During the last 10-15 years, very substantial research funding has been made available to develop programs in the area of AI. In the early stages, confidence is high and much of the discussion has been notable for the distinct lack of modesty of aims. John McCarthy (Head of Stanford University's AI Lab) stated in 1973 that:

As the confidence of AI researchers increased, they began to look for more and more areas within which their new technology could be set to work. As McCarthy suggests, it seems that the AI researchers had to conquer the world all over again. In many ways, it was perhaps surprising that it took them so long to discover law as a potential application domain. Whilst modern lawyers may be amongst the cheapest professionals to train, they are far from being the cheapest to hire. The mounting cost of professional legal help, together with the increasing complexity of many modern legal cases was already a cause of widespread concern. In complex criminal cases, especially those dealing with fraud, insider dealing and the like, people had already begun to ask whether juries were still capable of fulfilling their function. In civil cases, greater use was being made of arbitration and other procedures to keep disputes away from the courts.

Once one begins to talk of bringing in "expert assessors" to replace the jury in criminal cases, or to assist the judge in complex civil cases, then it only remains to determine how that expertise could, or should, be made available. If this expertise could be manipulated by a computer, then we may have found yet another area where the "technological fix" could be utilised to solve mounting social problems. No wonder that a good deal of the initial work in the area of AI and Law, has been done on tax and social security matters - areas which are paradigmatically, complex and detailed. In the area of tax it is notoriously expensive to avail of the expertise. In the area of social security it is notoriously difficult to find the expertise. What self-respecting lawyer would want to assimilate the complexities of the social security system, and to have only the prospect of impecunious clients?

The prospect is then that the new artificial intelligence techniques would be a great leveller, in that they would be able to help both rich and poor alike. In fact, not even artificial intelligence experts are so public spirited, and such social security systems as are being put into effect, are being provided to government departments - the public rich as opposed to the private rich.

Why should the computer scientists think that the law is a suitable area in which to develop their techniques? I have argued that these developments have been encouraged by recent legal theorists, who have unwittingly provided the computer scientists with the rationalistic and formalistic framework within which their techniques can flourish.

Most notable in this regard are the claims of H.L.A. Hart, which we discussed earlier, to the effect that:

Law is a system of rules.

The law consists of clear rules which do not require judgment in their application from case to case.

The law is identified by the use of a Rule of Recognition

The law changes in accordance with Rule of Change

Devlopment and change comes about in the law in accordance with "rules of change" which allow for the orderly transition from one state of affairs to another. This, of course, is equally a factual matter.

As Neil MacCormick has pointed out in his book on Hart, this view of legal philosophy (derived as it is from the empiricism of David Hume) suggests that law may be understood as a manifestation of attitudes to patterns of behaviour - both of which may be understood as a matter of "fact". The idea that attitudes should be seen as matters of fact, is one which finds a resonance within the judicial community. It has often been said in legal cases that the state of a man's mind is a much a matter of fact as the state of his digestion.

If this view of legal philosophy is correct, it provides us with a view of rules which have a "core of certainty" where the person applying the rule has no choice but to act in accordance with the rule's settled meaning. The use of discretion is not ruled out altogether - it is acknowledged by the idea of "open texture". This is apparently necessary because we are to some extent unsure about matters of fact, and we also have an indeterminacy of aim. However, it is clear that in Hart's account of things, this aspect of open texture is rather marginal to the operation of the legal system. He speaks of there being a "fringe of open texture". In what follows, I will be suggesting that our computer scientists have found rather more of this "open texture" than they would like, and considerably more than Hart allows for.

In one sense, the development of computers should have allowed us a good opportunity to road-test the Hartian hypothesis. One of the difficulties which has apparently faced a proper formulation of the rule of recognition is that of its complexity. With computers, which deal particularly well with complex, but non-judgmental matters, we have an ideal opportunity. We are no longer troubled by complexity. So long as we have clear, certain and non-judgmental rules to deal with, we can, given the memory capacity and speed of operation of modern computers, quickly give articulate expression to these complex entities, as is done with similarly complex entities in mathematics. So let's have a look and see what progress is being made with intelligent computer applications in law.

Our task now is to look at some of the attempts which have been made to computerise legal knowledge. In doing so, we will find out a great deal about the view which computer scientists have of law and which is not too far removed from that which Hart had. Probably the major shift which is going on is that they are finding more and more "open texture" and surprisingly little of clear rules and cores of certainty. This has led to an alteration of objectives, and in some cases, one can detect a noticeable shift in the descriptions of what it is that they have been trying to do.

It has to be said that the rate, and stage at which many of the programs are being abandoned, should not only give rise to cause for concern, but lead us also to question the estimation of the progress which has been made. We find that after a number of years tackling a particular problem, a number of the leading researchers are beginning to say that whilst they have solved all the theoretical issues in the way of the program's succesful completion, (and having spent a good deal of research money in the process) they will not however continue with its implementation. To sidestep the "further development" aspects in this way and to give up just when the going gets easy does not fit in well with one's knowledge of human experience. Whilst the initiators of some of these programs are now taking to dealing with new, and significantly different issues, which will no doubt present them with ever increasing intellectual challenges, we should perhaps take the opportunity to do a bit of stock-taking. I would like to know precisely which "theoretical issues" have been solved, and which have been ignored - or even worse, which of the theoretical issues they may yet be unaware of. This is not my attempt to speculate about the future of philosophy - but rather to suggest that our computer scientists may have been rather less aware of the theoretical issues than they ought to have been.

Because no country wants to be left behind in the race to develop smart machines, governments and funding bodies have been willing to provide money for the development of computer research in some areas because of the obvious success of computers in some others. This is rather similar to the idea that academics should be appointed to teach on the basis that they were once able to learn. It does not follow that they make good teachers, or that the computer fix will work in all areas. The experiences in the area of computers and law might more properly seen as examples of failures rather than as limited successes. This has to do with the lack of interdisiplinary input at the research and design stage, and a lack of clarity about the real research goals. Given what a number of people in the area are now saying, it appears that the research programs are designed more to advance computational skills than to make a serious contribution to the solution of legal problems. The legal funding bodies should be mindful that things are not always as they seem.

