be appealing to its expressed doctrinal foundations. In addition, one cannot say that a particular factual situation is relevantly more like one case than another without directly or indirectly invoking teleological or policy considerations, as relevancy is determined by teleology.

Legal doctrines, precedents, and policies are not unique norms; and doctrinal arguments, arguments from precedent, and policy rationals are not unique forms of reasoning. Doctrinal reasoning is a specialized form of rule-based reasoning. Precedential reasoning is a form of case-based reasoning, and policy arguments are a form of goal-based reasoning, or what is sometimes referred to as practical reasoning. What is unique within the legal context is the complexity of the arguments, and the interplay among the three kinds of norms. Case-based problem solving in non-legal areas generally does not have a doctrinal strata, nor the element of binding authority which is entailed in the application of precedents in legal reasoning. A comprehensive legal argument for the opening hypothetical problem would refer to discussions in some of the more important textbooks on the law of torts about the doctrine of strict liability for damage resulting from something escaping from land; would cite leading cases such as Rylands v. Fletcher itself, which authoritatively state the doctrinal rule about non-natural uses of land; would further cite cases where the doctrinal rule was applied to similar sets of facts; and would justify the application in terms of some underlying policy or teleological consideration, such as the common feature that in each of the cases where the defendant was held strictly liable for a loss caused to someone else, the defendant was making a profit from an action which created a non-reciprocal risk to others. In support of the teleological argument, academic literature in the area of tort theory would be cited. The argument itself would be an integrated and unified whole, even though three different kinds of standards or norms, legal doctrine, decided cases, and a principle of fairness, were invoked.

Legal Reasoning

Legal concepts, as we all know, are ambiguous, obscure, vague, imprecise, nebulous, indeterminate, and fuzzy [Berman & Hafner, 1988]. The concept of duty of care in the law of negligence, for example, can be used to refer to at least four very different kinds of issues; the question of whether or not the law will impose a standard of care; whether or not the defendants actions created a foreseeable risk of harm; whether the defendant's actions met a reasonable standard of care; and whether or not the damages claimed were too remote from the initial negligence [Smith, 1984 at 1-14]. Almost every teacher of the common law delights in pointing out the circularity of legal reasoning. The occupier has a right to exclusive possession because she has a leasehold estate and she has a leasehold estate because she has the right to exclusive possession. The defendant is liable because she owes a duty of care and she owes a duty of care because she is liable.

It has often been noted that legal concepts, doctrines, and cases are to be found in opposing pairs, one which favours the plaintiff's argument, and the other which favours that of the defendant. The plaintiff will argue remoteness issue in terms of proximate cause and the defendant will argue it in terms of a foreseeable duty [Smith, 1984 at 91-129]. Or the plaintiff will argue a remoteness issue in terms of the application of Wagon Mound II, [Overseas Tankship (U.K.) v. Miller S.S. Co. Pty., [1967] A.C. 617] while the defendant will argue the issue in terms of the application of Wagon Mound I, [Overseas Tankship (U.K.) v. Morts Dock & Enrg. Co. [1961] A.C. 388]. In a dispute which turns on the meaning of a statutory provision, one party will cite the "Golden Rule of Statutory Interpretation", which dictates that the clause should be interpreted in light of the teleology or purposes of the piece of legislation, while the other party will cite the "Clear Meaning Rule", which dictates that the section or sub-section should be given the meaning which the words themselves carry on their face. The bifurcated nature of legal arguments is a product of the adversarial system of litigation, and the duality of norms is possible because the relationship between legal doctrine and facts is contingent rather than necessary. The similarity between like cases is metaphorical rather than homomorphic.

Legal reasoning, justified in terms of legal doctrine, is indeterminate because the concepts are ambiguous. There are no necessary relationships nor clear rules to relate doctrine to fact with the result that the outcome of the issue depends on an arbitrary choice of doctrinal classification or conceptualization; and the doctrine is often bifurcated between apposing sets of concepts, rules or norms. An analysis of the policies which lie behind legal doctrine and cases, and a study of the economic impact of alternative ways of resolving cases or the practical implications of following one legal doctrine over another, reflect the same kinds of bifurcations and indeterminacy as do precedent and legal doctrine. Even if one looks behind the doctrines, precedents, and policies which constitute the law of torts, and grounds legal arguments in underlying philosophical moral or political principles one will still be faced with conflicting theories based upon alternative philosophical presuppositions.

