[Home] [Help] [Databases] [WorldLII] [Feedback]
	Murdoch University Electronic Journal of Law

LEGAL EXPERT SYSTEMS - A HUMANISTIC CRITIQUE OF MECHANICAL LEGAL INFERENCE

"Suppose I am in a closed room and that people are passing in to me a  series of cards written in Chinese, a language of which I have no  knowledge; but I do possess rules for correlating one set of  squiggles with another set of squiggles so that when I pass the  appropriate card back out of the room it will look to a Chinese  observer as if I am a genuine user of the Chinese language.  But I am  not; I simply do not understand Chinese; those squiggles remain just  squiggles to me." [*]

   1: INTRODUCTION

 Computerisation of the legal office is an ongoing process.  The range  of non-legal applications now in common use include word processing,  accounting, time costing, communication and administration systems.  [1]  More recently it has been demonstrated that computers can be  used as research tools, particularly in the retrieval of primary  legal materials.  Prominent examples include the LEXIS, SCALE and  INFO1 databases, now familiar to many practitioners. [2]  Some moves  have also been made towards "conceptual"  text retrieval systems. [3]

 Flushed with successes in projects such as DENDRAL, [4] PROSPECTOR,  [5], and MYCIN, [6] computer scientists have now turned to law in  order that they might "widen their range of conquests". [7]  The  interaction between computers and the law has now spawned a large and  disparate discipline, boasting eight centres for Law and Informatics  in Europe, as well as growing numbers of similar centres in North  America, Japan and Australia.  The "resource" of expert legal  knowledge, "often transitory, even volatile in nature" is seen worthy  of nurture and preservation.  The use of legal expert systems is seen  capable of preserving indefinitely and placing at the disposal of  others the wealth of legal knowledge and expertise. [8]  The idea is  not new, being anticipated by writers such as Loevinger [9] and Mehl  [10] as early as 1949.

 Yet lawyers have generally greeted "legal expert systems" - seen by  some as the natural progression in the use of computers - with  apathy, ignorance or resistance. [11]  This article argues that such  opposition is justified when proper regard is had to the implications  arising from the computational foundation for such systems. 

 It is necessary for the following analysis to clearly distinguish two  fundamentally distinct classes of computer applications to law:  decision support systems, and expert systems. [12]  "Decision support  systems" are powerful research tools or "intelligent assistants"  designed to support decisions taken and advice given by human  experts.  "Legal expert systems" are designed to make decisions and  provide advice as would a human expert.  Richard Susskind, a British  researcher whose work [13] constitutes the major theoretical  grounding of legal expert systems, states:

 "Expert systems are computer programs that have been constructed (with  the assistance of human experts) in such a way that they are capable  of functioning at the standard of (and sometimes even at a higher  standard than) human experts in given fields . . . that embody a  depth and richness of knowledge that permit them to perform at the  level of an expert."[14]

 Legal expert systems are a type of knowledge based technology.  With  the explosion of applications, "expert system" is quickly becoming an  imprecise term. [15]  The definition used by Feigenbaum will be  acceptable for the type of systems examined in this article:

 "An intelligent computer program that uses knowledge and inference  procedures to solve problems that are difficult enough to require  significant human expertise for their solution.  Knowledge necessary  to perform at such a level, plus the inference procedures used, can  be thought of as a model of the expertise of the best practitioners  of the field."[16]

 In terms of programming technology, the knowledge based approach has  been described as an "evolutionary change with revolutionary  consequences", [17] replacing the tradition of

 data + algorithm = program

 with a new architecture centred around a "knowledge base" and an  "inference engine" so that:

 knowledge + inference = expert system.

 In fact, there are four essential components to a fully functional  expert system:   1.  the knowledge acquisition module;  2.  the knowledge base;  3.  the inference engine; and  4.  the user interface.

 Knowledge acquisition is the process of extracting knowledge from  experts.  Given the difficulty involved in having experts articulate  their "intuition" in terms of a systematic process of reasoning, this  aspect is regarded as the main "bottleneck" [18] in expert systems  development.  The knowledge base stores information about the subject  domain.  However, this goes further than a passive collection of  records in a database.  Rather it contains symbolic representations  of experts' knowledge, including definitions of domain terms,  interconnections of component entities, and cause-effect  relationships between these components.  In legal expert systems this  usually consists of formalised legal rules obtained from primary and  secondary sources of law.  Another layer of rules may also be  obtained from less formal knowledge not found in published  literature, [19] such as "practitioner's hand books and internal  memoranda within legal practices". [20]  These heuristics [21] add  "experiential" to "academic" knowledge. [22]  

 An inference engine consists of search and reasoning procedures to  enable the system to find solutions, and, if necessary, provide  justifications for its answers.  The nature of this inference process  is described in detail in Section 2.  The user interface is critical  to the commercial success of expert systems, particularly in the  legal field, to enable lawyers with little or no expertise in  programming, to gain access to the encoded knowledge.  Typically this  is in the form of prompting for information, and asking questions  with "yes", "no" and "unknown" responses. 

 Artificial intelligence, the foundation for legal expert systems, has  run up against both practical and theoretical difficulties.  While  computers can beat the average human at "clever" tasks such as  playing chess, they are "impossibly stupid" over tasks taken for  granted such as speaking a language or walking across a room. [23]   This casts doubt on whether many human activities do, as some  artificial intelligence researchers suggest, consist of suppressed  computational algorithms.  There has been severe criticism by those  who claim that knowledge, by its very nature, is not amenable to  representation on a computer, [24] or that they achieve no more than  simple competency. [25]  Early misconceptions about the ease with  which powerful and knowledgeable systems could be built for use by  relative novices have given way to concern about real problems of  knowledge elicitation and knowledge modelling. [26]  Some opponents  are convinced that the claims of artificial intelligence are  exaggerated and their objectives unreachable. [27]

 Section 2 examines the nature of the inference engine, and suggests  that its deductive procedures rest in pattern matching routines.  It  also explores issues of "fuzzy" and "deontic" logic.  Section 3  explores the implications for knowledge representation, and questions  whether devices such as "semantic networks", "frames" and "case based  reasoning" are anything more than elaborate pattern matching  constructs.  Section 4 demonstrates that the need to be amenable to a  deductive inference engine involves unacceptable distortion of law  both at a practical and theoretical level.  Section 5 argues that  legal reasoning necessarily involves resort to social context and  purpose, which is not tractable within current technology.  Section 6  suggests that researchers ought to abandon legal expert systems, and  instead concentrate on computerising more mechanical tasks such as  legal retrieval and litigation support.  A summary and conclusion is  contained in Section 7.

   2: PATTERN MATCHING AS THE CORE OF AN EXPERT SYSTEM

 The core of any expert legal system is its inference engine.  This  Section investigates the nature of computer inference at a basic  level, and argues that it is little more than a pattern matching  exercise.  It is also argued that more sophisticated approaches, such  as fuzzy logic, and deontic logic, are no more than extensions of the  same principles.

 2.1  The Nature of Computer Inference

 Computer inference is undertaken by a simple strategy known as modus  ponens.  This means that the following syllogism is assumed to be  correct:

 A is true (fact)  If A is true then B is true (rule)  \ B is true (conclusion)

 Computer deduction is obtained by conditioning the consecutive  execution of instructions on matching, or failing to match, values in  storage registers.  The identical syllogism is obtained by a computer  using a routine in the following terms:

 1.  check the value of register X1  2.  compare the value of X1 to a value in register A  3.  if X1 = A then:  4.  change the value of register X2 to the value in register B

 The computer is conditioned by the value placed in register X1,  either by the user or by satisfaction of some prior rule.  A is taken  to be true if the value of X1 is equal to a particular value in  register A, representing some real world condition.  The rule "if A  then B" is contained implicitly within the structure of the routine  by conditioning step 47 on the satisfaction of the condition X1 = A  (i.e. A is true).  The modus ponens is completed by step 47 which  alters another register to equal a value B, thereby asserting that B  is true.

 A programmer in BASIC or PASCAL, for instance, has some relationship  in mind between the data supplied to the program and the output to be  produced by computation.  The input data are stored in the machine's  memory, and the programmer's task is to devise a sequence of  instructions to manipulate the data in accordance with the  relationships she envisages.  In such a case the inference engine  constitutes the implicit algorithm contained within the sequence of  instructions.  Examples of such algorithmic knowledge based systems  include Hellawell's CORPTAX, [28] CHOOSE, [29] and SEARCH, [30] all  implemented in BASIC. 

 2.2  Logic Programming

 Normal programming can at best maintain logical relationships  implicitly within the program's structure.  A number of researchers  are now involved in logic programming, [31] which has been seen as  the real technical breakthrough in this field.  Some have extended  logic programming to the point of writing expert systems by means of  another logic based application, such as DARWIN. [32]  Logic  programming allows the programmer to specify logical relationships,  not in terms of sequential instructions, but in terms of some  symbolic language. [33]  It is then up to the machine to compile the  set of sequential instructions to maintain the desired relationship.   A logic programming system can be regarded as a kind of rule-based  system where the inference engine becomes a "mechanical theorem  prover", [34] a machine for answering questions of the form:

 Do axioms (A0 . . . An) logically imply B ?

 The claimed advantages of rule-based logic systems over conventional  programs are perspicuity and modularity. [35]  Perspicuity is  obtained by separating the rules (the knowledge base) from the  logical operators (the inference engine).  This has important  implications for system maintenance and debugging.  Modularity exists  since the knowledge is split into small and independent rules. 

 Legal rules, written in symbolic language, are manipulated through a  process of "forward" and "backward chaining".  A set of IF-THEN  rules, constituting a "search space", [36] are compared against a set  of facts to reach a logical conclusion. [37]  In an expert system  forward chaining simply involves matching the IF conditions to the  facts, according to a predetermined order, which under the rules,  dictate a conclusion. [38]  Susskind describes this as a "control  strategy" which "triggers" and "fires" the rules.  [39]

 Backward chaining starts with the legal conclusion and searches for  justifying antecedents in the knowledge base.  In terms of  programming this technique is more difficult since the search of the  knowledge base is not along a single "path" but involves  identification of all possible rules leading to the required  conclusion. [40]  In essence, it matches THEN variables with their IF  antecedents and compiles a list of the paths thus generated.

 A "goal driven" expert system predetermines a conclusion and  identifies the legal arguments and reasoning that can be used in  support of that conclusion. [41]  In logic programming terms this is  no more than backward chaining across the search space.  These  processes of forward and backward chaining form the core of expert  systems inference procedure. 

 One notable feature of logic programming is the Horn clause, seen as  a suitable extension to the "simple" predicate logic already  outlined.  It is of the form:

 A if B0 and . . . Bn where n >= 0.

 which consists of a single conclusion, A, and any number of  conditions (B0, B1, . . .Bn).  For example, the Horn clause:

 X is the father of Y if X is a parent of Y and X is male

 is a Horn clause with one conclusion and two conditions.  Factual  premises, such as "X is male" can be expressed as a Horn clause with  one conclusion and no conditions.  The significance of such a clause  is that symbolic logic is not limited to IF-THEN statements, but may  be extended to IF-AND-NOT-THEN statements.  In terms of actual  programming, however, the Horn clause is implemented as a bundle of  IF-THEN assertions; each condition being checked separately for a  pattern match, and the routine halting on failure to match.    The pattern matching approach of logic programming is  not relaxed but in fact tightened by the use of "integrity  constraints", such as IF-THEN-ELSE structures, to close off the  potential for negation by failure [42] and counterfactual conditional  [43] difficulties.