Thorne McCarty, in developing the TAXMAN programs suggested that legal reasoning was clearly amenable to computational techniques. In 1980 he stated that:

He now accepts that, "we have not advanced very far at all in these past ten years." It is incumbent upon us to assess the reasons for this lack of progress, and to determine whether the answers which are now being proposed have any greater prospects of success than those put forward 10 years ago. I will argue that the reasons for the lack of progress are not due to new problems which have been encountered along the way, but because of failures to address methodological problems which have been present all along.

If we look to McCarty's recent report concerning his own progress in the area, he points out how important it is to distinguish practical from theoretical objectives. On the practical side we have "rule-based" expert systems which do "backward-chaining inference" from specified goals. Referring to Susskind who "cites a consensus among diverse jurisprudential theorists" McCarty suggests that the system developed by Susskind and Capper is "extremely sophisticated in its legal analysis." (ibid) p190. I shall argue that McCarty (along with many of the other people involved in the development of legal expert systems) have not, even yet, developed a proper understanding of the sophistication of legal analysis. McCarty points to the following types of systems which are under development:

Legal planning systems:

Apparently, the development of a tax planning system would require:

(a) a language in which to describe the desired artifact

(b) a statement of the constraints on an acceptable system
(c) we must then solve a massive search problem

These techniques, he announces, when applied to law, "would give us a true legal planning system." He seems to suggest that the techniques can be developed, and then applied to law, as if this could take place in a two stage process. I shall argue that whilst this a common approach to the solutions of problems by computer scientists, it does not hold for legal analysis. Also, by substituting "massive searches" for "pattern matching routines" McCarty indicates that he is placing the emphasis on the devlopment of more computing power and expertise - more grunt. I would suggest that there cannot be a succesful engagement with the nature of legal analysis until McCarty, or some members of these research teams, really begin to study law and make a serious attempt to understand the application domanin. They have clearly failed to do this so far, and as a result, seriously undermined the value of the research in which they have been engaged.

It appears that he has now abandoned any further development of those programs, without, it should be added having ever attempted a full implementation of them. Indeed, only one of his students actually attempted even a partial implementation. [McCarty: 1990 p195] His more recent efforts have gone into the "invention" of a "language of legal discourse", (the title of his recent book). He now refers to LLD as rich, complex and subtle, and suggests that the "proof procedures for the language are extraordinarily complex". One can see that he has a new hobby horse, but that in terms of the methods he employs, he has failed to change his ways. The failure so far (he supposes) has to do with the lack of sophistication of computational methods - more rigourous "proofs" are called for - it has not occured to him that the problem could really be in his lack of understanding of legal dynamics.

If we look at the developments of one of the major British teams - that at Imperial College, London with Kowalski, Sergot, et al, we can see clear evidence of a similar approach. Whilst considerable resources were expended in formalising the British Nationality Act in terms of the symbolic logic involved in "extended horn clauses", they accept that the program they have developed would not in its present state be of any use in either assisting a legal expert or of itself providing legal expertise. Why? Because their representation of the BNA was undertaken with no expert legal assistance. They have therefore produced something which they now claim to be a "layman's reading of the provisions". It is clearly the case that very few laymen would be able to make very much sense of their reading of the BNA when restructured in extended horn clauses. They do add that whilst there are "no outstanding technical obstacles" to be overcome to finish off a program in this way, they will not of course do so, because "this stage of the work can often involve a considerable amount of work and extra progamming effort". [Kowalski and Sergot: 1990]

Whilst not willing to change their ways, the computer scientists have been keen to find new application domains in which to "develop" these techniques - and, of course, to avail of the grants which go with them. Yet in the last 10 years, despite the increasing attempts to develop computer programs in the area of artificial intelligence applications to law, there have been none that have succeeded or even come close to success - the bureaucratic information management systems such as those developed by Mead and Johnston involve a different approach. Despite the notable failures, the continuing emphasis is on the computational side of the problems. The problem is not with the nature of legal knowledge, but with how that knowledge is to be represented in computer programs. Anne Von-der-Leith Gardner, one of McCarty's PhD students who has recently published the product of her research, recognises that the law does not come ready packaged in formalistic terms, but she then proceeds to deal with it as if it did. But as she goes on to say in the computer jargon, "this stage of the process is as yet unimplemented". The assumptions which she works with regarding the differences concerning statute law and case law are also incorrect.

It need hardly be said that in the last few years, the general perception of the utility and use of computers has taken a quantum leap. They are now part of the standard equipment in the home, the office and the school. Whilst the transition from pencil and paper to computers has been difficult for some of us, the up and coming generation will have used computers before they had teeth, or could walk or talk. We now have school children who have never known life without them - and they often understand the way in which new programs work more quickly than their teachers. It may even be the case that the pace of change in the next few years will accelerate rather than lessen. Computers are now being used to design, develop and manufacture the next generation of computers. Many, if not all of us, will have to make strategic decisions which may well have a profound effect upon our life plans.

Do we regard these things as another one of those passing fashions which will in time run its course and fade away? If so, our decision to stick to the ways of thinking and working which we have been used to, will be a demonstration of our wisdom, and will show that we do not get distracted too easily. By keeping our eye on the ball, whilst others go off to play with every new toy, we will forge ahead with the essentials, and end up ahead of the rest of the field. Ronald Stamper has pointed out that doubling the finanacial investment in capital expenditure in the US banking industry - largely as a result of IT - has given no significant improvement in labour productivity, and only slightly improved capital productivity..

But what if we are wrong in this? What if the changes we are witnessing are structural, substantial, irreversible and more rapid than at any other time in human history? Stamper also points out that each decade, the power of computers rises about 100 fold, their communications ability about 25 fold, and their reliability about 5 fold. Many of us could well find that within a short space of time, what we regard as our skills may well be regarded by others as quaint and funny - but that they no longer have any need for them. What will future employers think when you explain that you "know lots of law" but cannot type and that you do not know how computers work? If the experts we are considering are right, they will have expert systems which will tell them what the law is but which you (or your fellow students) cannot use. A scary prospect maybe when you look at it in that way. But surely we are smarter than that - we would see it coming. But would we? Why was it that the many thousands of compositors working with lead type to set up our books and newspapers, just a few years ago, did not see that the need for their skills would disappear virtually overnight? The answer to this apparent lack of foresight may well depend on at least three factors:

A Belief: irrational maybe, but pervasive. It is that if things have been a certain way for many years, maybe even hundreds of years, then although change may be evolutionary or incremental, it could not now be radical or revolutionary.