There are no formal limits or internal barriers to the kinds of arguments which may be made within the trial process, nor are there rules which limit the kinds of norms which may be invoked in a legal argument before a court of law. Much of legal argument is at the doctrinal level. Lawyers will sometimes make policy arguments, and will even occasionally resort to the philosophical arguments based on moral theory. There are no rules which prevent them from doing so. Nevertheless, the tolerance of the court will diminish the farther one gets away from strictly doctrinal analysis.

Legal Reasoning and Information Costs

The extensive literature critical of the imprecision and ambiguity of legal reasoning, appears to rest on the supposition that it is flawed, second best, imperfect, and defective because it is not mathematical or deductive. We will argue in this paper that when one takes into account information costs, legal reasoning will be found to be efficient and effective. Almost every human action or task, whether parking a car, performing brain surgery or carrying out microscopic genetic engineering, entails a certain degree of imprecision. The greater the degree of inexactitude which can be tolerated, consistent with the successful achievement of the task or the goals of the action, the easier the action or task will be. The nature of the task will dictate the degree of inaccuracy which will be acceptable. Humans tend to reduce the information transaction costs of action by minimizing the costs of information gathering by maximizing the benefits of imprecision [Zadah, 1984; 1990]. Our language and discourse abounds with imprecise terms such as high, low, far, near, hot, warm, cold, expensive, cheap, young and old. All of these imprecise terms could be replaced with exact figures in terms of measures of distance, temperature readings on a thermometer, precise dollar and cents figures, or measures of time. When asked how old a person is, the answer is generally in terms of a number of years such as 32 years old. We seldom give the age precisely as 32 years, 8 months, two weeks, three days, and 18 hours.

We follow rules to reduce the information costs of action rather than paying the costs of finding the optimum conditions for the action. The optimum time for a child to go to bed will differ each day according to a number of factors such as the time the child awoke, the nature of the day's activities, or what the child has planned for her the next day. Rather than going through a lengthy process of calculating the optimum time for the child to retire each night, the parent will set a rule that the child's bed time is nine p.m., which, when measured against the optimal time, will be a little too early on some evenings and a little too late on others. An arbitrary age is set for people to have the franchise even though some individuals are mature enough to vote long before they reach the age of 18 while others may never achieve the mental qualities which justify participation in the political process. The age is set because the costs of establishing mental capacity on an individual bases would be much greater than the costs of allowing a limited number to exercise the franchise a little later or earlier than the exact time when they achieve voting maturity.

The inverse relationship between the degree of toleration of imprecision and the costs entailed in resolving disputes through litigation reveals a great deal about the nature and structure of legal proof and argument. The higher the exactitude required in proof of facts, the longer the process of proof will be. We are willing to tolerate a higher degree of imprecision in proof in a civil action than we are in a criminal case because, generally, only money is involved in a civil case, whereas one's life and liberty are jeopardized in a criminal action. This difference in toleration of indeterminacy in proof is reflected in the difference in the degree of proof required---beyond the balance of probabilities in a civil action, and beyond a reasonable doubt for a criminal charge. The rules of evidence do not require every factual assumption entailed in a legal argument to be proven. The acceptance of factual propositions without proof entails a certain tolerance for inexact information. Certain factual assumptions are taken to be sufficiently well established that the judge can take "judicial notice", without them having to be proven. Thus the length and costs of a trial are substantially reduced by placing certain kinds of information outside of the requirements of proof.

There are certain matters which are often quite difficult to prove, such as whether a child has the mental capacity to form the requisite intent in order to be held responsible for a criminal offence. In order to reduce the costs of proof, the law selects a particular age and arbitrarily asserts that below that age a child is presumed not to have the necessary mental capacity. Such an arbitrary age entails a certain degree of imprecision since it treats children with very different mental capacities all the same. One need only prove that the child is below the particular age (easily established by a birth certificate), and the child is presumed not to have the mental capacity to commit a crime. This assumption is irrebuttable. Other presumptions are rebuttable, such as the presumption of death if a person has been missing and unheard of for at least seven years.

