® InfoJur.ccj.ufsc.br

Operational Expert Systems in Canada, Pergamon Press, Elmsford, New York: 1991
The Nervous Shock Advisor:
A Legal Expert System in Case-Based Law

CAL DEEDMAN, LL.B., LL.M. AND J. C. SMITH, B.A., LL.B., LL.M.

THE NERVOUS SHOCK ADVISOR (NSA) was developed as a means of determining whether or not it was possible for noncomputer scientists to build a legal expert system in the realm of case law using a commercially available shell. At the time this undertaking was conceived, we were optimistic, but by no means certain, that such a thing could be done. Others in the field had taken the position that the available shells were unsuitable for building legal expert systems because of the nature of law (Capper & Susskind, 1988; Susskind, 1987).

DEVELOPMENT PROCESS

Problem Selection

Preliminary discussions about the project began in May 1986. Nervous shock was chosen as a subject area because it seemed manageable. This choice of topic was in accordance with the consensus among experienced expert system builders that the domain chosen should have well-defined limits. Nervous shock is a narrowly circumscribed domain within the broader context of the law of negligence. In essence, it is a tort (i.e., civil wrong) that deals with claims for shock caused by some traumatic event. Most U.S. jurisdictions have a similar cause of action called "emotional distress." Although "nervous shock" is a term coined by lawyers that has no recognized medical significance, it deals with roughly the same phenomenon that the medical profession calls "posttraumatic stress disorder." In law, the classic case is the situation where the claimant is a bystander who witnesses something awful happen to someone else and suffers shock as a result, for example, a mother who sees her child killed by a careless driver. However, there are many other less straightforward cases where the courts may or may not award damages for nervous shock. Our aim was to build a system capable of dealing with these so-called "hard cases" where the outcome was by no means clear.

Knowledge Acquisition

Every expert system, of course, represents an attempt to capture some human expert's knowledge. Long recognized as an expert in both jurisprudence (i.e., legal philosophy) and the law of negligence, J. C. Smith served as our expert. Smith had expounded a "Deep Structure" theory of law in a number of books (Smith, 1976, 1984; Coval & Smith, 1986). The theory claimed that while legal decision making is essentially a rule-governed activity the rules that are applied are not the "surface" rules of traditional legal discourse. Building NSA was viewed by Smith as a means of testing his theory as well as building an expert system in a particular area of the law.

The essential legal issue that NSA was designed to grapple with was one of remoteness of damages. In other words, how far down the chain of strict causality would the law go in compensating victims of negligently induced nervous shock? The traditional legal test of remoteness is that a person is legally liable for negligently inflicted damage that is "reasonably foreseeable." Unfortunately, this test is really just a rhetorical formula that is recited to justify legal arguments or decisions. In itself, it has no predictive value. It is necessary to look to the decided cases to get any idea of what the phrase actually means. The issue was further complicated by the existence of two lines of conflicting authority on the degree of foreseeability required. One line of cases held that the damage suffered by the victim of the negligent act must be foreseeable as "probable." The other line of equally persuasive cases said that the damage need only be foreseeable as "possible," a standard which clearly cast a much wider net of liability.

Cal Deedman acted as the knowledge engineer, who would debrief the expert and build the system. Deedman was a lawyer with 10 years' experience in legal practice, mainly litigation. He had returned to law school in 1986 to do a master's degree on legal expert systems. Deedman was to build NSA as part of his thesis work. In contrast to Smith, Deedman's orientation was thoroughly pragmatic as a result of his having spent all his professional life in the "real world" of practice. His primary motivation was to build an expert system that would be useful to lawyers, thereby demonstrating that such a thing could be done by a noncomputer scientist using existing software tools.

The knowledge acquisition process began by Deedman reading Smith's books to familiarize himself with the author's general ideas. A lengthy series of interview sessions followed, during which these ideas were discussed and a conceptual structure of the law of nervous shock was gradually hammered out.

It is important to realize that no logical overall structure of this area of the law had previously existed. Most nonlawyers find the lack of organization difficult to believe and somewhat disconcerting. However, this formlessness is characteristic of the Anglo-American common (case) law tradition, as opposed to the European civil law which is codified. Basically, a body of case law consists of a large collection of cases that have been accumulated over time, often hundreds of years. The cases are supplemented by commentaries found in the legal literature. An area of the common law grows by a process of accretion, somewhat like a coral reef, as new cases are decided and added to its corpus. The problem of teasing a formal structure out of a formless mass of cases is further complicated by the fact that the decisions in some cases are sometimes logically inconsistent with the decisions in others. Indeed, it is for this reason that from time to time lawyers and judges are obliged to indulge in the rather complicated mental gymnastics. It is the lawyers' attempt to reconcile the irreconcilable; the legal equivalent, one might say, of the alchemical exercise of squaring the circle. In all the foregoing respects, nervous shock was a typical slice of the common law.