A Psychology: to explain and support the belief. A fear of the unknown and of our inability to deal with it. If the whole of our life has been structured around things being done a certain way - earning a living at the factory / university, friends and maybe family work there too, our work, our social life and maybe even our dreams revolve around that understanding of the world. The prospect of the factory / university just not being there next week is too awful to contemplate. So we get on with our lives, we do not think of such dreadful things and we may even hope that by wishing it so, it will remain so. Recently some students complained about my teaching because it was "different" from the year before - they did not want to be treated like guinea pigs. My reply was that you should expect our teaching to be different EVERY year - this is a university, a research and development environment. Change is (or should be) ubiquitous - risk taking essential.

A state of affairs: based on ignorance. It affects us all. No doubt we all like to think that we are knowledgeable, and with television, pocket organisers and the fact that we live in the 1990s rather than the 1890s, we naturally know more things than people used to know. But it is undoubtedly the case that no matter how many things we know, in relation to the things which can or should be known, our intellectual grasp on the world is tiny.

Of course, I may feel that I know about cars, (or law) having made them for 20 years. I take the view that I am not a narrow minded person, and might well believe that if anyone knows about the sort of changes which may come about in car design, then I am in a better position than anyone else to see it coming. The answer is, of course, that revolutions do not always come from within, but may also come from without. The challenge may not be to envisage new styles of cars but life without cars at all. One can readily see that if I am at present making a contribution as an important and valuable member of society, then it might be difficult for me to envisage how that society could continue to operate without making use of the knowledge and skills for which they are at present paying quite handsomely. You as students will be faced with difficult choices. At the present moment in time the ANU has increased its numbers of law students from an intake of around 80 a year to some 230 a year. Most of the other law schools are doing likewise. In addition to this, there are some 14 new law schools in the process of being established. We now have more students in training than we have legal practitioners in practice. By the time you graduate, a law degree per se will not be the passport to success that it used to be.

If all you are learning at law school is substantive "black-letter" law, then by the time you graduate, even those skills may not be needed. Computers may be able to manipulate that sort of information much better than you can. You need to think about how you can gain a competitive edge - over your fellow students, and over the computers. Your best hope of survival may be to understand what the computers can and cannot do - and then ensure that you can do something else. In a straight competition between a law student and a computer, you have not chance of winning. Your additional problem is that many of those responsible for teaching you are skilled practitioners of the old ways of doing things. The question then arises as to how the old relates to the new, and the extent to which they will be able to inform you about it.

We know in general, of course, that sometimes the new ways of thinking are not necessarily incremental developments from the old. We do know that the emergence of the Commonwealth of Independent States may well have been part of a bloodless revolution, but few would doubt that it was nevertheless a revolution on a scale not seen before in human history. We may pause and think "what a funny old world" when we see former heads of state in handcuffs or on the run from the police, unrestricted public access to KGB files, and Mikhail Gorbachev looking for employment. We might be less aware of what it all really means to the millions of people directly affected by it. But nevertheless, we still feel that our lives will carry on much as before. But are we really any more in control of our destiny than were the former citizens of the USSR? And if we cannot exercise control, can we influence the outcome? Presumably we can, but it may take some effort. Whether we are prepared to make that effort might determine whether we end up as winners or losers in this lottery of a lifetime.

We can see that the new methods of typesetting based on computers and electrical impulses bear no relationship to the old mechanistic methods based on the manipulation of pieces of lead with tweezers. There is nothing in the background, training and upbringing of a compositor which would have equipped them to appreciate the dire threat which integrated circuit boards posed to their very livelihood. The question which we now have to ask is whether we would be in any better position to see whether this was happening now in other areas? Two of the areas which will be of particular interest to us will be education and the law.

One might suggest that education is an area which is already being subjected to fundamental changes, although we might sometimes wonder to what extent we are participating in them. When Ronald Stamper said in 1989 that "the lecture form survived until comparatively recently", and that "the traditional lecture has been made obsolete by the invention of xerography" we might assume from this that many of our universities are more like a living museum than a modern educational institution. Change may be more noticeable at the present time in our junior and high schools, where the new technology has an obvious presence - but one might be tempted to say that the show has only just begun. Computer based tools now give you access to basic materials through the use of hypertext, many universities now set tutorial problems through CAL and use them in the assessment of completed tasks.

The revolutionary view will not try and adapt the new tools to the old ways of thinking - instead, whilst speaking of the "electronic textbook", we will be thinking of interesting ways of providing information when needed at a level which is appropriate to the needs of the user. It is the user who will be able to determine the level of complexity of the information which is provided, rather than having this set in advance by the information provider. You will in the near future compile and print your own textbooks.

However, this will require the information to be structured in ways which will require significant adaptations from present methods. Instead of being presented serially, as in the present textbook, the information will have to be presented more 3-dimensionally. This can be through the use of hypertext, statistical matrices for the analysis of precedents as in SHYSTER, or for the presentation of legal argumentation in Pam Gray's model where she speaks of 3-dimensional knowledge structures and "virtual reality".

Could it really be suggested that one of our most ancient institutions - the legal system - could be under threat? The incremental view would point to the fact that lawyers now use database retrieval systems to recover full text of precedents and statutes, and for document recovery as part of the process of litigation support. They use sophisticated word processors to put together partnership agreements and wills, and some of the more go-ahead firms now utilise "electronic data interchange" to complete their contracts. The revolutionary view would suggest that this is no more than putting a technological fizz into the old institutions. We still have the same people doing much the same as before - we still think in terms of bits of paper except that we now get machines to move them around instead of doing it by hand.

The lawyers, (and law students) like the compositors before them, fail to appreciate that it is not change that they are facing but the prospect of redundancy. Many of them are blissfully unaware of the fact that at the present moment, right around the world, there are vast sums of money being invested which, if successful, will strike at the very heart of the lawyers' monopoly. There are people, believe it or not, who can envisage a future without lawyers. If "the law" (as opposed to legal documents) can be put into and be manipulated by an "intelligent machine" or "expert system", then when I have a problem I could negotiate directly with the other party whilst being advised about the law and strategy by my computer based system. Should we fail to agree, we can each submit our case "on disc" to the public authority, who will process the matter by computer and automatically issue any relevant orders.

It might all sound rather sci-fi. But then again, the idea of "smart bombs" which travel for hundreds of kilometres, hugging and negotiating rugged terrain, only to enter a specified building through a previously designated entrance, sounded distinctly sci-fi just a few years ago. But as we now know, they worked perfectly well in recent disputes for all that.