The same inverse relationship between levels of toleration of imprecision in proof and the costs of litigation apply equally to the nature and structure of the legal arguments entailed in litigation. Legal reasoning presents us with an intriguing dilemma. The imposition of liability or denial of liability for damages caused by some one's actions must be justified. All reasons for acting or failing to act must ultimately rest on cause-effect relationships between human action and desired ends or goals. The discourse of action, reason, and justification is in the final analysis teleological. Yet the reasons for judgement in case-based reasoning contains very little discussion of the goals and purposes of the law. Rather, the discourse of legal case-based reasoning tends to be highly doctrinal. An action is justified either in terms of its cause effect relationship with a desired event or state of affairs, or because it falls under a normative rule or practice. This holds equally true for legal reasoning. The doctrinal justification in terms of rights and duties is justification in terms of a normative practice, while a policy justification is in terms of a desired outcome. The justification of an outcome in a case decided in terms of the application of a precedent lies within the doctrines and policies upon which that precedent rests.

The function of legal reasoning is to justify a decision which resolves a dispute. Justification can take two forms. Acts, decisions or states of affairs can be justified by showing that they fall within the rules of a practice for which the justification is taken for granted, and they can be justified by showing that they bear, or will have, a cause-effect relationship to a desired state of affairs. The division within utilitarianism between act utilitarianism and rule utilitarianism centers on a controversy about the preferred method of justification using the utilitarian measure of the greatest good for the greatest number. If preference is measured in terms of a cost-benefit analysis then rule utilitarianism is in general the preferred form of utilitarian justification because the information costs involved in decision are much less. If information costs were not taken into account, the preferred form of justification would be act utilitarianism, because the utilitarian measure would be made precisely and individually for each act. It is for this same reason that doctrinal justification is preferred over policy justification in legal argument, while policy analysis is preferred over doctrinal analysis, where one need not be overly concerned about information costs.

Deciding cases by the application of doctrinal rules avoids the necessity of arguing the nature and appropriateness of the foundations of the justificatory practice. By accepting imprecision at the level of the justification of justificatory practices themselves, we are able to greatly reduce the costs of litigation. Equally, It is not difficult to recognize why the discourse of legal reasoning is doctrinal rather than teleological. If legal arguments, rather than being doctrinally structured, were formulated in terms of practical reasoning, they would go on interminably. There is no authoritative text for what are the goals of a particular law or laws in general. There is little consensus concerning what they ought to be. The teleology of the law changes over time. A legal rule can serve one purpose at one period of time, and a different purpose at another period of time. The actual effects or impact of laws are either unknown or uncertain, and are inevitably controversial.

What are the values of the community? The community has no single set of shared values, since it is made up of groups and individuals with disparate interests that manifest themselves in the community by way of conflict and competition. The question then becomes which interests and related values ought to prevail. This becomes a moral issue which is beyond the capacity of the decision procedure of the court to resolve. The more that doctrine is separated from policy and justificatory principle the more circular the reasoning becomes. The more circular the reasoning the greater the degree of indeterminacy in legal argument. The greater the degree of indeterminacy in legal argument the less expensive is the trial process.

The very best justification for an action, or a decision related to an action, such as on whom should any resulting loss lie, is that which furnishes the justification in terms of the cause-effect relations of the action or alternatives of choice on desired outcomes, goals or values. Since it is extremely difficult to discover what outcomes maximize our individual and shared values, and what the cause-effect relationships are, the information cost of this kind of justification is extremely high. Justification in terms of rules and practices is second best, but desirable, because the justification of the rule or practice can generally be taken as a given. One needs only show that the action or issue falls under the rule or within the practice. The information costs of this kind of reasoning are very low compared to the costs of justifying an act or outcome of an issue in terms of direct cause and effect. In the former, all of the teleological issues have to be explored and proven. In the latter, most of it can be taken for granted.

Practical reasoning is essential for the justification of the law. Issues of practical reasoning, however, require a great deal of complex factual information about the cause and effect relationships between alternative actions or solutions to legal disputes. The gathering, presentation, and evaluation of factual information requires processes such as public hearings, royal commissions, investigative committees, media discussion, ministerial studies, and parliamentary debate, all or any of which are usually beyond the resources of the individual litigant and the decision making ability of a single judge, or a small group of judges. Consequently, practical reasoning is generally inappropriate for the courtroom decision procedure. There would be no way to limit legal argument if its discourse was practical rather than doctrinal. Imagine, for example, the added costs of trials in each action to resolve a conflict of interest arising from incompatible use of neighboring land, if the parties had to prove their case in terms of the optimal use as defined by the best interests of the community, rather than meet the necessary doctrinal requirements for establishing a cause of action or a defence thereto in nuisance. The relationship between cost and precision underlies pronouncements by judges such as that of Mr. Baron Parke in Egerton v. Brownlow, [10 E.R. 359 at 408] where he states, "This (public policy) is a vague and unsatisfactory term, and calculated to lead to uncertainty and error...To allow this to be a ground of judicial decision, would lead to the greatest confusion. It is the province of the statesman, and not the lawyer, to discuss, and the legislature to determine, what is the best for the public good and to provide for it by proper enactments."