Given the nature of case law, what the cases said had to be considered as determining the issue in the event of any fundamental conflict between the cases and Smith's theoretical perspective. NSA would be of no use to lawyers if the advice it gave represented the idiosyncratic view of an expert rather than the existing state of the law. A legal opinion that cannot be justified in terms of case authority has little persuasive force. While the framework of NSA was being built, therefore, exhaustive legal research was undertaken to ensure that all the reported cases on nervous shock had been considered before the design was finalized. A total of 97 reported cases were gathered from among all the commonwealth jurisdictions governed by the same law. These included cases from Canada, Great Britain, Australia, and New Zealand.

Once the cases had been gathered, they were reviewed in the light of Smith's preconceived ideas. Each case was then summarized as tersely as possible and classified according to its key features. Most cases were reduced to a digest of a just few paragraphs with the idea that the user should be able to grasp the material facts almost at a glance. Reading more than a screen or two of text on a monitor is a tedious process. Furthermore, NSA would have taken up too much disk space if the full text of all the relevant cases was stored. Also, the case report was available in the library if the user wanted to read the whole thing.

The only exception to the procedure of summarizing cases was made with respect to a case that Smith deemed to be the leading case on the law of nervous shock. We decided that this NSA should display this case in full text if the user wanted to see it. The rationale behind this was as follows. There is usually one case from the highest court in any jurisdiction that reviews and re-states the law on a particular subject. The persuasive force of this case makes it essential that a lawyer be aware of what it says if he or she is dealing with a case of the same type, whether or not the facts are similar. Since judges assume that lawyers have read these leading cases, to be unfamiliar with them can be a source of considerable embarrassment.

In short, the knowledge acquisition methodology we used could best be described as a concurrent top-down-bottom-up approach (i.e., "Hypothetico-deductive"). The process of attempting to formalize Smith's expert knowledge turned out to be a useful exercise in clarification for him. He was surprised to discover that he was by no means as certain about some things as he had imagined, even though he had been dealing with the area for years. On a number of occasions he had to rethink his ideas when faced with deciding what the outcome should be in some hypothetical case. We found the "what-if" method of creating hypothetical cases a good way of focusing our discussions and resolving ambiguities.

The exercise of mapping out the entire area encompassed by the law of nervous shock provided us with an interesting example of the human tendency to think in watertight compartments once we have organized our knowledge into categories and these categories have hardened like cement. As mentioned earlier, the textbook case of nervous shock involves a situation where the plaintiff suffers shock as a result of something terrible that happens to somebody else. However, it is also possible for plaintiffs to recover damages for nervous shock in circumstances where they themselves are the victims of the terrible event. Smith realized this in one sense, yet he had conceptualized those cases as belonging to another area of the law commonly referred to as the "thin skull" cases. "Thin skull" refers to a situation where the victim suffers a degree of harm out of all proportion to the seriousness of the negligent act because of a peculiar susceptibility on the victim's part. Smith had not thought of these cases as falling within the domain of nervous shock. This conceptual blind spot was discovered while sifting through the cases amassed in the course of our research. The most surprising thing about this oversight was that it seemed so obvious in retrospect.

Eventually we were satisfied that we had discovered the material elements of a nervous shock case. Elements are the conceptual building blocks out of which cases are built. They are a related set of abstractions that must be satisfied in any given fact situation for a case to succeed. To take a simple example, in a theft case the elements that the prosecution must prove to make a case against the accused are as follows:

1. property--something capable of being stolen

2. Ownership--belonging to somebody else

3. intent or lack of consent-was unlawfully

4. deprivation--taken

5. identity--by the accused

6. jurisdiction--at a certain time and place.

If one of these elements is missing, the case falls apart. The elements can be thought of as the links in a chain. Like a chain, a case is only as strong as its weakest link. In any given case, the facts that will satisfy a particular element will be different; however, each element must be proved in some way for the case to succeed.

The elements of a criminal case are fairly easy to ascertain because the criminal law is set out in a code. On the other hand, we had to figure out for ourselves what the elements of nervous shock were because they had never been clearly identified. All that had existed before was the nebulous "foreseeability" test. In our estimation, the material elements of nervous shock were as follows:

1. Intent. Fault was presupposed. We simply needed to know whether the act was negligent or deliberate because no issue of remoteness arises in the latter case. A person is legally responsible for the consequences of a deliberate act.