The question really is then, are we sufficiently confident that it is all just hype, such that we are safe to ignore it? Many, who were initially very sceptical of all this, now feel that anybody whose livelihood depends on the processing of information, in particular, lawyers and academics, would be very foolhardy to ignore these developments in the hope that they might just go away. They will not - but you might have to.

"Things aren't always what they seem - skimmed milk's often sold for cream"

This is not to say, of course, that all the work which is being done in this area is sound and sensible. There is undoubtedly much money being wasted, and a good number of people are doing things which will, in the fullness of time, amount to very little. On the other hand, there are people who are engaged in work of the utmost importance and who will, before too long, initiate changes of the revolutionary sort. It is then important for those of us who are engaged in the provision of information, to take a serious interest in these developments - indeed our future security depends upon it.

All of the current students in the university systsem should be aware of the issues raised in this area, because only by having an awareness of the issues involved can they make an intelligent appraisal of where the sense and the nonsense of it all lies.

The problem of communication is important. At the moment we have

• computer scientists who do not possess much in the way of knowledge of law and legal theory. Indeed, Professor Alan Tyree has recently said that theoretical issues are a priori and therefore irrelevant.
• We have lawyers who do not have special knowledge of computers and computer science.

This is, in my view, a dangerous situation which must not be allowed to continue - otherwise it will undermine your ability as lawyers / law students to understand what you are doing. As I shall demonstrate, considerable sums of money have in the past been wasted because of the failure to appreciate the need for effective interdisciplinary communication.

As part of the process of mutual education, we need to have some understanding of the dynamics which stand in its way. One of the principle reasons for the difficulties which we will encounter is due to what might be called "the professionalisation of knowledge". Computer scientists and lawyers often use language which is not immediately comprehensible to each other. Nowadays, with the language of the computer scientists, including the neologisms which they use and the notations of symbolic logic which they employ, many would feel that the computer scientists have beaten the lawyers at their own game. As Austin stated, lawyers have often worded their opinions with a studied ambiguity and perplexity to conceal the fact that they often are unclear as to what they are talking about. Now that the computer scientists are using similarly complex language about computerised legal systems, we should be in grave danger if lawyers cannot understand that sort of talk. This could clearly have a major impact on the status, role and autonomy of the legal profession.

It is to one particular aspect of that work that we will devote our attention here - that which involves the use of "legal reasoning". The computer specialists have, perhaps with good reason, been led to believe that legal reasoning is another area of legal work which is amenable to computational techniques. What is needed is a critical evaluation, a testing of the assumptions being employed in the various systems. This is essential to a proper understanding of prospects for development. This part of the discussion is not intended to be hostile to either of the two main disciplines. An understanding of the issues being dealt with by lawyers - how does the law work, and to what extent is the pursuit of "justice" relevant to this process, will be of central importance to anyone who has a thoughtful engagement with legal institutions and dispute resolution. Equally, with the recent spectacular developments within the field of computer science, it would be a bold person who would say, in any area, that computers were not either helpful or relevant. They will inevitably transform our world in many areas of life, and in only a fraction of the time which it took to accomplish major social transformations during the industrial revolution.

If we think of the processes of interpreting statutes and precedents in law we will see a tension between stability and change; between what the lawyers do and what they say they are doing. This will provide the groundwork which will enable us to understand later, why it is that some computer scientists see the cases and statutes as rules, The Imperial College team from London - Zeleznikow at LaTrobe and others claim that we have to look for a "deeper structure" to work on eg Smith from British Columbia, and our own James Popple. Perhaps before we get in too deep, we should discuss some of the basic aspects of the way in which computers work. As Austin has said, if you sometimes understand just the basic principles of certain areas of knowledge, then you do not necessarily need a detailed knowledge to enable you to appreciate what is possible and what is not. We should now look at some of the basic principles relating to the way in which computers operate.

Only if we understand how computers work will we be able to understand why the research into expert systems has developed in the way in which it has. Essentially we are faced with a problem - if the computers are governed by a mechanistic and logical procedure, how could they be used to develop creativity? If lawyers are governed by precedents, how could they initiate legal change? The following discussion is taken from Weizenbaum Computer Power and Human Reason

We do not have time here to go into the ways in which computers work in any great detail. At the most basic level the computer is an electrical device, just like a door bell, which even when "on" is in fact switching itself on and off very fast. The computer, just like this, has active and passive moments - which can be seen like the children's game of musical chairs. When the music plays people change positions - when it stops, they all remain as they are. Imagine bulbs instead of people, some on, some off. When the music plays (the active phase) you can change as many bulbs as you like from on to off, but each one can only be changed once. At each quiet time, the arrangement of lit and unlit bulbs will be different. The active phase can be quite short, because if you have some assistants so that each bulb has its own switcher, then the active phase lasts only as long as the flicking of one switch.

Now imagine each of these bulbs comprising a unit of 2 bulbs. The unit is wired up so that as one bulb in the pair goes on, it switches the other bulb in the pair off. So if a current is transmitted to the off bulb, it goes on, but in doing so switches the other one in the pair off. Hence each pair of bulbs flip-flops between which one is on and off. The wires leading from our flip-flop pair, lead to other flip-flops but the two wires from any one flip flop may go to 2 different flip flops. Now, of course, it is no good if the computer simply keeps reproducing the existing arrangement - its value is that it allows for transformations in the arrangement of these on-off switches. Now we can think either in terms of lights being on or off - of positive or negative charges, or in the notation used in computer programming, of 0s and 1s. The essential thing is that we have a system based on 2 values, and we need a means by which those values can change other than by going along and flicking the switch yourself. The way a computer does this is through the use of GATES. This is the means by which the computer can change a signal from one flip-flop into a different signal to the next one. The simplest example is the

NOT gate: if it receives a 0 it then transmits a 1 and vice versa.

it has one signal and one output.

0 = 1

1 = 0

the AND gate has two inputs and one output. It transmits 1 only when both inputs are 1, ie in our previous example, when the wires running to it are both from lights which are "on".

so 0 0 = 0

0 1 = 0

1 0 = 0

1 1 = 1

the OR gate has two inputs and one output. It transmits 1 whenever either or both inputs are 1 - otherwise it transmits 0

0 0 = 0

0 1 = 1

1 0 = 1

1 1 = 1

If you want to see how a combination of these gates can be used to make up a simple adding machine, look at the example in Weizenbaum. All we need to get from this is some understanding about the basic operations of a computer and to realise that it makes basic but routine transformations to its arrangement of symbols.