Not only is there a cost to the search and verification of factual information within the confines of a given time frame dictated by the nature of the dispute between the two parties and the capacity of the judicial system, there are risks related to the outcome of each individual dispute. So long as the practice of following precedent is followed, the outcome of an individual dispute will create risks on a variety of others who are not parties to that dispute, and therefore have no input into the information gathering process.

Doctrinal arguments, on the other hand, are much more concise because they are generally circular within their own frame of reference. The essence of a legal argument is to persuade the judge to adopt a particular doctrinal analysis of the case. If one can persuade the judge to analyze the facts within the confines of a particular legal doctrine, the conclusion will normally follow as a matter of course. The proper boundaries of the appropriate sphere of the courts and that of the legislature are determined, we would suggest, more by factors relating to information costs than by philosophical principles.

The doctrine of precedent, that relevantly like cases should be decided alike is an application of the principle of universalizability. Legal judgments are universalizable because they are teleological. They entail a cause-effect relationship between the act which is required or prohibited, and a valued or desired state of affairs. The universalizability of legal judgments, therefore, is a special case of the uniformity entailed in the discourse of cause and effect [Smith, 1976 at 88-108]. The presumption that a legal precedent or case is correctly decided entails that the outcome or holding of the decision is justified in terms of the teleological considerations which furnish the normative foundation for the legitimacy of law.

Argument by precedent justified in doctrinal terms is a very economical, Spartan, and elegant form of reasoning. If one had to argue for each case what the appropriate moral or political justificatory principles are, what the teleological considerations dictate, and how the particular case should be measured against them, the length of time required to make such arguments, and to prove the teleological claims, would be interminable. Argument by precedent, however, avoids this very kind of argument which, although highly relevant, would add tremendous costs to the litigation process. The underlying structure of an argument by precedent runs something as follows:

1. A legal decision that a particular person has a duty to refrain from doing a particular act under a specified set of circumstances may be prima facie presumed as right, just, and proper if it conforms with the underlying teleological factors which form the normative foundations of the legal discourse.

2. If a new dispute is similar to a correctly decided case in all relevant respects, then the resolution of the new dispute will be right, just, and proper if the same resolution is adopted in the new case.

Thus arguments by precedent, using a doctrinal discourse which permits patterns of reasoning which are circular, are classic examples of the inverse relationship between our tolerance for imprecision and the costs of the task. Arguments by precedent are much more imprecise than they would be if all of the underlying teleological assumptions were argued and attempted to be proven. We give much more value, however, to the relative simplicity and speed of the process of argument by precedent than we do, to the degree of precision we would gain if each case had to be argued in its entirety in terms of the underlying teleological considerations at stake. The moment legal issues are argued in teleological terms which require evidence of a factual nature, the cost of litigation becomes prohibitive. High litigation costs might be tolerable for large and wealthy corporations, and the legal system does not prohibit this kind of argument where the litigants are prepared to pay the costs of obtaining a greater thoroughness in argumentation. However, for the majority of litigants the costs of litigation are far too high as it is.

The driving force which best explains the development of case law is the maximization of the benefits of imprecision. Precedential reasoning requires little external information beyond the specific facts of the case. This method of argumentation attempts to find the best match as between the facts of a case and the facts of a precedent within the confines of doctrinal structures which in and of themselves require no external factual information. The machine of legal reasoning is case-based. It is precedent which carries the burden, not doctrinal reasoning nor practical or teleological reasoning. These other factors serve precedent, and not vice versa: they serve only the function of dealing with relevancy in determining which facts are to be matched with which facts of the precedent. Precedential reasoning is a process of classification which requires little external factual information or reliance on controversial moral and philosophical principles. The ambiguity, circularity, and bipolarity of the doctrines articulated in precedents allow a great deal of flexibility in the process of matching the facts of a dispute to the facts of a precedent. Thus, at the doctrinal level, legal reasoning is highly indeterminate and relatively inexpensive so far as information costs are concerned.