 Constructed on the basis of Horn clauses, PROLOG and variations have  been the platform for most logic programming projects, both in logic  and procedure.  APES, [44] implemented in PROLOG, is one widely used  augmentation. [45]  Using PROLOG as a symbolic logic structure  involves rendering the domain knowledge in terms of Horn clauses,  rewriting them in PROLOG syntax, and then executing the result as a  program.  PROLOG may itself provide a procedural basis for expert  system platforms.  Horn clauses may be backward chained as a  procedure, working from conclusions to conditions and, as a sub-task,  pattern matching each against its knowledge base, or user input.  The  program statements can thereby mix conditions which express legal  rules with procedures to prompt the user for additional information.   An example is Schlobohm's system to determine "constructive ownership  of stock" under United States revenue laws. [46]  The LEX [47]  project is a more sophisticated application of the same principles.

 2.3  Fuzzy Logic

 Fuzzy logic is an attempt to escape the perceived inadequacy of  binary logic. [48]  Zadeh introduced the concept of the fuzzy set  [49] to provide a formal way of speaking about imprecise concepts,  such as "large" and "small".  Rather than requiring precise values to  be attached to particular characteristics, a spectrum of values,  broken into categories, is used to match concepts, analogous to  concepts in cognitive psychology. [50]  The object of fuzzy logic is  to convert continuous measurements into approximate discrete values.   For example, a rule of the form:

 A PERSON IS A MINOR IF UNDER 18 YEARS OF AGE

 can be rendered by the following simple routine:

 1.  check register AGE  2.  if AGE < 18 then:  3.  set register PERSON to value 123

 where 123 represents "minor".  The spectrum of values "less than 18"  is the fuzzy category.  On a more complex level, matching can take  place not only between ranges of values, but fuzzy sets.  In binary  logic, two concepts will be identical if and only if their membership  functions, that is, their defining characteristics, exactly coincide.   For instance, if F is a class of subsets of X, a set of  characteristics defining legal concepts, then for Y and Z:

 Y is identical to Z iff fY(x) = fZ(x) for all x

 Rather than matching "identical" sets, fuzzy logic matches "closely  identical" or "sufficiently close" sets. [51]  To ascertain  "closeness", a probabilistic metric is constructed.  For example

 D(Y,Z) = Integral [{fy(x) - fz(x)}^2.p(x).dx]

 where p(x) is some probability distribution on X.  D(Y,Z) is  therefore a metric that depends on the choice of p(x).  Using these  definitions, one can test "closely identical" by inferring that:

 Y is identical to Z iff (1 - d(Y,Z)) >= d

 where d is some arbitrary threshold which can itself be used to  trigger the operation of a rule. 

 "Fuzzy logic" is therefore one way of rendering continuous or  approximate concepts into terms amenable to computer deduction.  It  is, however, no more than an extension to logic programming  techniques.  Critics suggest that fuzzy logic is no more than  oversophistication of arbitrary approximation; that its appearance of  precision is spurious, and that its philosophical basis is uncertain  when applied to legal concepts. [52]

   2.4  Deontic Logic

 In legal expert systems the nature of law as a normative system, [53]  has given rise to a perceived necessity for incorporation of deontic  logic. [54]  Whereas traditional and classical logics provide formal  canons for reasoning with empirical statements that have truth value,  deontic logics provide standards for reasoning with statements which  lack truth value [55] in the sense that they describe norms or  imperatives.  They cannot be characterised as true or false or  logically related to each other or to statements of fact. [56]   McCarty has consistently argued for "intuitionistic" rather than  classical logic as the basis for representing legal concepts.  He  sets out some theoretical suggestions, as yet unimplemented, for the  semantics of normative concepts in legal expert systems. [57]   General foundations were laid by von Wright, [58] who developed a  system of logic based on possibility and necessity.  According to  McCarty, "permission" exists in the union of all states and substates  in which an action is necessarily true given the conditions of these  states.  This forms the "Grand Permitted Set" or a boundary condition  for legality. [59]  "Obligation" exists in the intersection of these  sets. [60]  In programming terms, "permission" entails backward  chaining from the proposed action to all the states of the world.   "Obligation" then is moving forward from all states of the world to  find a common action. 

 McCarty designed a language called LLD which allegedly possessed  distinct advantages for legal applications in its use of action terms  and deontic language. [61]  However, his unimplemented proposal is  problematic.  LLD attempted to represent law in count terms, mass  terms, states, events, actions, permissions and obligations.   However, even McCarty admits that LLD failed to represent purpose,  intention, knowledge and belief. [62]  Jones demonstrates that a  striking feature of McCarty's theorem is that an obligation to act  did not logically imply its permission.  In particular he  demonstrates that under McCarty's analysis, one could logically  derive "permission to poison the King from an obligation to serve  him". [63]  The only difference from logic programming in the  suggested implementation of LLD lies in the use of fuzzy categories.   That it does not stray too far from traditional logic programming is  obvious since LLD is constructed almost entirely from Horn clauses.  [64]  McCarty therefore fails to tackle more difficult and  fundamentally philosophical problems in deontic logic. [65]

 Stamper's LEGOL [66] project proposed a number of extensions to  enable the system to handle concepts such as purpose, right, duty,  judgment, privilege, and liability; [67] yet these were never  implemented. [68]  His latest project, NORMA, [69] has the object of  relating all formalised symbols directly to the notions of agent,  intention, and behaviour.  However, it is doubtful whether this goal  can be achieved, given that his languages are based in typical  control structures such as sequencing of rules, if-then branches, and  iteration, [70] hence easily rewritten as a logic program. [71]   Sergot suggests that both Stamper's work and McCarty's LLD have  simply taken standard semantics of logical formalism and presented  their own variant. [72]

 Deontic logic may in any case be non-computational.  Since the limit  to current technology ultimately lies in the mechanistic linking of  discrete relationships, the modelling of any "normative" aspect of  law will not by its very nature be amenable to computer processing: 

 "For there is not much sense in asking how . . . by having 255 in  register 1234567 licences coming to have the number 128 in register  450254925." [73]

 Perhaps in light of this limit to technology, both the Oxford Project  [74] and the Imperial College Group [75] have avoided deontic logic.   Susskind reduced deontic logic to predicate logic by treating the  normative aspect of law as merely linguistic. [76]  Deontic labels  were attached to different varieties of mechanical cause [77] and  effect. [78]  Normative statements were simply rewritten into  declarative symbolic language. [79]

 The important implication from this work on deontic logic is  recognition of the error in equating "logic" as understood by a  computer with "logic" as understood in wider contexts. [80] In  particular, reference can be had to MacCormick's distinction between  "formal" and "everyday" logic, with the latter being based in common  sense. [81]  Researchers such as Stamper appear to be aware of such  difficulties, but find themselves constrained by the existing  technology.  Whether legal reasoning can be computational is  addressed in Section 5.

   3:  KNOWLEDGE REPRESENTATION AND THE PROBLEM OF CLASSIFICATION

 The previous Section demonstrated that the process of computer  inference was limited to an elaborate process of pattern matching.   This Section investigates the implications for knowledge  representation;  in particular, that it is necessary for knowledge to  be represented in terms of IF-THEN rules.  It is also argued that  more "sophisticated" representation techniques, such as "semantic  networks" are no more than elaborations of this basic structure. 

 3.1  Pattern Matching and Open Texture

 To be implementable, the knowledge base must be structured so as to  be amenable to deductive inference procedures.  In theory, it is  possible to use any form of symbolic logic as the representational  formalism as long as it is appropriate to a deductive inference  engine.  This condition requires that the knowledge base must be in  the form of pattern matching rules.  In logic programming,  computation is deduction, and the task of the programmer is therefore  to pose a problem suitable for a deductive process. [82]

 The major difficulty encountered is what broadly may be termed  "semantic indeterminacy". [83]  Not all legal rules are appropriate  for application in all situations. [84]  Legal expert systems have  been acknowledged to be only capable of solving problems referred to  as "clear cases of the expert domain". [85]  Yet what is a clear  case?  The Oxford Project defined a clear or "easy" case as one  easily solved by an expert, yet hopelessly difficult for non-experts.  [86]  Gardner [87] drew the distinction between hard and easy cases  by describing the latter as situations whose verdict would not be  disputed by knowledgable and rational lawyers, whereas they may  rationally disagree as to the former. 

 The real answer for legal expert systems lies in the nature of the  computation process.  When presented with the facts of a case, the  expert system must decide whether or not a rule applies.  Since "fact  situations do not await us neatly labelled, creased, and folded" [88]  the difficulty lies in subsuming particular instances under a general  rule. [89] A "hard case" is therefore one where the system fails to  match the appropriate pattern, thereby preventing a rule from firing.   As computer logic relies on pattern matching, knowledge  representation necessarily must encounter problems of classification.  [90]   What is "ultimately beyond the grasp of a computer," states  Detmold, "is not complexity, but particulars". [91]  This difficulty  is often termed "open texture".  The notion of open texture is  obtained from Hart's analysis.  In the now infamous regulation:

 NO VEHICLES ARE PERMITTED IN THE PARK

 the open-textured term here is vehicle.  The difficulty in terms of  legal expert systems is how the program can classify an object as  being a "vehicle" falling under the rule.  Hart suggests that general  words like "vehicle" must have a set of standard instances in which  no doubts are felt about its application.  There must be a "core of  settled meaning", but there will be, as well a "penumbra of debatable  cases". [92]

 In terms of computation, a case is within the core of settled  meaning, and is classified as "easy" where there is a matching  pattern in the knowledge base.  Cases in the penumbra of doubt are  hard, since they cannot be classified by the system.  The difficulty  for legal expert systems, then, is to build a system for  classification, so that the pattern matching process can take place.

 Skalak [93] suggests there are three theoretical models of  classification:  - the classical model  - the probabilistic model  - the exemplar model

 All three models have been extensively used in expert systems  technology.  The following analysis demonstrates that the first two  models have little to distinguish them in practical effect, and  together constitute an inadequate response to the problem of  open  texture.  The exemplar model has been used as justification for case  based reasoning approaches, but the term has been misused, and such  cases are argued instead to fall into the "probabilistic" model.  The  exemplar model is returned to in Section 6, where it is used as the  basis for suggested development of computer applications to law.