Now bearing in mind the active and passive phases which we referred to, we can see that at any time, we can read the arrangement of lights - or values which the computer holds, in such a way that they are meaningful to us. We can perhaps see the analogy with semaphore flags. The computer has an arrangement whereby a number of these flip flops are wired up together to make up a register - the power of the computer comes down to the values which are held in these registers. All of these values are expressed in the binary system of 1s and 0s. A collection of 8 flip flops will make up an 8 bit register which will give us the ability to express all letters and values.

The internal storage of the computer depends on an arrangement of these registers, maybe a million or more. In order for the computer to be able to operate it has to be able to identify each of these registers - each one has to have its own unique address. The computer needs to be able to store these addresses in a central access point - which is of course, itself, a register. A 20-bit register can store 1, 048,576 addresses.

This addressability function of the computer is very important. It explains why full text retrieval is so fast, but not really "full text searching". We can retrieve the occurence of words without really having to search through the full text at all. When we are setting up the data base, we give each occurrence of a word its own address. So in addition to the data base which is the full text, we also have a data base which is the indexed and addressed version of all the occurences of the words. Very often, what appears to be full-text searching, is in fact only index searching - that is every bit as good as if the full text were searched, but is much more efficient, although it increases substantially the actual "text". But text storage is where computers can give us enormous gains.

All we have to appreciate is that the complexity of the computer is made up by operating in a mechanistic fashion to the values which are found in the registers. Weizenbaum has an interesting illustration of the way in which the operation of a computer can be likened to working with a toilet roll and some black and some white pebbles. Thinking issues through at this level of "machine code" is very complicated, so programers have developed a series of higher level languages in which we can express ourselves more readily, and which then depend on other computer languages to translate them through the lower level languages and eventually, by the use of ASSEMBLY languages into this machine code to operate the system. If this simplistic analysis is correct then, it might raise fundamental problems for anyone attempting to get the computer to engage in certain types of activity.

With the emergence of semantic and neural networks, parallel processing and supercomputers, are we witnessing a qualitative, or merely a quantitative change? Can computers make the transition from dealing with data, to reasoning with concepts? Is an "expert system" really expert? Are the computers becoming clever, or like the magician, does their speed help to create illusions? As Weizenbaum has pointed out - "if astrology is nonsense, then computerized astrology is just as surely nonsense".

We should perhaps mention the famous "Turing Test" (named after Alan Turing, an early pioneer of expert systems) which suggests that if a machine can arrive at the same outcome as a human expert, then it is fair to say that the machine engages in expert behaviour. John Searle, a philosopher of language, uses his Chinese Room argument to illustrate the point that this really conceals more than it reveals. Mimic-ing expert behaviour, he argues, is a long way short of engaging meaningfully in expert behaviour. Mimic-ing the rainfall over certain States will not make anyone wet, just as mimic-ing the flow of money through the Australian banks will not make you rich.

We will look at a number of different research projects as a way of illustrating the different techniques and approaches which have been utilised in developing legal expert systems. First, we will focus on the more logical or rule following aspects of law, and show how different the approaches can be whilst working within the same broad paradigm. Here we will look at the ICG's logic programming and Thorne McCarty - prototypes and deformations - an important part of our project which we will come back to deal with in greater detail in the next chapter. Later, we raise the issue of the extent to which legal doctrine can be taken at face value, and shows ways in which we can understand legal cases, yet not be reliant upon the wording of the judgments which are delivered in those cases - A Kowalski and Smith and Deedman's Deep Structure, and Popple's mathematical matrices. We shall also consider how the computational techniques can be developed to enable us to work with multiple or competing meanings within the law - Pam Gray and Skalak and Rissland's development of argumentative strategies. Then we will look briefly at the theory of Ronald Stamper, who argues that our computational structures have to take seriously a theory of meaning (semiotics) if they are to be of any value to us.

As the systems are evolving, there is a growing recognition of the necessity of having the lawyers involved in the construction of the system, rather than have this done through "knowledge engineers". Whilst the aims are more modest - the prospects of success are greater. Instead of expecting the machines to engage in clever thinking, we now only expect them to retain the results of clever thinking. By consulting with an expert, we can see how they would tackle a problem. If we can clearly articulate how an expert would tackle a task, then by embodying this knowledge in a machine-base, we could enable others with less expertise, to perform at a higher level than that at which they would otherwise be able to perform. The analogy might be with an automated office manual. The advantage which a system such as this has over the "office manual" being that much unneeded complexity is hidden from the user until they are able to develop their understanding to enable them to cope with it. In this way, the computer system may well embody expertise, but it does not do the job of an expert.

As well as "delegating expertise", there is also a significant push in the direction of developing these systems as "aids to experts". In the area of conceptual modelling, it is suggested that skilled lawyers could utilise these systems to enable them to better view or articulate the structure of legal argument. Just as designers of cars, or architects, can change a few of the parameters within which they are working, and then view a 3-dimensional model of the amended structure; so too lawyers could structure legal arguments and then see how the alteration of certain assumptions would affect that argumentative structure.

If we can see these last two possibilities as "storage of expert knowledge" and "tools for experts", we can see that both of them involve a significant adjustment of the original idea - that of being able to think like an expert. These last two possibilities suggest that the systems are more likely to reflect the expertise but cannot generate it. The important question is the extent to which (and how) these representations are tied into, or derived from the basic knowledge base.

If we were to come across a number of people who were running along and jumping in the air in an attempt to fly, we could of course join them, and attempt to run faster or jump higher. We could try jumping off a cliff or catapulting ourselves into the air. We could attempt to justify this to ourselves by saying that "if its good enough for them, then its good enough for me," and feel very noble about our sense of solidarity. But we would still not be able to fly. What is needed at this stage, is not more people doing it in different ways, but some people to try and think the issues through. To relate the problem to our existing body of knowledge, both theoretical and practical, and to attempt to gain some insight into the nature of the problem. As Weizenbaum says:

The logic programmers who we will look at in the next chapter have said that there are no serious problems with knowledge acquisition - because the legislation is already written down Clearly, these researchers do not distinguish between the writing (which is the legislation) and the meaning of that writing. However, we are now in a position to examine more clearly what we might mean by "expert system".