Deep Structure Analysis

In a dispute between two individual parties the search for information is limited by their own economic situation; the environment of the dispute is ambiguous; information is vague; experts are imperfect; arguments are forensic; doctrine is nebulous, circular and bifurcated; policies are contentious; and moral and philosophical principles are controversial. How is it, then, that legal reasoning functions at all? If we are to represent legal reasoning in computer based information systems, we must be able to explain the machinery in the decision process which does bring stability of expectations, consistency, and some form of rationality to the judicial decision process of the law, despite the high degree of indeterminacy entailed in legal doctrine.

The function of indeterminacy in legal reasoning is to reduce the cost of decision making by excluding from direct argument in court contentious matters and issues entailing high information costs. These important factors are allowed to function at an intuitive level. If they were brought into contention in court, information costs would sky-rocket. Practical reasoning, policy matters, moral principles, and political ideology are essential considerations for the resolution of disputes between individuals. Our point is not that these do not function in the judicial decision making process, but that they function without disputation because the information costs of arguing them are generally more costly than what could be gained by having to prove the claims made about them. The arguments, controversies, and debates surrounding such issues are on-going. They entail and require a time frame far beyond that within which disputes between individual litigants function, and information which is beyond that which individual litigants can generally afford. For reasons of economy, the arguments themselves within individual disputes must be kept more or less isolated from these grander issues. This isolation is not a matter of rule or principle, but is dictated by the limited resources of the litigants and the economic limitations imposed by maintaining a dispute resolution system at the expense of the taxpayers. The arguments which take place in a court of law do so in the context of ongoing processes of reasoning which are external to the specific case. The arguments are to be found in scholarly journals, texts, and political debates, and in the ideological struggles within any society. There is, however, an underlying "deep structure" which runs through these external debates and the legal doctrines, precedents, and limited legal policies and principles which are directly argued in court. Traditional legal theory, contrary to what Susskind [1987] argues, is inadequate for the task of representing legal knowledge in artificial intelligent information systems. The various forms of rule skepticism such as American Legal Realism or Critical Legal Studies, while recognizing the influence of teleological considerations deny the existence of structure. Those theories of law such as natural law or sociological jurisprudence which privilege teleology, disavow the importance of analytical structure in legal theory. An adequate understanding of legal reasoning requires a legal theory which is both analytical and teleological [Smith, 1976; Coval and Smith 1983; Coval and Smith 1986].

The Yale Law Journal [1991, vol. 100 at 1449] recently published a piece entitled "The Most-Cited Articles from the Yale Law Journal". Nearly all of the named articles entailed a teleological analysis of law or an area of law in terms of what we have been calling "deep structure." Why is it that the kind of analysis which lawyers, students, academics and judges find the most interesting seldom appears in legal argument or in the text of legal judgments? The reason is simple. The information costs of proving claims made in terms of policy and practical reasons are too high to be feasible in a court of law.

Representing Legal Knowledge in Legal Information Systems

The emerging field of artificial intelligence and law offers a new testing ground for legal theories. Presuppositions about the nature and structure of legal knowledge can be tested in the process of attempting to simulate legal reasoning in the computer. Attempts to make machines think like lawyers by representing legal knowledge in the form of doctrinal rules, has failed [Gardner, 1987]. Nor has there been any success in giving a computer-based legal information system a set of first principles from which it can deductively reach conclusions. In learning about human intelligence through the process of simulating intelligence in the machine, we are haunted by an uncertainty about our own thought processes. It would seem that we are not clear about what kind of thinking is ideal and what kind is second best. The issue may be reduced to the simple but difficult question of whether mathematical and logical reasoning is better than less formal thought processes such as practical reasoning, and the somewhat related question as to whether a deductive system of knowledge is in some way superior to a body of knowledge which is neither systematic nor logical. The way in which we comparatively evaluate mathematical--logical--deductive reasoning and less systematic and structured discourse has critical implications for legal theory as well as for artificial intelligence. As far as human thought is concerned, we need not privilege one form over the other since humans can do both, and appropriately choose which form is appropriate for which purposes. For machines, however, it is different. The machine calculates better than humans do. On the other hand, even a young child can manage natural language better than can any computer-based intelligence system. Which is the ideal and which is the second best. Which is the norm and which is aberrant? If mathematical/logical reasoning is best, and a logically deductive system is ideal, then humans should reason more like computers. Artificial intelligence would be the ideal model for real intelligence. If, on the other hand, practical everyday reasoning is the ideal, and deductive, mathematical, and logical reasoning is needed only on particular occasions for particular purposes, then machines will become more intelligent as they become better able to simulate human reasoning. This controversy is replicated in the context of artificial intelligence, as reflected in Rogers Shank's separation of theorists into "the neats" and "the scruffies". Those who work in the field of artificial intelligence and law separate into the same two camps. The "neats" seek to build logical and deductive models of legal reasoning, while the "scruffies" attempt to represent legal reasoning heuristically.