2. Victim. It was important to know from the outset whether the plaintiff as well as a third party had been a victim of the shocking act. If so, different considerations would apply with respect to elements 3, 4, and 5.

3. Harm. The harm suffered by the victim(s) had to be serious enough to support a valid claim. The degree of harm affected what needed to be established in order to satisfy element 5.

4. Relationship. Unless the plaintiff was also a victim of the shocking act, it was necessary for some sort of relationship to exist between the third-party victim and the plaintiff.

5. Exposure. The plaintiff had to be in some way exposed to the shocking event or its after effects. The extent of exposure required varied depending on the seriousness of the harm.

6. Damage. Except in unusual cases like trespass, every lawsuit requires proof of tangible damage to succeed. Nervous shock was no exception. Furthermore, the courts have recognized certain specific mental and physical symptoms as constituting evidence of nervous shock. A mere assertion on the part of a claimant that shock had been suffered would be insufficient to prove the case in the absence of evidence of this type.

NSA was designed with a chain-link structure (Fig. 6.1) and an "early kick-out" feature. Each material element had to be satisfied in turn for the hypothetical case the system was dealing with to succeed. Although NSA was to decide legal issues, it would do this by posing purely factual questions to the user. If the outcome of the consultation was successful, the conclusion would be justified by precedent cases matched with the facts supplied by the user. If the user's facts failed to satisfy any link in the chain of elements, the consultation would terminate at that point. The reason for the failure would be given to the user and the negative outcome supported by case authority.

We grouped the precedent cases into three categories: on point, relevant by analogy, and contra. The on-point cases were those in which all the material elements of the case exactly matched the user's facts. The relevantby-analogy cases were those that were similar, but not exactly the same.

Based on Smith's expert opinion, NSA would predict success in some cases, even where there was no on-point precedent. In those instances, the outcome would be justified with cases relevant by analogy. Justification in terms of analogy was inevitable given that NSA was designed to be able to deal with "hard cases." In fact, legal reasoning is basically reasoning by analogy. A lawyer tries to persuade the judge that the decision in some precedent case which favors the client's position should be applied to the facts before the court, even though the facts of the precedent may not be exactly the same. Indeed, if the facts were the same, there would really be nothing to argue about and the case would be unlikely to wind up before the court. A lawyer's skill lies in being able to argue successfully that the differences are not material. The opposing lawyer's stance, of course, is to take the opposite approach and argue that they are.

The contra cases were those that, in Smith's opinion, had been wrongly decided. They were aberrations that simply could not be reconciled with the law as laid down in the overwhelming majority of other cases. However, for the sake of completeness, in relevant circumstances NSA would give them to the user with an indication that they went against the weight of legal authority.

In situations where NSA predicted success, users would be given the option of perusing two other sets of cases. First, those that matched the particular symptoms of nervous shock displayed by the client. Second, those in which the defendant had won. The latter category was to be included to give the user an idea of the cases the other side might rely on, and to balance out the plaintiff-oriented perspective the system. No matter what the outcome of the consultation, users were to be offered the opportunity of looking at the full text of the leading case on nervous shock.

Knowledge Representation

We decided to use a rule-based ("production system") shell to build NSA because we believed that legal knowledge could best be represented by rules.

After some shopping around, we chose M.l, a product marketed by Teknowledge Inc. of Palo Alto, California. (M. I is no longer sold. Teknowledge decided some time ago to get out of the software marketing business and shift its focus to consulting and building custom systems.) M. I was a direct descendant of EMYCIN, the first shell system. EMYCIN was a spin-offof MYCIN, the well-known expert system developed at Stanford University in the 1 970s to diagnose and prescribe treatment for bacterial infections of the blood.

M. 1 seemed like the best, reasonably priced tool on the market for our purposes for a number of reasons. For one thing, it ran on an IBM PC with 640K of RAM memory. At the time NSA was being developed, IBM Canada Ltd. had donated a large quantity of hardware and software to the law school as part of a cooperative project on law and computers. Therefore, whatever tool we chose had to be IBM compatible.

M. I had a number of other attractive features. For one thing, it had built-in certainty factors. We felt that any opinion NSA gave should be weighted in terms of certainty. Lawyers are accustomed to talking about a case's chances of success or failure in percentage terms among themselves and to clients. M.1 also had an explanation facility that we felt was essential. Less important features were that its rules were written in an English-like syntax and any standard text editor could be used to create the knowledge base.