 3.2  The Classical Model

 In the classical model, a concept is defined by necessary and  sufficient conditions.  Hafner [94] suggests that these conditions  can be formally represented by knowledge structures involving  decision rule hierarchies, taxonomic hierarchies, or role structures.   Decision rule hierarchies specify the conditions under which a  concept is true or false.  "Vehicle" may be defined by a set of  characteristics such as "four wheels", "engine" and so on.  In  programming terms, this means that the IF antecedents are themselves  THEN consequents based on sets of prior conditions which constitute  the "definition" of a term.  This quickly builds into a "decision  tree" structure. [95]  Statute law, particularly statutory  definitions, are seen particularly suitable for rendering into what  amounts to typical Horn clauses. [96]  A prominent example of this  technique is the modelling of the British Nationality Act 1981. [97]   Others include the United Kingdom supplementary benefits legislation,  [98] and STATUTE, now used by some Australian government departments.  [99]

 Taxonomic hierarchies define sub-types of concepts, placing different  objects into groups and sub-groups.  For instance, the class  "vehicle" may have among its sub-classes "car", which in turn may be  further sub-classed into "Toyota", "sedan" and specific instances  based on model types, years, and so on. [100]  Any taxonomic  hierarchy can, however, be represented in terms of a chain of  decision rule hierarchies, or IF-THEN rules.  Role structures  are  modelled in "frames", and "semantic networks".  A semantic network  [101] is a collection of objects called "nodes".  These are connected  by "links". [102]  Typical links include "is a" links to represent  class-instance relationships and "has a" links to represent  part-subpart relationships.  Interconnected, these may quickly build  into a complex web of relationships.  A frame [103] is a subset of a  semantic network, being a representation of a single object with a  set of "slots" for the value of each property associated with the  object.  All links in both semantic nets and frames are, however,  functionally equivalent to taxonomic hierarchies.  Hayes demonstrates  that both semantic networks and frames are no more than elaborate  logic programs, and concludes that they hold no new insights. [104]   The semantic network may be forward and backward chained as a set of  logical rules, just as would a rendering of a set of Horn clauses in  PROLOG. [105]      For instance, in McCarty's TAXMAN project [106] the domain was  modelled in terms of objects such as corporations, individuals,  stocks, shares, transactions &c.  Each object is described by a  "template", being a collection of the object's properties, such as  name, address, size, and value.  These properties are then linked and  indexed.  Each "bundle of assertions" constitutes the object's  "frame". [107]  These structures are aimed at answering questions  such as:

 "Does the taxpayer and her family have a controlling interest in the  stock of a company which is a partner in a partnership which owns an  interest in XYZ Ltd?" [108]

 In TAXMAN 2 McCarty proposed a more elaborate semantic net based on a  "prototype-plus-deformation" model.  Essentially this sets one frame  as being the default for each class of object, with incremental  modifications to slots based upon fuzzy categories.  Unfortunately  the concept was never implemented.   As a result, McCarty offer no  solution to algorithmic issues such as how to choose, index, and  search the space of prototypes, and their relationships to actual  cases. [109] 

 3.2  The Probabilistic Model

 Some argue that legal concepts cannot be adequately represented by  definitions that state necessary and sufficient conditions.  Instead  legal concepts are incurably open-textured. [110] Typically an expert  system associates some kind of "certainty factor" with every rule in  its knowledge base, obtained from probabilities, or fuzzy logic, or  some combination of the two. [111]  Firstly, probabilities are used  alongside facts and rules, as a "slot" in the knowledge base.  To  each fact and rule is attached a certainty factor between zero and  one.  Concepts are mechanically linked, but the final output includes  a composite probability.  For example:

 A is true (0.8 chance)  If A is true then B is true (0.75 chance)  \ B is true (0.6 chance = 0.8 x 0.75)

   Secondly, fuzzy logic is called into play when classification of  facts involves weighting particular features.  In the former the rule  "fires", but a certainty level is attached to each fact  and rule.  In the latter, the rule will only fire at defined  threshold certainties.  If the weighted average of a set of  characteristics add to a threshold amount, the facts are classified  accordingly. 

 3.3  Case Based Reasoning

 Using precedents by induction and analogy is seen as advantageous in  overcoming apparently intractable problems of classification. [112]   However, both analogy and induction are inherently non-computational.  [113]  Case based reasoning is one attempt to imitate these  techniques, allegedly based on an exemplar model.  In the exemplar  model, the user is presented with prototypical instances, or "mental  images", on which to base her classification.  This approach differs  significantly from the two previous models  in that it is primarily  designed to leave the task of classification to the user.   Most case  based legal expert systems instead use a database of examples linked  to the decision given in particular cases.  When presented with a new  case for decision the system will attempt to match the case under  consideration with the examples, either stereotypical [114] or  actual, in its database to extract those which appear to be most  similar.  On that basis it will attempt to predict the outcome of the  new case.  For instance, Popple's SHYSTER [115] applies rules until  the meaning of some open texture concept is required.  At this point  a case based reasoning mechanism attempts to resolve this  uncertainty. [116]

 A matching algorithm is used to measure the similarity of cases in  terms of "case features".  Each case is modelled as a frame with  significant features contained in "slots". [117]  These features are  then weighted by some statistical method. [118]  The object of  constructing these "similarity metrics" [119] is to retrieve the most  "on-point" cases. [120]  Using weighted characteristics is described  as a "dimensional" [121] approach, such as Betzer's "3-D" system  which uses relative weights in a "procedure sweep" to fill in "gaps"  in the knowledge base. [122] 

 In addition to the facts, the cases themselves are often weighted in  a manner "meaningful to lawyers", usually to reflect some sense of  stare decisis.  For instance, the weights might be determined by the  level of the tribunal. [123]

 Although "case based" reasoning allegedly possesses an advantage over  rule based systems by the elimination of complex semantic networks,  [124] it suffers from intractable theoretical obstacles. Apart from  the question of choice of a matching algorithm, without some further  theory it cannot be predicted what features of a case will turn out  to be relevant.  Too often, "legally significant parameters" [125]  are facts deemed important by the programmers, [126] with no  grounding in any articulated theory, even though the utility of such  systems depends critically on the set of attributes selected. [127]    Both selection of attributes and the choice of associated weights are  therefore highly arbitrary. [128]  

 On this analysis, case based reasoning constitutes an extension of  the probabilistic model rather than a true exemplar model, in which  the task of classification is left to the user.  The potential of  this latter model for building applications is examined in Section 5.

 4: PHILOSOPHICAL IMPLICATIONS OF THE DEDUCTIVE INFERENCE ENGINE

 The previous Sections have demonstrated that the task of knowledge  representation was to provide a symbolic representation of knowledge  in a form amenable to the deductive inference engine.  The primary  reference point was logic programming, that is, formalisation of the  law into a set of Horn clauses.  It was also argued that techniques  such as "fuzzy logic" "semantic networks" and "case based reasoning"  are no more than elaborations of logic programming, [129] and not, as  some would argue, "second generation" systems going beyond deductive  inference. [130] 

 This Section carries this analysis beyond the practical and into the  philosophical.  It has been thought inescapable that a legal expert  system which attempts to emulate the reasoning processes of a lawyer  must embody theories of law that must in turn rest on more basic  philosophical assumptions. [131]  Building a legal expert system is  thus described as not being just an exercise in computer programming,  but requires "solid and articulated" jurisprudential foundations.  [132]  Researchers in this field appear, however, to have discounted  or ignored the value of close analysis of the field's theoretical  assumptions.  This Section demonstrates how, to avoid theoretical  obstacles, the nature of law, its epistemological basis, and the task  of jurisprudence have all been subjected to unacceptable distortion.

 4.1  Isomorphic Representation or Distortion?

 The activity of legal knowledge representation is said to involve the  operation of interpretative processes whereby the formal sources of  part of a legal system are scrutinised and analysed, so as to be both  faithful in meaning to the original source materials, and in a form  which is computer encodeable.  This principle is termed  "isomorphism". 

 Doubts have been raised as to whether it is possible to meet both  objectives.  It is said that the knowledge engineer must desist from  imposing her own interpretations, lest she be universally condemned  for misrepresenting the law. [133]  Yet it is difficult to reconcile  Susskind's claim of isomorphism with his admission that the process  is in fact one of complete "reformulation" or "rational  reconstruction". [134] 

 Although isomorphism requires the formalisation of rules to be  sufficiently expressive to capture their original meaning, [135]  Levesque and Brachman have demonstrated that there is a significant  trade off between the expressiveness of a system of knowledge  representation and its computational tractability. [136]  Susskind  admits that it is not possible, without "extensive modification and  inconvenience" to accommodate legal knowledge within the restrictive  frameworks offered by currently available computer programming  environments. [137]

 Moles provides a typical example of the modelling of British coal  insurance claims. [138]  This involved taking statutes and cases and  then "translating" them into six different structures using three  separate applications into the target representation language.  After  being "translated, cut up into bits, precised, further analysed into  [frames], which are then stored in another structure", he suggests it  would be a "miracle" if they were "isomorphic" to the original texts.  [139]  It would appear that terms such as "isomorphism" may be no  more than "syntactic sugar" [140] used to "sweeten" the acceptability  of what must be a distorting process. 

 4.2  Law as a System of Easily Interpreted Rules

 Statutory interpretation has been predominantly characterised as  involving a literal interpretation, [141] particularly in tax law,  [142] allegedly idiosyncratic in being construed both literally and  strictly. [143]  In case law, legal sources cannot be as easily  "formalised" or "normalised", [144] but must to some degree be  interpreted.  However, Susskind takes a dangerous step in suggesting  that the task of the knowledge engineer is to "sift the authoritative  ratio decidendi from the text  eliminating obiter dicta and other  "extraneous" [145] material.  Moreover, he argues that this can be  easily extracted, not by a thorough examination of the case but by  reading the headnote alone. [146]  Representation of cases in  knowledge bases typically are compressed into a "headnote" style,  including citation, court, date, facts, and holdings. [147]  Cost may  also be a factor behind this characterisation of law.  Susskind  suggests that knowledge engineers need only avail themselves of the  services of the legal expert to "tune" the knowledge base. [148]

 This view, however, that the law is a formal rule-governed process,  ignores a great deal of learning stretching back over a century  -  and more recently in the form of critical legal studies - arguing  that the law is far from determinate. [149]  The law is at least an  "elastic" phenomenon in which students have traditionally been taught  and encouraged to "flip" legal argument. [150]  The conception of  legal decision-making as a formal rule-governed process has been  eroded by a judicial move towards "realist" scepticism of rigid rule  structures.  For example, members of the Australian High Court have  indicated a rejection of formalism and adoption of a more active  assessment of legal principles with respect to justice, fairness, and  practical efficacy. [151]

 Advocates of case based reasoning attempt to accommodate realist  criticism by suggesting that fact patterns can explain legal  decisions independently of any "surface discourse" of law. [152]  The  critical assumption is that judges decide even hard cases in a rule  based manner.  Levi [153] supports this view in arguing that legal  reasoning, while not being purely a system of applying the law  mechanically to facts, does embody rules obtained by analysing the  similarities and differences in decided cases.  Such researchers  argue analogously to theorists, such as Goodhart, who suggest  examination ought to be focussed on the facts treated as material and  immaterial by the court. [154]