Human language and processes of reason were never invented; rather, they evolved. There are two very important aspects of human knowledge. First is that much of our processing of information goes on at the periphery and outside our conscious awareness. This permits humans to efficiently and quickly tap into and process a vast reservoir of knowledge and experience at a much faster rate, instead of having to consciously recall and think about what each word means, and how the meanings are related in the form of a sentence, and what the connection is between the meaning of the utterance and the vast matrix of knowledge possessed by each sentient human. The other aspect of human knowledge is, as discussed above, the capacity to reduce information costs and increase the speed of knowledge processing by being as imprecise as possible without greatly decreasing efficiency. If we accept that those practices of human communication which we have developed to serve our needs work very efficiently by maximizing the benefits of indeterminacy and intuition, why should legal reasoning be any different? Artificial, mathematical, and logical languages are infinitely too simplistic to deal with the complex conceptual problems surrounding human action. To properly evaluate our methods of legal reasoning, we need to examine the process in the context of the complexity and indeterminacy which surrounds human action and conflicts of interest.

In the field of artificial intelligence the interrelationship and complexity of human knowledge and the intuitive aspect of knowledge processing by the human mind is simulated in the computer by using heuristic rules and conceptual networks rather than rule-like logically deductive structures of knowledge. The indeterminacy of human language and concepts is dealt with by using "fuzzy sets" or "fuzzy variables", that is "classes of objects in which the transition from membership to non-membership is gradual rather than abrupt [Zadah, 1990 at 99; 1984]. According to Professor Lotfi Zadah, [1990 at 100] the pioneer of what has become known as fuzzy logic, "Fuzziness...is a concomitant of complexity....In fact, it is the capability to manipulate fuzzy concepts that distinguishes human intelligence from the machine intelligence of current generation computers."

The UBC Faculty of Law Artificial Intelligence Research (FLAIR) Project has successfully constructed a number of expert systems which can give an opinion on a difficult legal issue in the particular areas of the law covered by the systems, by representing legal knowledge in the form of deep structure rules which are formulated on the bases of a teleological analysis of the law [Smith & Deedman 1987; Deeman & Smith, (1992); MacCrimmon 1989; Kowalski, 1991; Smith, Gelbart, & Graham, 1992.]. The legal knowledge contained within a system is represented by taking a fuzzy doctrinal rule and restating it as a fuzzy factual rule. The fuzzy factual rule is then "defuzzified by breaking it up into a set of single more precise individual factual rules. For example the Land Use Advisor takes doctrinal rules relating to strict liability, such as "the non-natural use of land" rule in Rylands v Fletcher, and restates it as a fuzzy rule about strict liability for non-reciprocal risks. The system then correlates a corresponding set of fairly precise rules which compare and measure reciprocity. These rules are correlated with a set of questions which are asked of the user, and with the decided cases in the database, which are entered in terms of fields containing the facts about the nature of the risk in each particular case.

Almost all legal concepts and categories are fuzzy. There are, for example, a set of five necessary conditions for recovery in the tort of malicious prosecution: (1) commencement of (2) judicial proceedings of a particular kind, (3) terminated favorably for the plaintiff, (4) brought without reasonable cause, and (5) with malice. Each of the above constitute a fuzzy set. Each case where one of the above is in issue constitutes a particular instance of the fuzzy set. Each of the above can be stated in the form of a fuzzy doctrinal rule. Each individual decided case can be stated in the form of a precise rule for that particular set of facts. All of the decided cases on each doctrinal issue can be stated in terms of precise rules which constitutes the set of the fuzzy factual rule which corresponds with the doctrinal rule [Kowalski, 1991]. When a specific case raises an issue, it can be conceived in terms of whether the particular case falls within or without the fuzzy rule. A particular case falls within the fuzzy rule if it is relevantly similar to a case which forms a precise rule within the fuzzy set which constitutes the fuzzy rule. A particular case falls outside the doctrinal rule if it is not relevantly similar to any case which can be stated as a precise factual rule within the set which constitutes the factual fuzzy rule. One case is relevantly similar to another to the degree that it shares the same teleological structure. A case shares the same teleological structure when the same goals or interests are at stake, with similar causal relationships to other sets of like goals or interests.