Basically, the representations of law are contained within the system as a system of propositions and consequences connected in the system in an IF .. THEN pattern which can be seen as representing rules and conclusions. .

The inference engine uses forward and backward chaining - as the names suggest, this means that one can work

forward from rules to conclusions,

backwards from conclusions to antecedent rules

When an object is supplied by the user, the system can invoke and display all the rules where that object is used. Any rule can be used to start a new problem ie any rule can be a goal.

A PARSER is used to construct a knowledge base from standard text files. The phrases are taken verbatim from target materials, and the parsing heuristics automatically effect the transformation of these names so that questions and explanations can be provided.

Rather than give a technical explanation here of what is involved, let me give you an illustration by one of the first pioneers of such a system. In the mid 60s, Joseph Weizenbaum developed a system called ELIZA. The idea was that the program would "converse" in English. A person would sit at a computer, and as they typed in comments the computer would respond in the manner of a psychotherapist. In fact, the program which constructed the responses was a parser of the same sort which the Datalex system uses. The ELIZA system used a language analyzer and a script. The script is a set of rules which could be given to an actor to enable the actor to improvise around a theme. It sets the level and direction of "responses". Weizenbaum chose a Rogerian psychotherapist for the computer to model, because the technique here was to draw the patient out by reflecting or echoing the patients statements, albeit in slightly different form. .

The "conversation" took the following form:

The patient speaks first - the computer responses are in in capital letters:

Men are all alike

IN WHAT WAY

They're always bugging us about something or other

CAN YOU THINK OF A SPECIFIC EXAMPLE

Well, my boyfriend made me come here

YOUR BOYFRIEND MADE YOU COME HERE

He says I'm depressed much of the time

I AM SORRY TO HEAR YOU ARE DEPRESSED

Its true, I am unhappy

DO YOU THINK COMING HERE WILL HELP YOU NOT TO BE UNHAPPY

I need some help, that much seems certain

WHAT WOULD IT MEAN TO YOU IF YOU GOT SOME HELP

Perhaps I could learn to get along with my mother

TELL ME MORE ABOUT YOUR FAMILY

My mother takes care of me

WHO ELSE IN YOUR FAMILY TAKES CARE OF YOU

My father

YOUR FATHER

You are like my father in some ways

WHAT RESEMBLANCE DO YOU SEE

You are not very aggressive but I think you don't want me to noice that

WHAT MAKES YOU THINK I AM NOT VERY AGGRESSIVE

You don't argue with me

WHY DO YOU THINK I DON'T ARGUE WITH YOU

You are afraid of me

DOES IT PLEASE YOU TO BELIEVE I AM AFRAID OF YOU

My father is afraid of everybody

WHAT ELSE COMES TO MIND WHEN YOU THINK OF YOUR FATHER

Bullies

DOES THAT HAVE ANYTHING TO DO WITH THE FACT THAT YOUR BOYFRIEND MADE YOU COME HERE

The author says that he was shocked by the way in which this was taken up. He intended it to illustrate the effective information processing power of computers, to those who may have had no specialist knowledge of how computers work. Yet within a short while people were talking about the possibility of automatic psychotherapy being provided by computers - not unlike, perhaps, the way in which people now talk about legal advice being provided by computers. Weizenbaum was shocked by the way in which some psychiatrists could have suggested that "the human therapist can be viewed as an information processor.. with a set of decision rules." He felt that even psychiatrists did not understand the basis of their own discipline. He was also shocked at the way in which people became emotionally involved with this process - at times insisting on their right to have their "consultation" with his program - in private - what he described as powerful delusional thinking in otherwise normal people. He also suggests that "language is understood only in contextual frameworks" an important point which we will take up shortly in the discussion on logic programming.

Bench-Capon has laid great emphasis on this need for isomorphism which, he claims, ensures similarity of structure between the knowledge base and source documents (we will look at this in some more detail in the next chapter). Yet, if we look at what he says here, we find that we must treat this claim with considerable scepticism. The documentary sources comprise legislation, delegated legislation, administrative guide-lines and precedents. The text is:

1 translated into computer readable form

2 relevant sections are identified

3 the sections are copied and summarised

4 Entity, Attribute and Value triples are identified

5 Class Hierarchies and Rule Bases are formed from the EAVs

6 Another program takes the Class Hierarchy and Rule Base and puts them through further compilation and translation procedures

To claim that all this "maintains the original structure" is surely very misleading. How does one know what sections are "relevant"? To what extent can a summary or translation maintain the integrity of the original? Of course, many of these researchers were unaware of the extent of the difficulties, because they had not included people with legal expertise as part of the research team. Bench-Capon acknowledges that a surprising factor of the research was that there were no practising or academic lawyers involved. Given the size and length of the project (65 researchers, 30 or more at any given time, working over some 5 years) and the fact that this was a feasibility study to explore the application of knowledge based systems to law - this was clearly inexcusable. This acknowledgment is similar to that made by Sergot and Kowalski, when they say that lack of legal expertise has meant that they have only structured a "layperson's" understanding.

After his earlier work on the TAXMAN programs, McCarty has gone back to try and think through the fundamental difficulties which had arisen in their earlier approach. He said that they had to acknowledge that:

As a result he moved in the direction of developing a theory of legal argument, rather than a theory of correct legal outcomes. The theory is that legal concepts are represented by

This work relates to the psychological study of categorization.

But this sort of theory makes enormous demands upon the knowledge representation language. Whilst the transformation is SYNTACTIC, yet it must correspond to the significant legal SEMANTIC relationships - yet the TAXMAN systems did not have an adequate semantic basis. How do we define a prototype - what is a permissable transformation? Without adequate constraints on these, anything could be transformed into anything.

The answer is, according to McCarty, a "Language for Legal Discourse".

This allows one to represent - states events actions - and modalities (deontic modalities are like the "ought" type attitudes to things) over actions - permissions and obligations. He explains that the relationship between the syntax and the semantics of the system is based on an intuitionist logic or semantics. He also claims that the use of prototypes and deformations is reflective of judicial work, at the appellate level, and suggests that his work is complementary to that of Ashley and Branting (who have articles in the 3rd vol of the conference proceedings). However, he says that their work is more useful for organising large data bases rather than for getting to grips with real insights into legal argument (p188). He suggests finally that the newly emerging field will be "Law and Cognition" - watch this space.