 Stone, however, argues that there is a critical distinction between  the ratio which explains the decision and the one which binds future  courts.  More often than not, the critical facts are those treated as  material by the later court, and even if they are identifiable, they  can be stated at multiple levels of generality. [155]

 Other legal systems, particularly in some parts of Europe, may be  more suited to this characterisation of law.  For example, in the  Scandinavian legal system, one overall guiding principle is the  prohibition of decisions which are non liquet [156], which is  considered a serious fault. [157]

 4.3  Change

 A severe impediment to the routine use of knowledge based technology  for practical legal applications lies in the unresolved problems  associated with the "maintenance" of such systems, that is, how to  continually update the system with primary sources. [158]  One group  describe how after exhaustively studying over 1,000 cases under the  Canadian Unemployment Insurance Act, it was amended in 1990 rendering  their work irrelevant. [159]  Most approaches are inadequate, either  for expressly assuming a constant state of the law, or avoiding  primary sources and instead modelling directly the heuristics of the  expert. [160]

 The logic programming approach of the Imperial College group, whereby  the expert system is formalised to correspond to individual sections  of a statute, is argued to be easily modifiable.  However, a fully  functioning expert system requires a layer of pertinent heuristic  knowledge to avoid a "layman's reading" of an Act. [161]  Once the  formalisation is structured, explained and augmented in this way,  modifying the system is no longer straightforward.    Schlobohm  suggests that, as a result, human experts would have to modify the  heuristic rules whenever the law changes, and the entire system  containing the new rules would then have to be debugged. [162]  Similarly, use of modular approaches such as the Chomexpert system  have proved inadequate. [163]   It is difficult to encode statutory  rules at even the most basic level without making inappropriate  commitments as to how they will be interpreted in future. [164] 

 4.4  Epistemology of Law

 Susskind suggests that law is not an abstract system of concepts and  entities distinct from the "marks on paper" that are the material  symbols of it. [165]  The difference between legal expert systems and  scientific systems such as PROSPECTOR and MYCIN lies, according to  Susskind, in that scientific laws are to be "discovered" in the  empirical world in general, while legal rules can be extracted, as an  acontextual linguistic exercise, from scrutiny of formal legal  sources. Under this analysis, knowledge engineers need go no further  than the written text, hence Susskind argues that the "bottleneck" of  knowledge acquisition is effectively dissolved. This is a dangerously  narrow epistemology to adopt, [166] since researchers in this area do  not sufficiently distinguish between the writing, and the meaning of  the writing. [167]  In Section 5 it is argued that meaning can only  be found in a social context.

 4.5  The Nature of Jurisprudence

 Although the foregoing suggests that many theories in jurisprudence  conflict significantly with important assumptions of expert systems  technology, many of these fundamental theoretical difficulties have  been downplayed or eliminated.  When faced with theories which imply,  for instance, that there is no future for expert systems,  some  researchers have expressly rejected the usefulness of jurisprudence.  [168]   Critical legal theory is therefore characterised as  "unacceptable". [169]  Even if jurisprudence is wholly ignored by  knowledge engineers, they suggest that the only risk is that the  systems they design might be of some "inferior quality". [170]

 Others "rationally reconstruct" jurisprudence into an acceptable  form.  For instance, Susskind asserts that the activity of any "legal  science" is to impose order over unstructured and complex law by  recasting it into a body of structured, interconnected, coherent, and  simple rules. [171]  Smith and Deedman go further and argue that the  task is to transform apparent indeterminacy into a completely  rule-governed structure. [172] 

 The same can be said for the portrayal of the epistemology of  jurisprudence.  Just as "complex" law is recast into "simple" rules,  the task of Susskind was to take "confused and perplexed"  jurisprudence, and obtain "consensus" over relevant issues.  What  Susskind does to find "consensus" in legal theory, is to allegedly  statistically sample the literature. [173]  However, the "sample" was  limited firstly to works of analytical jurisprudence, and secondly to  British writings from the mid-1950's. [174]  Perhaps unsurprisingly,  the influence of H.L.A. Hart's concept of law as a system of rules  was overwhelming.  As an adjunct, Susskind further asserted that to  be "jurisprudentially impartial", that is, to embody no "contentious"  theory of law, an expert system must reason only with rules.  Any  facility for reasoning with non-rule standards [175] was rejected out  of hand.  A significant internal inconsistency emerges when it is  appreciated that Susskind believed that it was sufficient  justification for use of rules that this "consensus" identifies legal  rules as necessary but insufficient for legal reasoning. [176]

 Perhaps the clue to why these works were chosen lies in the fact that  they constituted "the source materials with greatest potential given  the overall purpose of the project".  Susskind notes that his work  was intended to "eliminate much of the future need for extensive  scrutiny of non-computationally oriented contemporary legal theory".  [177] Here the inference engine is most clearly "driving"  jurisprudence. 

 4.6  Jurisprudence Turned on its Head

 Niblett claimed that "a successful expert system is likely to  contribute more to jurisprudence than the other way around". [178]   If the suggestions of researchers such as Susskind are taken  seriously, they turn jurisprudence on its head.  Theory is not used  as a basis for practice, but instead implementability in technology  is used as the touchstone for accepting the truth or falsity of the  theory.  A particular feature of artificial intelligence literature  is that its rigour lies not in experimental corroboration, or any  theoretical soundness, but implementability. [179]  Hofstader [180]  suggests that so long as the artificial researcher takes care to  construct theories which can be written down as a sequence of  algorithmic or computational steps, these theories can be  implemented, thereby "confirming" the theories underlying the  process.  Implementability per se leads to a self-perpetuating  methodology: since an artificial intelligence researcher will use  concepts of computational theory to construct theories, it is  necessarily implementable. 

 Legal expert systems researchers fall into this model by rejecting  "unacceptable" legal theories, and reformulating the remainder in  computational terms, to eliminate potential obstacles to the  prosperity of their research programme.  This abandonment of serious  inquiry into jurisprudence by researchers into legal expert systems  may give credence to Kowalski's fears that the field may have cut  itself off as a specialist discipline and established its parameters  prematurely. [181]  Brown notes that at a 1991 conference, few if any  papers questioned the basic assumptions of the field. [182]

 Niblett claimed that "a successful expert system is likely to  contribute more to jurisprudence than the other way around". [183]   The foregoing demonstrate that these words ring true.  Law and  jurisprudence, to form an acceptable basis for expert systems  research, has been reformulated in computational terms, to eliminate  philosophical "technicalities". [184]  

 4.7  Failure to Recognise Limitations

 Leith has argued for a rejection of legal expert systems on the basis  that they simplify the law to such an unacceptable extent that they  have little or no value in legal analysis. [185]  Yet while some  engineers of legal expert systems may be fully aware of the  limitations already discussed, it is not necessarily the case that  other researchers, and more importantly, the users of these programs  will also be so mindful.  This article agrees with Leith's implied  suggestion that many accounts of work in this area refuse to  acknowledge that there are significant limitations.  For example,  McCarty felt able to say:

 "[Law] seems to be an ideal candidate for an artificial intelligence  approach: the "facts" would be represented in a lower level semantic  network, perhaps; the "law" would be represented in a higher level  semantic description; and the process of legal analysis would be  represented in a pattern-matching routine." [186]

 Susskind has, however, admitted that expert systems might not be  amenable to corporate, commercial and tax law, but would be apposite,  for example, but to limited instances such as the Scottish law  relating to liability for damage caused by animals. [187]  Such  limitations, often given little attention, should be made clear, and  "plausibility tricks" avoided. [188]   There is a very real danger  that users will significantly overestimate the value of the analysis  they obtain from such a program, particularly in light of the wealth  of optimistic literature and when it is described as "expert". 

   5: LEGAL REASONING AS AN INTRACTABLY COMPLEX SYSTEM

 The previous Section demonstrated how law and jurisprudence have been  unacceptably distorted to be amenable to expert systems technology.   Moles suggested that researchers have deliberately ignored  fundamental problems since they were committed to the use of a  "particular computing tool", and not to the understanding of law.  [189]  This article identifies this tool as the inference engine  itself.  The following section addresses the non-computational nature  of legal inference.

 5.1  Search for a Deep Model

 There has been a growing trend in legal expert systems to speak of  "deep knowledge" or "conceptual knowledge" as something distinct and  preferable to "shallow" knowledge. [190]  McCarty calls for the  development in law of "deep" systems akin to CASNET [191] in which  the disease is represented as a dynamic process. [192]  The depth of  a system has been described as the extent to which programmes contain  not only rules for mapping conclusions onto input scenarios, but also  a representation of the underlying causes. [193] 

 The Imperial College Group suggest that deep structure in legislation  is the isomorphism to that legislation, on the basis that each Horn  clause represents some clause in the legislation. [194]  In addition,  case based reasoning has been described as employing a "deep  structure". [195]  The advantage stemming from both of these  descriptions is that they cast deep structure into computational  terms. [196]  This is another example of technology driving the  underlying theory.

 On the other hand, McCarty argues that resolution of the difficulties  of open texture are related to a sense of "conceptual coherence".  [197]  In addition, while theoretical approaches are emerging  to cope with problems of legal change, [198] a unifying theme is a  striving for an undefined "normative enrichment". [199]  This Section  argues that deep structure is to be found in social context and  purpose, which are non-computational.

 5.2  Interpretation in a Social Context of Shared Understanding

 Law is not, as legal expert systems would portray it, self-contained  and autonomous, [200] but in fact is embedded in social and political  context.  That legal concepts draw upon ordinary human experience is  precisely what makes them so difficult for an artificial intelligence  system. [201]  Whenever human behaviour is analysed in terms of  rules, it must always contain a ceteris paribus condition; in  essence referring to the background of shared social practices,  interests and feelings.  Even if we accept Susskind's  characterisation of law's ontology as going no further than the  "marks on paper", semantic problems will still arise since these  marks are not created in a vacuum, but are the result of purposive  social interaction, and must be so interpreted. [202]

 Using one recognised example, the injunction:

 DOGS MUST BE CARRIED ON THE ESCALATOR

 can only be interpreted based on the understandings, for instance,  that a dog's small feet may become trapped in slots and moving parts;  that humans generally feel some concern for dogs, and therefore do  not wish to see them "mangled".  Thus an adequate interpretation of  any rule requires that we locate it in a complex body of assumptions.  [203]

 Minsky noted that intelligent behaviour presupposes a background of  cultural practices and institutions which must be modelled if  computer representations are to have any meaning. [204]   Wittgenstein's arguments that the meaning of language must be based  in social use and a community of users are worth rereading in the  light of Searle's Chinese room analogy. [205]  How can the computer  have this sort of direct access to language? [206]  Kowalski and  Sergot admit that a computer must operate by "blind" and "mechanical"  application of its internal rules. [207]

 5.3  Open Texture as an Intractable Problem

 If legal reasoning was really some "pointing" [208] or "cataloguing"  [209] procedure, Hart's suggestion that the task of legal  institutions is to approach greater refinement in definition by  adjudicating [210] on particular cases, may be attractive. [211]   Open texture may then be marginalised by a progressive refinement of  categories; in computational terms, weaving a more elaborate semantic  net.  To model social context in a knowledge base, however, may be an  impossible task. 