Conclusion

In conclusion we wish to argue that when the practice of following precedent is applied, and there are no express contradictions in the legal doctrine, legal decisions will be consistent, and if one understands the teleological structure which is isomorphic for the doctrine and the decisions, one will be able to maintain a fairly high degree of predictability as to how like cases will be decided in the future, or the probable outcome of a dispute, given a particular set of facts. Furthermore the decisions will be understandable and explainable in terms of this underlying teleology. Legal reasoning is neither necessarily objective or essentially just. On the other hand, legal doctrine is not nonsensical or irrational. It is not a flawed version of an ideal system of inference. It is practical and effective precisely because of the indeterminacy which is entailed within the process. The complexity of human action and conflicts of interest and the high costs of information require that many of the important factors relevant to decisions not be contested in the form of legal argument in the court. The high costs of argumentation dictates that the controversies which surround much of what is relevant for particular disputes are best kept outside the court room. The process relies on the judge's common sense and intuitive appreciation of the teleological structure which underlies legal knowledge, ranging from the philosophical to the practical, and on the ability to manipulate the doctrinal structure accordingly.

Judges, lawyers, and academics can be conceived of as functioning within the context of a vast information system which reflects a complex matrix of goals and their orderings. Each case which decides in favour of a plaintiff or a defendant resolves a conflict of interests by hierarchically ordering the goals which are pitted against each other in the dispute. The doctrine of precedent produces trends of consistency, with corrections taking place where different circumstances result in conflicting sets of ordering. Legal doctrines are constantly being restated, refined, and modified. Legislation makes more prominent and massive corrections to the goal structure, which in turn will filter their way through the individual cases. If we are to successfully simulate legal reasoning in computer based legal information systems, we will need to take into account a powerful and fundamental aspect of human reasoning---the ability to gain the maximum cost benefits from indeterminacy.

References

Berman, Donald & Hafner, Carole, (1988). Obstacles to the Development of Logic-Based Models of Legal Reasoning, in Computer Power and Legal Language, ed. C. Walter, Quoram Books, New York, p. 183.

Coval, S. C. & Smith, J. C., (1983). Rights, Goals, and Hard Cases, in Justice, Rights, and Tort Law, eds., M. D. Bayles & B. Chapman, D. Reidel, Boston, p. 239.

Coval, S. C. & Smith, J. C., (1986). Law and its Presuppositions, Routledge & Kegan Paul, London.

Deedman, Cal & Smith, J.C., (1992). The Nervous Shock Advisor: A Legal Expert System in Case-based Law, in Operational Expert Systems Applications in Canada, eds., Ching Y. Suen & Rajjan Shingail, Pergamon Press.

Gardner, A., (1987). An Artificial Intelligence Approach to Legal Reasoning, MIT Press, Cambridge.

Kowalski, A., (1991). Case-Based Reasoning and the Deep Structure Approach to Knowledge Representation, in The Proceedings of the First International Conference of Artrificial Intelligence and Law, Accociation for Computing

Machinery Press, New York, p. 21.

MacCrimmon, M., (1989). Expert Systems in Case-Based Law: The Hearsay Rule Advisor, in The Proceedings of the Second International Conference of Artrificial Intelligence and Law, Accociation for Computing Machinery Press, New York, p. 68.

Smith, J. C., (1976). Legal Obligation, Athlone Press of the University of London.

-----.(1984). Liability in Negligence, Carswell, Toronto.

Smith, J. C. & Deedman, Cal, (1987). The Application of Expert System Technology to Case-Based Law, in The Proceedings of the Second International Conference of Artrificial Intelligence and Law, Accociation for Computing Machinery Press, New York, p. 85

Smith, J.C., Gelbart, D., & Graham, D., (1992) Building Expert Systems in Case-Based Law, in vol. 4, nu. 3, Expert Systems with Applications, p. 335.

Susskind, R. (1987). Expert Systems in Law: A Jurisprudential Inquiry, Oxford University Press.

Zadah, Lotfi A., (1984). Making Computers Think Like People, in IEEE Spectrum, August, p. 26.

-----.(1990). The Birth and Evolution of Fuzzy Logic, in International Journal of General Systems, vol. 17, p. 55.