See Skalak and Rissland's article in 3rd conference proceedings. Their focus in this paper is on creating legal argument. Unlike the logic programmers who simply try to attain the "unambiguous meaning of cases and statutes" these researchers explicitly focus on the adversarial aspects of legal argument. This involves mapping not only the rule, but whether one is for or against it, and whethere there are precedents or alternative paths to the desired result. The whole thing has to be construed as a decision tree - not unlike Pam Gray in her recent work on legal argumentation. They look to

Basically one has to think of lines of argument in support of, or against, a rule and which can depicted in the form of a decision tree. Their top-down approach means that one first selects the line of argument and whether one wishes to confirm or to discredit it, before getting to the level of cases. The computer is then a control structure for developing argumentative moves.

Andrzej Kowalski deals with "case based reasoning and the deep structure approach to knowledge representation" also in the 3rd conference proceedings. Unlike many of the others who see law as a coherent body of rules or principles, he takes the view that legal prinicples are often expressed in contradictory pairs, and that as law is an adversarial system, not an impartial seeking of the truth. In his view, theory postulates that judges use deep structure fact patterns to decide cases. These are sub-conscious, like the rules of grammar, and go beneath doctrinal law to the facts of cases to establish a basis on which the computer can reason. Case based reasoning therefore has to do with the dynamic control of cases, so that the entry of new cases will effect legal outcomes. This will require cases to be ranked and ordered.

The Malicious Prosecution Consultant system began as rule-based system, Rules were thought to be basic, he suggests, because they are fundamental units of knowledge representation in law. To move from a rule based system to a case based reasoning system, the specific rules are deleted, but their "ghosts" remain although they do not directly effect the outcomes. The more general rules are retained.Then rules for controlling the retrieval and the display of cases are then added.

Each case then has a profile - there are 8 result slots which represent the courts finding on particular factual ATTRIBUTES - this is the replacement for the specific rules of the system. Almost as if the Factual Attributes stand in place of Rules. Each Attribute is then given a VALUE. It is worth noting that when one talks of "value" in this context, one is not speaking of value in the more generally understood sense, but of whether that attribute is present or not - value is limited to YES / NO / and maybe DON"T KNOW.

One can then search the database for the identification of attributes, and for the weighting of attributes of a particular type or configuration. The expert contribution to an expert system is to provide the general structure within which problems may be solved. This structure is inevitably founded on a rule-based model of the domain, whether or not the rules are explicitly stated.

This system has been developed by Graham Greenleaf, Andrew Mowbray and Alan Tyree. It is a system which combines a number of features - free text retrieval, expert system and hypertext - most systems developers focus on one or other and there are few integrated systems. Softlaw's STATUTE also uses hypertext within an expert system environment. Datalex attempts an integrated approach within the context of privacy law. It has three "engines" with a common interface:

· Expert system shell, (inference engine)

· Free text Retrieval

· Hypertext

Each system needs its own material, or representations (knowledge base) on which it can work -

· For the expert system it is a rule base:

· Free text retrieval uses a concordance

· Hypertext uses a hypernet

each is conceptually distinct from the texts, or sources which we would normally regard as "the legal material". With expert systems, the transformation from "the materials" to a "knowledge base" is always problematic. Some account of knowledge representation is needed. Some other systems may more easily be seen as data + work with "information representations". Privacy law is of more recent origins, not unlike the work on the British Nationality Act, which requires a range of materials, such as statutes, cases and "
"guidelines" which might be of an administrative, or more formal nature.

This is the non-linerar text presentation or text navigation - rather like snakes and ladders between the content of different text files. In the materials which have been developed in our contract law teaching, we have linked them together in the following manner. COLIN is the "Contract OnLine Information Network":

The intention is that the one system will provide a teaching tool for university lecturers, a learning tool for university students, and a research tool for solicitors, barristers and others interested in finding out about the law. University teachers and students will probably enter the system by way of the textfile overview, and then into the Code and lectures. Legal researchers can go straight into the case or legislation files at the appropriate level.

The material is cross related - cases use statutes, the meaning of statutes depend on the cases. Various points in the text can be marked as nodes which may either be listed separately, or linked to other relevant text. Lists of chapter or topic headings can be produced with each item on the list linked to the respective portions of text. The text within the system can be card-based or continuous ie each link can be to a single portion of text, or to a place within a large textfile. The text to which a node is linked may be jump-to, or pop-up. The links may be one to many or many to one - Different strategies can be employed to limit the spaghetti effect, and possible confusion which may result from being lost in hyperspace, by routing through a cross reference index. Automatic marking up of text can be important in terms of the commercial viability of these systems. There is probably no need to say much more about free text retrieval at this stage, other than that in this system it works on a 5 level concordance - chapter, article, section, paragraph and word. It uses synonyms, and searches for exactness rather than nearness. Retrieved dcouments are not ranked by likely relevance. Boolean and proximity connectors may be used and there can be Boolean connector links to previous search request. Retrieved terms can take one directly to relevant material using the hypertext links.

The major factor which distinguishes the researchers which I will deal with in this section, and those of the previous section, is not so much the computational aspects of their system, but their overall relationship to, and understanding of, the application domain. The logic programmers clearly had a certain mind-set or paradigm within which they wanted to work, and were looking for an information system to fit it. The researchers in this section are much more aware of the environment within which their computers have to operate. Even in their earlier work they pointed out for example, that whilst classical logic serves the mathematician pretty well it does so by simplifying problems - "by putting the observer out of the picture". They also indicated early in their work, their awareness of the complementary relationship between statute law and case law in the sense that each provides a commentary on the other. [Jacob, 84]. They point to the creative aspects of categorisation - something which I see as an essential aspect of rule creation and application. They also acknowledged that deductive reasoning is entirely peripheral to the work of the judges - all of which is in stark contrast to the claims of the previous theorists we have looked at.

Semiotics - a serious engagement with theory

The most significant challenge provided from this perspective is upon the idea that words themselves can have meaning without reference to the social framework within which that meaning is developed [Stamper, 87]. This is referred to as a naive assumption which begs all the important questions:

The view taken here, rightly emphasises that legal disputation is not a case of applying settled meanings, but of disputation and negotiation as to what those meanings should be. This is in accordance with the views of the legal theorist John Austin who suggested that rules are merely the shorthand way we try to explain the point at which a settlement to a dispute is either agreed or imposed. Rules may represent the point at which we end up, but cannot represent the means by which we get there or the point at which we start. Wherever there is a dispute, we have by definition alternative formulations of a rule - and the same applies for any other rule which we put forward in an attempt to resolve that dispute. Rules must always look beyond themselves for further and better particulars [Moles, 87].