 Popper demonstrates that context entirely depends on point of view.  [212]  Harris suggests that any view of the law must be a  phenomenological one which takes account of shifting foci of  interest. [213]  The difficulty is that a great deal of social  context will not be "conscious" and expressible, but will constitute  a hidden set of assumptions on which human decisions will be based.   It is impossible to focus attention onto elements of that context  without creating a new subconscious context.  Polyani describes this  as the difference between focal and subsidiary awareness. [214] 

 As Berry [215] demonstrates, if people learn to perform tasks so that  important aspects of their knowledge are implicit in nature, then  knowledge engineers will be unable to extract this knowledge and  represent it in a meaningful way in an expert system. [216]  Husserl,  for instance, discovered that construction of even simple "frames"  involved coping with an ever expanding "outer horizon" of knowledge.   He sadly concluded at the age of 75 that he was a "perpetual  beginner" engaged in an "infinite task". [217]  This is a   fundamental difficulty with artificial intelligence in all its  applications. [218]

 5.4  Purposive Interpretation and Intention

 Hempel [219] argued that ad hoc modifications to a theory were  limited by the increased complexity of the theory and that, after a  certain threshold level of complexity was exceeded, scientists would  naturally and logically pursue simpler alternative theories. [220]   Here we may learn from science.  Certain physical and chemical  systems have been discovered that display uncanny qualities of  co-operation, or organise themselves spontaneously and unpredictably  into complex forms.  These systems are still subject to physical  laws, but laws that permit a more flexible and innovative type of  behaviour than the old mechanistic view of nature ever suggested.   The lesson from chaos theory is that seemingly complex systems can be  defined in terms of simple but not mathematically tractable models.  [221] 

 Legal reasoning is not mechanical. [222]  Social context and shared  understandings can be dealt with in terms of the simple, elegant, but  non-computational model of purposive interpretation.  Searle's  Chinese room analogy identifies intentionality as the benchmark of  the mental, and refutes claims that intentional mental predicates,  such as meaning, understanding, planning, and inferring, can be  attributed to a mathematical computational system. [223] 

 Susskind prefers to avoid purposive theories, [224] since such  theories imply that law is not simply a question of linguistic  pattern matching but instead involves examination of social practices  and human intentionality. [225]  Similarly, case based reasoning is  seen as a way around having to tackle "full blown" statutory  interpretation involving legislative intent. [226]

 Law is a practical enterprise, concerned to guide, influence or  control the actions of citizens.  Since any action is purposive, any  philosophy of action must be a philosophy of purposes. [227]  When a  court applies, say, the statutory term of our example, "vehicle", to  a particular contraption, the meaning of "vehicle" is found in an  analysis not only of the purpose of the law, but of the purpose for  which the vehicle was to be used. [228]  For example, Fuller  responded to Hart in these terms:

 "What would Professor Hart say if some local patriots wanted to mount  on a pedestal in the park a truck used in World War 2, while other  citizens, regarding the proposed memorial as an eye-sore, support  their stand by the "no vehicle" rule?  Does this truck, in perfect  working order, fall within the core or the penumbra?" [229]

 One could make similar arguments when a "NO DOGS ALLOWED" sign  confronts a seeing eye dog, or one that is stuffed or anaesthetised.  [230]  It is difficult to reconcile Hart's acontextual approach to  legal interpretation with his own view of actors within the legal  system holding an internal normative view of the rules. [231]  Following a rule equals "obeying the law" only where a purposive  personal commitment has been made to a rule structure. [232] 

 Applying modern literary and linguistic theory to the law, [233] some  suggest that no text has meaning without the active participation of  the reader, [234] and an "interpretive community" of which the reader  is a part. [235]  The use of figurative language, imagery and  metaphor is integral to legal discourse. [236]  Ideological symbolism  is inescapable. [237]  What counts as the relevant facts depends  entirely on context, [238] and cannot be determined by programmers ex  ante. [239]  Language is the very condition of intention, and  intention is the vehicle of meaning. [240]

 5.5  Humanistic Implications

 The implication of the foregoing suggests that the law cannot be  amenable to a legal expert system, as this involves denying social  context, purpose, and essentially humanity.  A humanistic critique  would argue if expert systems have any degree of success in modelling  "the law", the result would be "profoundly humiliating". [241]   Weizenbaum stated that if artificial intelligence fulfils its  promises then this implies that man is merely a machine. [242]  In  similar vein, the success of legal expert systems might imply that  the law itself is a machine, and that lawyers, perhaps even judges,  can be replaced by computers.

   6: THE WAY FORWARD

 The preceding sections have demonstrated that the use of a deductive  inference mechanism, and the consequent need for knowledge  represented to be amenable to such an engine, will lead to  unacceptable distortion of both the law, its philosophical  underpinnings, and its humanity.  How are lawyers then effectively to  utilise the information technology resource?  This article adopts the  basic message in Tapper's insightful 1963 piece. [243]  The range of  activities to which computers ought be used must be limited to  activities which can be reproduced by the machines.  Tapper  tentatively describes the distinction as one between "mechanical" and  "creative" tasks.  [244]  If the argument of this article is  accepted, the way forward involves relocation of the inference engine  from the computer to the human user.  This section explores  possibilities for "decision support systems", which presents material  to the user on which she alone performs the specifically legal  reasoning. [245]

 6.1  Decision Support Systems

 Recall that in the exemplar model, the user is presented with  prototypical instances, or "mental images".  This approach differs  from the other models of classification in that it is primarily  designed to leave the task of classification to the user of the  system.  The reason why the user, rather than the machine, ought  perform the legal inference is that legal reasoning is  non-computational, as Section 5 has demonstrated.

 Despite growing recognition that research perhaps ought to be  oriented towards "decision support systems", such systems have been  designed to first reason with the legal data and then present such  reasoning to the user to support her conclusion. [246]  This approach  is hazardous since it may predetermine the human conclusion to a  large degree. [247]  To dispute the computer inference the user would  require knowledge of the area of law to a degree where the computer  would not need to be have been consulted in the first instance. [248]

 Decision support systems, then, differ significantly from expert  systems in that the heart of the problem - the inference engine - is  relocated in the user of the system.  Computers should then be  utilised for mechanical and time-consuming tasks for which they are  best suited.  In particular, this Section suggests three significant  uses:   - structured legal information retrieval;  - calculation based on strategies; and  - litigation support and "legal econometric" systems.

 6.2  Legal Information Retrieval

 Firstly, searching for primary and secondary legal sources is a  costly and to a significant degree a mechanical exercise.  Efficient  retrieval of legal information is vitally important.  Tapper  suggested in 1963 that lack of resources to those operating outside  provincial centres, and concentration of materials within large  organisations was productive of injustice in favour of powerful  sections of the community. [249]  Modern statute and case databases  have gone some way to addressing this problem. 

 Generally, systems such as LEXIS, SCALE, and INFO1 use Boolean  keyword search routines, [250] but these have obvious disadvantages.  [251]  Some limited advances have been made with, for instance,  "Hypertext" cross-referencing, [252] and "probabilistic" elaboration  of keyword searches. [253]  It has long been assumed that retrieval  based on the meaning and content of documents, and indexed in terms  of legal concepts, [254] would be far more appropriate. [255]  A  variety of techniques have emerged for indexing, including use of  discrimination trees, [256] and explanation based generalisation.  [257] Research is progressing towards a "hybrid" approach of linking  case databases with statutory material and legal texts. [258]  Hafner  [259] has constructed a database on United States negotiable  instruments law designed to retrieve cases based on typical problems  which arise in legal disputes. 

 McCarty has suggested that it would be more fruitful to look at legal  argument than to develop a theory of correct legal decisions. [260]   Similarly Bench-Capon and Sergot suggest that open texture should be  handled by giving the user for and against arguments in borderline  cases.  If so, a computer system will be concerned, not with the  production of a conclusion, but rather with presenting the arguments  on which the user may base her own conclusions. [261] 

 On this basis, Ashley and Rissland have designed a system called  HYPO, [262] which emerged from Rissland's earlier work on reasoning  by examples. [263]  It does not use an inference engine for legal  analysis but instead aims for conceptual retrieval based on structure  of legal argument. [264]  The system's inference engine is used for  some statistical processes used to decide which primary materials to  retrieve.  The cases relevant to the issues identified by the user  are retrieved and arranged in terms of argument for and against a  decision in a new case. [265]  The system is further supplemented by  a set of "hypothetical" cases. [266]  The actual legal inference on  the basis of the material retrieved is left to the user of the  system, which distinguishes HYPO and similar text retrieval systems  from many of the case based reasoning systems earlier. 

 6.3  Calculations and Planning

 A second application would be to utilise the mathematical functions  of computer systems.  The computer performs inference, but  essentially calculates outcomes based on strategies already  formulated by an expert who has himself interpreted the legal  materials.  Michaelson's TAXADVISOR [267] is one example.  It  calculates tax planning strategies for large estates, based on  strategies obtained from lawyers experienced in tax advice.  There is  little more legal inference in this than calculating a share  portfolio to maximise return based on a broker's personal model.   Similarly, systems have been suggested which will assist in financial  planning, for instance by forecasting retirement pensions. [268] 

 6.4  Litigation Support and Jurimetrics

 Finally, a decision support system may more clearly focus on  litigation strategies.  These may be developed with the assistance of  expertise, or by techniques such as hypothesis and experiment. [269]   Such systems do in fact make inferences, but these are not inferences  of law, but inferences based on strategies already defined by  expertise or essentially what amounts to empirical research.  In that  sense, the inference procedures are extensions to the calculation and  planning examples.

 One example is the LDS [270] system, implemented in ROSIE [271] by  Waterman and Peterson.  It advises on whether to settle product  liability cases, and an advisable amount, based on factors such as  abilities of the lawyers, characteristics of the parties, timing of  claim, type of loss suffered, and the probability of establishing  liability.  The primary goal of LDS was not to model the law per se  but rather the actual decision making processes of lawyers and claims  adjusters in product liability litigation.  Another example is SAL,  [272] intended to advise on an appropriate sum to settle asbestos  injury claims.  In such systems, the computer is modelling non-legal  factors which may influence the outcome of a case, in order to assist  the lawyer in deciding her ultimate strategy.  In Australia the  Government Insurance Office has developed COLOSSUS, a sophisticated  system to detect possible fraudulent personal injury claims, and tag  them for investigation by its officers. [273] 

 Similar systems have also been suggested as aids not only in  litigation, but dispute resolution strategies. [274]  The information  contained in such systems may as an adjunct constitute an important  resource for sociological study, such as Bain's modelling of  subjective decisions of judges of particular varieties of crime in  the United States. [275]  In this case, the expert system constitutes  jurimetrics, a legal version of econometrics.  