After we had acquired M. I, Deedman was struck by a curious similarity between the inference engine's backward-chaining control strategy and the way lawyers think. The lawyer's goal in any particular case is dictated by the needs of the client. Although lawyers are constrained by the rules of the legal system and bound by a code of ethics, they are essentially intellectual mercenaries. They start with a goal such as recovery for the plaintiff in a particular case and then, working backwards, seek every available means at their disposal of achieving it. Being accustomed to the backward-seeking approach to problem solving may have made it easier for Deedman to learn M. 1.

Knowledge Encoding

Implementing NSA was a time-consuming and frustrating process. Unless one has the "hacker" mentality, writing and debugging code is not much fun, and even less fun to read about. Suffice it to say that the knowledge base was written by Deedman in consultation with Smith over a period of about 6 months, with many technical modifications along the way. The system was substantially completed by the beginning of January 1987.

A brief technical description of NSA is as follows. The uncommented knowledge base consisted of 225 IF-THEN rules occupying about 75K of memory. All facts were elicited from the user. Throughout the inference process, extensive use was made of patternmatching variables and the intermediate conclusions stored in M. 1's "cache" (i.e., Dynamic database). At the end of the consultation,the contents of the cache were used to build a profile of elements in the form of a list structure. This list structure was matched to a predefined table of lists representing all possible outcomes of the consultation. Each list was identified by a unique integer that acted as a pointer to a menu containing a collection of relevant case names. The menu was displayed through a linkage to the knowledge base written in C with the assistance of Doug Arnold, a computer scientist employed by the law school. The case digests of the cases selected from the menu by the user were retrieved from a flat text file indexedto a list of case names. The external files were quite large: the menu file was 162K and the text file, 191K. Since the M. 1 inference engine required 232K, the completed system needed to be run off a hard disk. Some examples of the MA code are given in Figures 6.2 through 6.4.

A sense of how NSA works is given by the sample consultation in the appendix to this chapter. The consultation is based on the hypothetical situation that precedes it. The remarks in square brackets are for clarification purposes only and would not appear in an actual consultation.

Knowledge Testing and Evaluation

NSA was first tested against a class of Smith's law students. The students were given a nervous shock problem to research. More than 85% of the students agreed with the opinion given by NSA as to the correct outcome. Thereafter, ongoing feedback about NSA was obtained by demonstrating the system to graduate students, faculty, and visitors to the law school. The response of the legal professionals who were shown the system was overwhelmingly favorable with respect to its usefulness.

During 1987, NSA was also shown across North America at three major conferences: the Association of American Law Schools Annual meeting in Los Angeles (January 6, 1987); the First International Conference on Law and Artificial Intelligence in Boston (June 4, 1987); and American Association for Artificial Intelligence in Seattle (August 1987).

IMPLEMENTATION PROCESS

Deployment Strategies

NSA covers such a narrow, specialized area of law that cases falling within its domain do not arise very often. In a sense, this is a plus because the system provides a rapid and reliable means of researching legal territory with which most lawyers are unfamiliar. However, precisely because such cases come along so infrequently, the system does not have the sort of everyday usefulness that is a feature of expert systems dealing with more mainstream types of subject matter. Therefore, rather than distributing the system to a whole series of locations where it might never be used, it was decided that the best strategy was to make its existence known to members of the profession through widespread publicity (newsletters, continuing legal education materials, etc.), and invite them to come and use it at the faculty of law if they had a nervous shock case.

This strategy has worked well. Since NSA was completed, a number of lawyers have used it to help them prepare actual nervous shock cases. Several of those lawyers had done their own legal research before coming to see us. In each instance, NSA was able to come up with relevant cases the lawyers had missed. One lawyer estimated that consulting NSA for 20 minutes had provided material that would have taken 50 hours to collect using conventional legal research methods. Even then, there was no guarantee that all of the precedent cases provided by NSA would have been found by the human researcher. Legal research tends to be a hit-or-miss proposition because ofthe way reported cases are scattered among the different series of law reports.

NSA was used by one judge presiding at a trial involving a nervous shock claim. The judge found the system useful, but decided to make a ruling at odds with its opinion. The judge's decision is under appeal. We are awaiting the outcome of the appeal with interest.

Technology Transfer Strategies

The main purpose of our work was to impose a formal structure on an area of case law so that the legal knowledge could be manipulated computationally. Since we had used a well-understood technology to achieve this end, the nature of our application was of primary interest rather than the mechanics of its operation. Consequently, NSA was much more intriguing to people in the legal field than to computer scientists. A number of law schools approached us and expressed interest in acquiring NSA as a teaching tool in torts courses and as a means of showing how computer technology could be of use in legal practice. In fact, we have provided the system free of charge to law faculties throughout Canada, the United States, and Australia.