Rules may be the manifestation of a consensus, but even then they are only an abstract and symbolic shorthand for more complex reasoning - in that event they cannot be the explanation for that consensus. It helps us to appreciate a neglected aspect of Austin's work - unlike Hart who explained rules in terms of prior rules, Austin explained rules in terms of social attitudes and pressures [Moles, 87]. The logic programmers, as we will see, have been misled by the Hartian, oversimplified legal philosophy. Whilst Leith is clearly correct in his view that the logic programmers he looked at were engaged in a more simplified project than that of Stamper's he did not really spell out in what way Stamper was trying to adapt to complexity. This is something which I would like to look at in the remainder of this section.

It follows from what we have just said, that the recognition of the abstract and symbolic nature of language and of the need for a theory of signs (semiotics) is important, in that it has correspondence with sociological insights whilst at the same time having theoretical depth which can serve to link jurisprudence with the theoretical contributions of other disciplines. For some years now, Stamper has been critical of the theorists represented in the previous section . He argued then, as I have done now, that they assume too much and work within a formalism as if nothing existed beyond it, and that they take as primitive concepts, those which need to be explained.

What is essential to any prospect of success in this area is to have a better understanding of the relationship between any logical formalism and the "real" world. To understand this one needs to look at the way in which these classical logics define symbols which intrinsically have no meaning at all - if we see words as symbols then there is a stark contrast between this view and that which seeks the unambiguous meaning of a word.

The problem can, of course, be avoided by:

The price to be paid is in the avoidance of responsibility, and hiding behind a false and misleading technicalism. If the role of the law is to establish boundaries and maintain them, even to allow them to move gradually in a controlled manner, then to assume that all kinds of boundaries are fixed, and fixed in an objective way, independently of any human agency, is to evade the central issues with which the law is concerned. Stamper uses the metaphors of bottling and transmitting knowledge as compared to the social construction of reality to which I have already referred [Stamper, 91] and in support of which I have argued in detail elsewhere [Moles, 87].

Stamper clearly acknowledges the consequences of this line of thought. If we adopt it, then certain implications would seem to follow. First, we regard reality as subjective and constructed by users within their informal, culture-based information systems, knowledge of which is transmitted through abstract signs whose meaning can only be recognised by appreciating the purpose and context within which those abstractions are formulated. To speak of "meaning" raises the issue of semantics which many writers prefer to reduce to a problem within the technical platform. This may not be unreasonable, where boundaries reflect a well established consensus - but it fails when that consensus breaks down and negotiations are needed to re-draw or re-establish the boundaries. This is why we have to see the resolution of legal disputes as being in the nature of performative utterances - as creating the rule (the boundary). For use, a sign (or rule) must always have an intention imputed to it by its creator and interpreter and this can only be understood in its context -

This emphasises the view that "context" means not just the relationship to other texts, but an understanding of the social framework within which those texts have meaning. It is this frame of reference, by its very nature, which is unexpressed or incompletely articulated within the texts themselves, yet which gives shape and meaning to them. "The frame of reference determines what you see and what remains invisible", which relates to what we were saying earlier about "organising frameworks".

The analogy which is most appropriate to law is Stamper's use of the idea of information as "giving form" to something, as a potter informs the clay. The common view, especially prevalent in discussions concerning the law is that "the system" takes raw data and converts it into information. The information plumbing metaphor has no room for the people who give meaning and intention to the signs nor about the relationships between people which are created, sustained and exploited through signs. Technical questions, he claims, are secondary to the organisational needs and dynamics.

We now have to allow for, rather than ignore, those matters which Fleck and Polanyi speak of - the importance of informal tacit knowledge. Indeed, says Stamper, this informal system is not something to be minimised or dismissed - it matters most. It is the concept of the norm which, he suggests, can be the link between the formal and informal systems. This requires us to map communication of the organisation in terms of its norms and responsible agents. The rules, or norms, being social constructs, must always be constructed by someone for some purpose. It follows then that there is no knowledge without a knower. The knowledge then, to be meaningful, has to be linked to those whose knowledge it is. This requires us to tie every item of knowledge to the agent responsible for it. Truth, then, is something which agents have to decide upon and the consequences for which they have to accept responsibility. Responsibility, here, plays the same role as truth in classical logic. Truth is not a primitive concept but a derived one, which is explained in terms of agreement among agents [Stamper, 92].

If we accept this, then it appears that we may be able to map the relationships involved in this understanding. To do so we need to appreciate the ontological dependencies (the way in which one type of behaviour depends upon another), and the ontological antecedents (the way in which an invariant cannot be realised without its antecedent). If we can bring into the picture the issue of time constraints (when each of these operative factors begins and ends) and complications such as group agents and an agent with many parts or roles to play, then we are approaching a degree of complexity which may best be mapped by what Stamper calls an ontology chart.

If this represents an accurate portrayal of this approach, then it seems to me, as a legal theorist, that it presents us with much greater potential for a way forward. In essence, it appears that this perspective requires us to model the social understanding of a group, rather than to think that we can crank anything of significance out of a text - without this. Bringing this dimension - the aspects of conceptual modelling - into the picture, is in my view essential from a theoretical perspective. I would want to argue that theory and practice are not in tension or conflict - merely two different ways of looking at the same thing. Better theories will give us fresh insights and maybe raise issues which have to be seriously tackled if progress is to be made. Tyree has recently suggested that theoretical objections are a priori and irrelevant [Tyree, 92]. His only concern is whether or not the system works. The evidence so far, from those pursuing the logic programming paradigm, is that their approach does not work within the legal domain - and they are no closer now to understanding why that is so than they were several years ago. From the perspective of legal theory, and epistemology in a wider sense, I would offer the opinion that Stamper’s approach has real potential which is lacking in that of the logic programmers. Whether it provides a strategy which is implementable, is something which I am not, as yet, qualified to judge - but I hope to be in a position to report further on that very soon.

The full text of Definition and Rule in Legal Theory is now up on the system. Chap 8