   7: SUMMARY AND CONCLUSIONS

 Computerisation of the legal office will continue, but the message  from this article is that researchers must be acutely aware of the  philosophical underpinnings of their work.  In particular, the  usefulness of legal expert systems is severely questioned.  Use of  such systems has involved an unacceptable level of distortion both of  the nature of law and of jurisprudence.  This is not a case of  "carbon", [276] "biological", [277] or even "neural" [278]  chauvinism, but a demonstration that expert systems technology have  made a poor choice of domain in law.  Blame has been laid for such  distortion on the core of the expert system: the pattern matching  inference engine.  Legal inference, on the other hand, relies on  purpose and social context, implying that computational models of  sufficient richness are not tractable. 

 This article suggests that, given current limitations of computer  technology, the quest for an artificially intelligent legal adviser  is misguided.  In the future, however, these limitations may be  overcome.  For example, work being undertaken in parallel distributed  processing is producing significant results with respect to low level  "intelligent" processes, including perception, language, and motor  controls.  This is based on the assumption that intelligence emerges  from interactions of large numbers of simple processing units, and  represents a significant break away from increasingly complex  rule-based structures. [279]  While this article cannot address such  possibilities within future technology, it is suggested that the  basic pattern matching /rule-governed principles will limit computers  for some time.  It is therefore suggested that researchers instead  investigate decision support systems as a more useful alternative.   Relocation of the inference engine will mean that knowledge  representation will no longer need be amenable to computational  inference, but human inference.  The computer's inference engine  should instead be used instead for searching procedures, and  computation.  Some possibilities have been noted.

 Ardent advocates such as Tyree suggest that despite their  difficulties, legal expert systems are a cost-effective second-best  solution.  The choice is portrayed not between human advice and  machine advice, but in an era of high costs of justice, between  machine advice and no advice at all. [280] 

 While economic factors are important, [281] humanistic factors must  not be forgotten.  Law plays an important role in modern  civilisation.  It must maintain a close relationship with the social  and political forces shaping society, and not merely regress into a  "technology", a tool to be used by competing social forces. [282]

   ENDNOTES

 *   J Searle, "Minds, Brains and Programs" (1980) 3 Behavioural and  Brain Sciences 417.

  1  For examples see NJ Bellord, Computers for Lawyers (Sinclair  Browne: London, 1983); and  T Ruoff, The Solicitor and the Automated  Office (Sweet & Maxwell: London, 1984).

  2  q.v. J Bing (ed.), Handbook of Legal Information Retrieval  (North-Holland: Amsterdam, 1984).

  3  See Section VI, infra.

  4  Inferring molecular structure from mass spectroscopy data; q.v.   RK Lindsay, BG Buchanan, & J Lederberg, Applications of Artificial  Intelligence for Chemical Inference: The DENDRAL Project  (McGraw-Hill: New York, 1980).

  5  Advising on the location of ore deposits given geological data;  q.v.  RO Duda & R Reboh, "AI and Decision Making: The PROSPECTOR  Experience" in W Reitman, Artificial Intelligence Applications for  Business (Ablex Publishing: Norwood, 1984).

  6  Providing consultative advice on diagnosis and antibiotic therapy  for infectious diseases; q.v. BG Buchanan & EH Shortcliffe,  Rule-Based Expert Systems: The MYCIN Experiments of the Stanford  Heuristic Programming Project (Addison-Wesley: Reading, 1984).

  7  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.11.

  8  Ibidem p.15.

  9  L Loevinger, "Jurimetrics: The Next Step Forward" (1949) Minnesota  Law Review 33.

  10  L Mehl, "Automation in the Legal World: From the Machine  Processing of Legal Information to the 'Law Machine'" in  Mechanisation of Thought Processes  (HMSO: London, 1958) p.755.

  11  RW Morrison, "Market Realities of Rule-Based Software for  Lawyers: Where the Rubber Meets the Road" (1989) Proceedings Second  International Conference on Artificial Intelligence and Law 33 at  p.35.

  12  c.f. RA Clarke, Knowledge-Based Expert Systems  (Working paper:  Department of Commerce, Australian National University, 1988) p.6.

  13  Primarily RE Susskind, Expert Systems in Law: A Jurisprudential  Inquiry (Clarendon Press: Oxford, 1987).

  14  Ibidem p.44; emphasis added.

  15  MJ Sergot, "The Representation of Law in Computer Programs",  Chapter One in TJM Bench-Capon, Knowledge-Based Systems and Legal  Applications (Academic Press: London, 1991) at p.4.

  16  P Harmon & D King, Expert Systems: Artificial Intelligence in  Business (John Wiley & Sons: New York, 1985) at p.5.

  17  R Forsyth, "The Anatomy of Expert Systems" Chapter Eight in M  Yazdani (ed.), Artificial Intelligence: Principles and  Applications  (Chapman & Hall: London,

  1986) pp.186-187.

  18  R Forsyth, "The Anatomy of Expert Systems" Chapter Eight in M  Yazdani, Artificial Intelligence: Principles and Applications   (Chapman and Hall: London, 1986) p.194.

  19  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.46.

  20  Ibidem p.47.

  21  F Hayes-Roth, DA Waterman & DB Lenat Building Expert Systems  (Addison-Wesley: London, 1983) p.4.

  22  e.g. The Latent Damage Adviser; q.v. PN Capper & RE Susskind,  Latent Damage Law - The Expert System (Butterworths: London, 1988).

  23  J Vaux, "AI and Philosophy: Recreating Naive Epistemology"  Chapter Seven in KS Gill (ed.), Artificial Intelligence for Society  (John Wiley & Sons: London, 1986) p.76.

  24  T Winograd & F Flores, Understanding Computers and Cognition: A  New Foundation for Design (Ablex: Norwood, 1986).

  25  HL Dreyfus & SE Dreyfus, Mind over Machine (Basil Blackwell:  Oxford, 1986).

  26  A Hart & DC Berry, "Expert Systems in Perspective" in DC Berry &  A Hart (eds) Expert Systems: Human Issues  (MIT: Cambridge, 1990)  p.11.

  27  e.g. J Weizenbaum, Computer Power and Human Reason: From Judgment  to Calculation (WH Freeman & Co: San Francisco, 1976).

  28  R Hellawell, "A Computer Program for Legal Planning and Analysis:  Taxation of Stock Redemptions" (1980) 80 Columbia Law Review 1363.   See also NJ Bellord, "Tax Planning by Computer" in B Niblett (ed.),  Computer Science and Law (Cambridge University Press: New York, 1980)  p.173.

  29  R Hellawell, "CHOOSE: A Computer Program for Legal Planning and  Analysis" (1981) 19 Columbia Journal of Transnational Law 339.

  30  R Hellawell, "SEARCH: A Computer Program for Legal Problem  Solving" (1982) 15 Akron Law Review 635.

  31  P Jackson, H Reichgelt & Fv Harmelen, Logic-Based Knowledge  Representation (MIT: Cambridge, 1989).

  32  Implemented in QUINTUS PROLOG; q.v. NH Minsky & D Rozenshtein,  "System = Program + Users + Law" (1987) Proceedings First  International Conference on Artificial Intelligence and Law

  170.

  33  Symbolic logic has had a profound influence in the artificial  intelligence field; for a description see I Copi, Symbolic Logic  (Macmillan: New York, 1973).

  34  MJ Sergot, "A Brief Introduction to Logic Programming and Its  Applications in Law" Chapter Five in C Walter (ed.) , Computer Power  and Legal Language (Quorum: London, 1988)  at pp.25-27.

  35  C Mellish, "Logic Programming and Expert Systems" Chapter  Nineteen in KS Gill (ed.), Artificial Intelligence for Society (John  Wiley & Sons: London, 1986) at p.211.

  36  F Hayes-Roth, DA Waterman & DB Lenat, Building Expert Systems  (Addison-Wesley: London, 1983) at p.66.

  37  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.208.

  38  R Forsyth, "The Anatomy of Expert Systems" Chapter Eight in M  Yazdani, Artificial Intelligence: Principles and Applications   (Chapman and Hall: 1986) p.191.

  39  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) pp.209-210.

  40  RI Levine, DE Drang & B Edelson, Artificial Intelligence and  Expert Systems (McGraw-Hill: 1990) Chapter Six, particularly at  pp.62-65.

  41  e.g. AW Koers & D Kracht, "A Goal Driven Knowledge Based System  for a Domain of Private International Law" (1991) Proceedings Third  International Conference on Artificial Intelligence and Law 81. 

  42  q.v. RA Kowalski, "The Treatment of Negation in Logic Programs  for Representing Legislation" (1989) Proceedings Second International  Conference on Artificial Intelligence and Law 11; P Asirelli, M De  Santis & M Martelli, "Integrity Constraints in Logic Databases"  (1985) 2 Journal of Logic Programming 221; K Eshghi & RA Kowalski,  "Abduction Compared with Negation by Failure" (1989) Proceedings of  the Sixth International Logic Programming Conference;  and JW Lloyd,  EA Sonenberg and RW Topot, "Integrity Constraint Checking in  Stratified Databases" (1986) 4 Journal of Logic Programming 331.

  43  TJM Bench-Capon, "Representating Counterfactual Conditionals"  (1989) Proceedings Artificial Intelligence and the Simulation of  Behvaiour 51.

  44  "Augmented Prolog Expert System"; q.v. MJ Sergot, "A Brief  Introduction to Logic Programming and Its Applications in Law"  Chapter Five in C Walter (ed.), Computer Power and Legal Language  (Quorum: London, 1988)  at pp.34-35.

  45  P Hammond & MJ Sergot, "A PROLOG Shell for Logic Based Expert  Systems" (1983) 3 Proceedings British Computer Society Expert Systems  Conference.

  46  DA Schlobohm, "A PROLOG Program Which Analyses Income Tax Issues  under Section 318(a) of the Internal Revenue Code" in C Walter (ed.),  Computing Power and Legal Reasoning (West Publishing: St Paul, 1985)  p.765.

  47  q.v. F Haft, RP Jones & T Wetter, "A Natural Language Based Legal  Expert System for Consultation and Tutoring - The LEX Project" (1987)  Proceedings First International Conference on Artificial Intelligence  and the Law 75.

  48  C Walter, "Elements of Legal Language" Chapter Three in C Walter  (ed.), Computer Power and Legal Language (Quorum: London, 1988).

  49  LA Zadeh, "Fuzzy Sets" (1965) 8 Information and Control 338.

  50  E Rosch & C Mervis, "Family Resemblances: Studies in the Internal  Structure of Categories" (1975) 7 Cognitive Psychology 573.

  51  M Novakowska, "Fuzzy Concepts: Their Strcuture and Problems of  Measurement" in MM Gupta, RK Ragade & RR Yager (eds), Advances in  Fuzzy Set Theory and Applications (North-Holland: Amsterdam, 1979) at  p.361.

  52  TJM Bench-Capon & MJ Sergot, "Toward a Rule-Based Representation  of Open Texture in Law" Chapter Six in C Walter (ed.), Computer Power  and Legal Language (Quorum: London, 1988) at p.49.