Maintenance Strategies

Our experience in building NSA in M.1 showed us that although purely rule-based systems work well they have a number of inherent shortcomings. For this reason, instead of spending time maintaining the original system, we decided to attempt to overcome these inadequacies by exploring alternative means of knowledge representation with a view to creating a second version of NSA.

The problems we encountered with the rule-based version of NSA (hereafter referred to as Version 1) were as follows. First, the path that the backward-chaining control mechanism took through the knowledge base was difficult to follow. The flow of control became more and more convoluted as the body of rules grew larger. Secondly, since not only the logic, but also the structure of the system was embodied in rules, the underlying conceptual framework was not readily discernible. Finally, updating the system to reflect changes in the law was difficult because of the many dependencies that existed in the interconnected web of rules. While these weaknesses could be palliated to some extent by good rule-base design, they could not be overcome entirely in a system that relied on rules to do everything.

Another major drawback of Version 1 was not related to its rule-based structure, but rather to the way in which case authorities were stored and retrieved. Given the need to justify legal arguments in terms of precedent, the case storage and retrieval mechanism constituted a central part of the system. In Version 1, the cases were stored in a large flat text file and individually indexed to an even larger menuing system. Depending on the outcome of a consultation, a particular menu containing a list of relevant case names was displayed to the user. The text of the case was displayed when the user selected the case name from the menu. Each termination point in a path of inference had its own tailor-made menu.

Even though the retrieval system functioned well, it was too cumbersome and difficult to maintain. The same case name would appear in many different menus if it was relevant to a number of possible outcomes. This meant that the name of a new case would have to be entered in a variety of locations if the court's decision was applicable to a wide range of circumstances. It was clear, therefore, that the data retrieval component of NSA also needed to be improved.

After our experience with Version 1, we began exploring ways of remedying the perceived shortcomings of the system. We eventually decided to adopt a hybrid approach to knowledge representation. The conceptual framework would be implemented as a series of related objects that would be referenced and reasoned about by rules. This approach would enable us to separate structure from logic in a clear-cut way and would automatically modularize the system.

Our other fundamental requirement was that Version 2 should be capable of being easily linked to a relational database in a way that would permit relevant cases to be retrieved and presented to the user in a dynamically assembled menu. We also intended to make use of the database to give Version 2 the flavor of a case-based reasoner by the way in which the precedent cases were handled. Rather than simply lumping the cases into three broad categories as we had done in Version 1, we would enable the retrieval system to rank them in terms of persuasive force by weighting certain key features such as jurisdiction, level of court, and the closeness of match with the user's facts.

With the foregoing considerations in mind, we started looking around for a suitable tool with which to build the new system. We considered a number of options and ultimately decided on Intelligence/Compiler, marketed by IntelligenceWare Inc. of Los Angeles (see Parsaye & Chignell, 1988). This software met our basic requirements and seemed scaled at exactly the right size for our purposes. It had the added desirable feature of offering hypertext as a medium that could be used for storing case digests. In effect, the hypertext feature would allow the user to break out of the confines of a specific termination point with its limited selection of relevant cases. If other cases mentioned in the body of a case digest were defined as hypertext terms, then clicking. on them would enable the user to jump from case to case and navigate through the case base in a nonlinear fashion.

Version 2 is now being built. We anticipate that it will be completed by December 1990. It will incorporate several changes in the law that have come about since Version 1 was completed. We intend Version 2 of NSA to be the first of a series of expert systems dealing with the law of negligence.

REFERENCES

Capper, P., & Susskind, R (1988). Latent damage law-- The expert system. London: sutterworths.

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

Parsaye, K., & Chignell, M. (1988). Expert systems for experts. New York: John Wiley & Sons.

Smith, J.C. (1976). Legal obligation. Toronto: University of Toronto Press.

Smith, J.C. ( 1984). Liability in negligence. London: Sweet & Maxwell.

Susskind, R. (1987). Expert systems in law. Oxford: Clarendon Press.

APPENDIX: SAMPLE CONSULTATION HYPOTHETICAL

A man is standing on a downtown street talking to a friend. His 10-year-old son is part way down the same block on the sidewalk watching workmen on a construction site where a new office tower is being built. The skeleton of the building is in place and work is going on at several different levels which are as yet unenclosed. On one of the upper floors, a workman has negligently failed to secure a piece of machinery properly. The machine begins to roll toward the edge of the building where it borders the street. One of the workmen yells a warning just before the machine rolls over the edge. The father turns and sees the machine fall and crush his son to death. He suffers nervous shock. He is often depressed about his son's death and sometimes contemplates suicide. Can he recover?

Consultation