  53  D Berman & C Walter (ed.), "Toward a Model of Legal  Argumentation" Chapter Four in C Walter (ed.), Computer Power and  Legal Language (Quorum: London, 1988) at p.22.

  54  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.225. 

  55  CE Alchourrsn & AA Martino, "A Sketch of Logic Without Truth"  (1989) Proceedings Second International Conference on Artificial  Intelligence and Law 165 at p.166.

  56  HLA Hart, "Problems of the Philosophy of the Law" in HLA Hart,  Essays in Jurisprudence and Philosophy (Clarendon Press: Oxford,

  1983) p.100; and H Kelsen, "Law and Logic" in H Kelsen, Essays in  Legal and Moral Philosophy  (Reidel: Dordrecht, 1973) at p.229.

  57  LT McCarty, "Permissions and Obligations - an Informal   Introduction" (1983) Proceedings International Joint Conference on  Artificial Intelligence-83;  LT McCarty, "Permissions and Obligations  - An Informal Introduction" in AA Martino & NF Socci (eds) Automated  Analysis of Legal Texts (North-Holland: Amsterdam, 1986).  Fora more  developed system on the same principles, see H-N Castaqeda, "The  Basic Logic for the Interpretation of Legal Texts" in C Walter (ed.),  Computer Power and Legal Language (Quorum: London, 1988) at p.167.

  58  GHv Wright, "Deontic Logic" (1951) 60 Mind 1.

  59  McCarty suggests that it is helpful to think of the set as an  "oracle" to be consulted when contemplating a course of action; see  LT McCarty, "Permissions and Obligations - an Informal  Introduction"  in AA Martino & NF Socci (eds) Automated Analysis of Legal Texts  (North-Holland: Amsterdam, 1986).  

  60  Note LT McCarty, "Permissions and Obligations - A Informal  Introduction" in AA Martino & NF Socci (eds) Automated Analysis of  Legal Texts (North-Holland: Amsterdam, 1986)Definitions 5-7.

  61  LT McCarty, "Clausal Intuitionistic Logic I: Fixed-Point  Semantics" (1988) 5 Journal of Logic Programming 1; LT McCarty,  "Clausal Intuitionistic Logic II: Tableau Proof Procedures" (1988) 5  Journal of Logic Programming 93.

  62  LT McCarty, "On the Role of Prototypes in Appellate Legal  Argument" (1991) Proceedings Third International Conference on  Artificial Intelligence and Law

  185 at p.187.

  63  AJI Jones, "On the Relationship Between Permission and  Obligation" (1987) Proceedings First International Conference on  Artificial Intelligence and Law

  164 at pp.166-168.

  64  LT McCarty & Rvd Meyden, "Indefinite Reasoning with Definite  Rules" (1991) Proceedings of the Twelfth International Joint  Conference on Artificial Intelligence.

  65  Particularly those of consequential closure; AJI Jones & I Pren,  "Ideality, Sub-Ideality and Deontic Logic" (1985) 2 Synthise 65.

  66  R Stamper, "The LEGOL-1 Prototype System and Language" (1977) 20  The Computer Journal 102.

  67   R Stamper, C Tagg, P Mason, S Cook & J Marks, "Developing the  LEGOL Semantic Grammar" in C Ciampi (ed.) Artificial Intelligence and  Legal Information Systems (North-Holland: Amsterdam, 1982) p.357.

  68  R Stamper, "LEGOL: Modelling Legal Rules by Computer" in B  Niblett (ed.), Computer Science and Law (Cambridge University Press:  New York, 1980) p.45.

  69  R Stamper, "A Non-Classical Logic for Law Based on the Structures  of Behaviour" in AA Martino & F Socci (eds), Automated Analysis of  Legal Texts (North-Holland: Amsterdam, 1986) p.57.

  70  S Jones, "Control Structures in Legislation" in B Niblett (ed.),  Computer Science and Law (Cambridge University Press: New York, 1980)  p.157.

  71  MJ Sergot, Programming Law: LEGOL as a Logic Programming Language  (Imperial College: London, 1980).

  72  MJ Sergot, "The Representation of Law in Computer Programs",  Chapter One in TJM Bench-Capon, Knowledge-Based Systems and Legal  Applications (Academic Press: London, 1991) at p.35.

  73  IE Pratt, Epistemology and Artificial Intelligence  (PhD  dissertation: Princeton, 1987) p.18; emphasis in original.

  74  Susskind and Gold.

  75  Including Bench-Capon, Cordingley, Forder, Frohlich, Gilbert,  Luff, Protman, Sergot, Storrs and Taylor; q.v. RN Moles, "Logic  Programming - An Assessment of Its Potential for Artificial  Intelligence Applications in Law" (1991) 2 Journal of Law and  Information Science 137 at pp.146-147.

  76  RE Susskind, "The Latent Damage System" (1989) Proceedings Second  International Conference on Artificial Intelligence and Law 23 at  p.29.

  77  On causality, note CG de'Bessonet & CR Cross, "Representation of  Some Aspects of Causality" in C Walter (ed.) Computing Power and  Legal Reasoning (West: St Paul, 1985) pp.205-214.

  78  e.g. SR Goldman, MG Dyer & M Flowers, "Precedent-based Legal  Reasoning and Knowledge Acquisition in Contract Law: a Process Model"  (1987) Proceedings First International Conference on Artificial  Intelligence and Law

  210 at pp.214-215 using Hohfeldian analysis of  rights; q.v. WN Hohfeld, "Some Fundamental Legal Conceptions As  Applied in Judicial Reasoning" (1917) 23 Yale Law Journal 16. 

  79  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.227.

  80  P Leith, "Logic, Formal Models and Legal Reasoning" (1984)  Jurimetrics Journal 334 at pp.335-336.  

  81  N MacCormick, Legal Reasoning and Legal Theory (Oxford University  Press: Oxford, 1978) at p.37.

  82  MJ Sergot, "A Brief Introduction to Logic Programming and Its  Applications in Law" Chapter Five in C Walter (ed.), Computer Power  and Legal Language (Quorum: London, 1988)  at p.26.

  83  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987)   pp.181-198, particularly at p.188.

  84  LE Allen & CS Saxon, "Some Problems in Designing Expert Systems  to Aid Legal Reasoning" (1987) Proceedings First International  Conference on Artificial Intelligence and Law

  94 at p.94.

  85  RE Susskind, "The Latent Damage System" (1989) Proceedings Second  International Conference on Artificial Intelligence and Law 23 at  p.28.

  86  Ibidem p.30.

  87  AvdL Gardner, "Overview of an AI Approach to Legal Reasoning" in  C Walter (ed.),Computing Power and Legal Reasoning (West: St Paul,

  1985) p.247.

  88  HLA Hart, "Positivism and the Separation of Law and Morals"  (1958) 79 Harvard Law Review 593 at p.599.

  89  G Gottlieb, The Logic of Choice: An Investigation of the Concepts  of Rule and Rationality (Allen & Unwin: London, 1968) p.17.

  90  OC Jensen, The Nature of Legal Argument (Basil Blackwell: Oxford,

  1957)  p.16; A Wilson, "The Nature of Legal Reasoning: A Commentary  with Special Reference to Professor MacCormick's Theory" (1982) 2  Legal Studies 269 at pp.278-280.

  91  MJ Detmold, The Unity of Law and Morality: A Refutation of Legal  Positivism (Routledge & Kegan Paul: London, 1984) p.15; c.f. RE  Susskind, "Detmold's Refutation of Positivism and the Computer Judge"  (1986) 49 Modern Law Review 125.

  92  HLA Hart, "Positivism and  the Separation of Law and Morals"  (1958) 79 Harvard Law Review 593 at p.607.

  93  DB Skalak, "Taking Advantage of Models for Legal Classification"  (1989) Proceedings Second International Conference on Artificial  Intelligence and Law 234.

  94  CD Hafner, "Conceptual Organisation of Case Law Knowledge Bases"  (1987) Proceedings First International Conference on Artificial  Intelligence and Law

  35 at pp.36-37.

  95  Note the diagram attached to RE Susskind, "The Latent Damage  System" (1989) Proceedings Second International Conference on  Artificial Intelligence and Law 23.

  96  MJ Sergot, "Representing Legislation as Logic Programs" (1985) 11  Machine Intelligence 209.

  97  MJ Sergot, F Sadri, RA Kowalski, F Kriwaczek, P Hammond, & HT  Cory, "The British Nationality Act as a Logic Program" (1986) 29  Communications of the ACM 370; and MJ Sergot, HT Cory, P Hammond, RA  Kowalski, F Kriwaczek, & F Sadri, "Formalisation of the British  Nationality Act" in C Arnold (ed.), Yearbook of Law, Computers and  Technology (Butterworths: London, 1986).

  98  TJM Bench-Capon, GO Robinson, TW Routen & MJ Sergot, "Logic  Programming for Large Scale Applications in Law: A Formalisation of  Supplementary Benefit Legislation" (1987) Proceedings First  International Conference on Artificial Intelligence and Law 190.

  99  q.v. P Johnson & D Mead, "Legislative Knowledge Base Systems for  Public Administration: Some Practical Issues" (1991) Proceedings  Third International Conference on Artificial Intelligence and Law 

  74.

  100  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.100.

  101  The concept can be attributed to Quillian; q.v. MR Quillian,  "Word Concepts: A Theory and Simulation of Some Basic Semantic  Capabilities" (1967) 12 Behvioural Science 410.

  102  WA Woods, "What's in a Link: Foundations for Semantic Networks"  in DG Bobrow & AM Collins (eds), Representation and Understanding:  Studies in Cognitive Science (Academic Press: New York, 1975) p.32.

  103  M Minsky, "A Framework for Representing Knowledge" in J  Haugeland (ed.), Mind Design (MIT Press: Cambridge, 1981) p.95.

  104  PJ Hayes, "The Logic of Frames" in D Metzing (ed.), Frame  Conceptions and Text Understanding (Walter de Gruyter: Berlin, 1979)  p.46.

  105  MJ Sergot, "The Representation of Law in Computer Programs",  Chapter One in TJM Bench-Capon, Knowledge-Based Systems and Legal  Applications (Academic Press: London, 1991) at p.48.

  106  LT McCarty, "Reflections on TAXMAN: An Experiment in Artificial  Intelligence and Legal Reasoning" (1977) 90 Harvard Law Review 837;  and LT McCarty, "The TAXMAN Project: Towards a Cognitive Theory of  Legal Argument" in B Niblett (ed.), Computer Science and Law  (Cambridge University Press: New York, 1980).

  107  PJ Hayes, "The Logic of Frames" in D Metzing (ed.), Frame  Conceptions and Text Understanding (de Gruyter: New York, 1979).

  108  Example adapted from MJ Sergot, "The Representation of Law in  Computer Programs", Chapter One in TJM Bench-Capon, Knowledge-Based  Systems and Legal Applications (Academic Press: London, 1991) at  p.46.

  109  KE Sanders, "Representing and Reasoning About Open-Textured  Predicates" (1991) Proceedings Third International Conference on  Artificial Intelligence and Law 137 at p.138.

  110  LT McCarty & NS Sridharan, "The Representation of an Evolving  System of Legal Concepts II: Prototypes and Deformations" (1987)  Proceedings of the Seventh International Joint Conference on  Artificial Intelligence 246.

  111  TJM Bench-Capon & MJ Sergot, "Toward a Rule-Based Representation  of Open Texture in Law" Chapter Six in C Walter (ed.), Computer Power  and Legal Language (Quorum: London, 1988) at p.47.

  112  SS Weiner, "Reasoning About "Hard" Cases in Talmudic Law" (1987)  Proceedings First International Conference on Artificial Intelligence  and Law

  222 at p.223.

  113  P Leith, "Logic, Formal Models and Legal Reasoning" (1984)  Jurimetrics Journal p.334 at p.356.

  114  KA Lambert & MH Grunewald, "LESTER: Using Paradigm Cases in a  Quasi-Prcedential Legal Domain" (1989) Proceedings Second  International Conference on Artificial Intelligence and Law 87.

  115  J Popple, "Legal Expert Systems: The Inadequacy of a Rule-based  Approach" (1991) 23 Australian Computer Journal 11 at p.15.

  116  Also note the GREBE system; q.v. LK Branting, "Representing and  Reusing Explanations of  Legal Precedents" (1989) Proceedings Second  International Conference on Artificial Intelligence and Law 103.

  117  RA Kowalski, "Case-based Reasoning and the Deep Structure  Approach to Knowledge Representation" (1991) Proceedings Third  International Conference on Artificial Intelligence and Law 21 at  p.23.

  118  KD Ashley & EL Rissland, "Waiting on Weighting: a Symbolic Least  Commitment Approach" (1988) Proceedings American Association for  Artificial Intelligence.

  119  MJ Sergot, "The Representation of Law in Computer Programs",  Chapter One in TJM Bench-Capon, Knowledge-Based Systems and Legal  Applications (Academic Press: London, 1991) at p.65.

  120  J Zeleznikow, Building Intelligent Legal Tools - The IKBALS  Project (1991) 2 Journal of Law and Information Science 165 at p.173.

  121  EL Rissland & KD Ashley, "A Case-Based System for Trade Secrets  Law" (1987) Proceedings First International Conference on Artificial  Intelligence and Law

  60.

  122  M Betzer, "Legal Resoning in 3-D" (1987) Proceedings First  International Conference on Artificial Intelligence and Law

  155 at  p.155.

  123  RA Kowalski, "Case-based Reasoning and the Deep Structure  Approach to Knowledge Representation" (1991) Proceedings Third  International Conference on Artificial Intelligence and Law 21.

  124  RA Kowalski, "Case-based Reasoning and the Deep Structure  Approach to Knowledge Representation" (1991) Proceedings Third  International Conference on Artificial Intelligence and Law 21 at  p.26.

  125  G Greenleaf, A Mowbray & AL Tyree, "Expert Systems in Law: The  DATALEX Project" (1987) Proceedings First International Conference on  Artificial Intelligence and Law 9 at p.12.

  126  e.g. SR Goldman, MG Dyer & M Flowers, "Precedent-based Legal  Reasoning and Knowledge Acquisition in Contract Law: a Process Model"  (1987) Proceedings First International Conference on Artificial  Intelligence and Law

  210; and MT MacCrimmon, "Expert Systems in  Case-Based Law: The Hearsay Rule Adviser" (1989) Proceedings Second  International Conference on Artificial Intelligence and Law 68.

  127  G Vossoss, J Zeleznikow & T Dillon, "Combining Analogical and  Deductive Reasoning in Legal Knowledge Base Systems - IKBALS II" in  Cv Noortwijk, AHJ Schmidt & RGF Winkels (eds), Legal Knowledge Based  Systems: Aims for Research and Development (Koninklijke: Lelystad,

  1991) p.97 at p.100.

  128  The weighting scheme used by Kowalski was:  Highest level court = 70; appeal level court = 50; trial level court  = 30.  Add 10 points for trial or appeals local to the jurisdiction.  Deduct 15 points for foreign jurisdictions, except England, then 10  points.  Add 1 to 5 points if case is recent: 1986 = 1 to 1990 = 5.  See RA Kowalski, Case-based Reasoning and the Deep Structure Approach  to Knowledge Representation (1991) Proceedings Third International  Conference on Artificial Intelligence and Law 21. 

  129  G Vossos, T Dillon & J Zeleznikow, "The Use of Object Oriented  Principles to Develop Intelligent Legal Reasoning Systems" (1991) 23  Australian Computer Journal 2.

  130  J Zeleznikow & D Hunter, "Rationales for the Continued  Development of Legal Expert Systems" (1992) 3 Journal of Law and  Information Science 94 at pp.102-103.

  131  RE Susskind, "Expert Systems in Law: A Jurisprudential approach  to Artificial Intelligence and Legal Reasoning" (1986) 49 Modern Law  Review 168 at p.171; see also RE Susskind, Expert Systems in Law: A  Jurisprudential Inquiry (Clarendon Press: Oxford, 1987) p.20.

  132  RA Kowalski, "Case-Based Reasoning and the Deep Structure  Approach to Knowledge Representation" (1991) Proceedings Third  International Conference on Artificial Intelligence and Law 21 at  p.21.

  133  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987)  pp.81-82.

  134  Ibidem  pp.21-23.

  135  TJM Bench-Capon & J Forder, "Knowledge Representation for Legal  Applications" Chapter Twelve in TJM Bench-Capon, Knowledge-Based  Systems and Legal Applications (Academic Press: London, 1991) at  p.249.

  136  HJ Levesque & RJ Brachman, "A Fundamental Tradeoff in Knowledge  Representation and Reasoning" Chapter Four in RJ Brachman & HJ  Levesque (eds), Readings in Knowledge Representation (Morgan  Kaufmann: Los Altos, 1985) at pp.66-67.

  137  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.49.

  138  q.v. TJM Bench-Capon & F Coenen, "Exploiting Isomorphism:  Development of a KBS to Support British Coal Insurance Claims" (1991)  Proceedings Third International Conference of Artificial Intelligence  and Law 62.

  139  RN Moles, "Logic Programming - An Assessment of Its Potential  for Artificial Intelligence Applications in Law" (1991) 2 Journal of  Law and Information Science 137 at p.144.

  140  Bench-Capon's own words; q.v. TJM Bench-Capon & J Forder,  "Knowledge Representation for Legal Applications" Chapter Twelve in  TJM Bench-Capon, Knowledge-Based Systems and Legal Applications  (Academic Press: London, 1991) at p.259.

  141  e.g. C Biagioli, P Mariani & D Tiscornia, "ESPLEX: a Rule and  Conceptual Based Model for Representing Statutes" (1987) Proceedings  First International Conference on Artificial Intelligence and Law 

  240, and the examples previously cited.

  142  e.g. DM Sherman, "A Prolog Model of the Income Tax Act of  Canada" (1987) Proceedings First International Conference on  Artificial Intelligence and Law

  127; also note TAXMAN and  like  projects cited.

  143  B Niblett, "Computer Science and Law: An Introductory  Discussion" in B Niblett (ed.), Computer Science and Law (Cambridge  University Press: Cambridge, 1980) at pp.16-17.

  144  For instance into ANF (Atomically Normalised Form) used with the  CCLIPS system (Civil Code Legal Information Processing System); q.v.  G Cross, CGde Bossonet, T Bradshaw, G Durham, R Gupta & M Nasiruddin,  "The Implementation of CCLIPS" Chapter Nine in C Walter (ed.),  Computer Power and Legal Language (Quorum: London, 1988) p.90.

  145  JP Dick, "Conceptual Retrieval and Case Law" (1987) Proceedings  First International Conference on Artificial Intelligence and Law 

  106 at p.109; although such material may classify as a source of  heuristics.  Susskind does not address this, however.

  146  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987)  pp.84-85;  citing HLA Hart, The  Concept of Law (Clarendon Press: Oxford, 1961) p.131.

  147  KE Sanders, "Representing and Reasoning About Open-Textured  Predicates" (1991) Proceedings Third International Conference on  Artificial Intelligence and Law 137 at p.142.

  148  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.61; contra G Greenleaf, A Mowbray &  AL Tyree, "Expert Systems in Law: The DATALEX Project" (1987)  Proceedings First International Conference on Artificial Intelligence  and Law 9.

  149  c.f. LB Solum, "On the Indeterminacy Crisis: Critiquing Critical  Dogma" (1987) 54 University of Chicago Law Review 462.

  150  J Boyle, "Anatomy of a Torts Class" (1985) 34 American  University Law Review 131; see also M Kelman, "Trashing" (1984) 36  Stanford Law Review 293.

  151  A Mason, "Future Directions in Australian Law" (1987) 13 Monash  Law Review 149, particularly at pp.154-155 and p.158; FG Brennan,  "Judicial Method and Public Law" (1979) 6 Monash Law Review 12; and M  McHugh, "The Law-making Function of the Judicial Process" (1988) 62  Australian Law Journal 15.

  152  e.g. JC Smith and C Deedman, "The Application of Expert Systems  Technology to Case-Based Reasoning" (1987) Proceedings First  International Conference on Artificial Intelligence and Law

  84.

  153  E Levi, An Introduction to Legal Reasoning (University of  Chicago Press: Chicago, 1949) pp.3-5.

  154  AL Goodhart, "The Ratio Decidendi of a Case" (1930) 40 Yale Law  Journal 161.

  155  J Stone, "The Ratio of the Ratio Decidendi" in Lord Lloyd & MDA  Freeman, Lloyd's Introduction to Jurisprudence (5th Ed.) (Stevens:  London, 1985) p.1164.

  156  "It is unclear".

  157  S Strvmholm, Rdtt. rottskollor och rottssystem (3rd Ed.)  (Norstedts: Stockholm, 1987) cited by P Wahlgren, "Legal Reasoning -  A Jurisprudence Description" (1989) Proceedings Second International  Conference on Artificial Intelligence and Law 147 at p.148.

  158  TJM Bench-Capon & F Coenen, "Practical Application of KBS to  Law: The Crucial Role of Maintenance" in Cv Noortwijk, AHJ Schmidt &  RGF Winkels (eds), Legal Knowledge Based Systems: Aims for Research  and Development (Koninklijke: Lelystad, 1991) p.5.

  159  P Bratley, J Frimont, E Mackaay & D Poulin, "Coping with Change"  (1991) Proceedings Third International Conference on Artificial  Intelligence and Law 69.

  160  e.g. P Hammond, "Representation of DHSS Regulations as a Logic  Program" (1983) Proceedings of the 3rd British Computer Society  Expert Systems Conference 225; and the Estate Planning System; q.v.  DA Schlobohm & DA Waterman, "Explanation for an Expert System that  Performs Estate Planning" (1987) Proceedings First International  Conference on Artificial Intelligence and Law

  18.

  161  RA Kowalski & MJ Sergot, "The Use of Logical Models in Legal  Problem Solving" (1990) 3 Ratio Juris 201 at p.207.

  162  DA Schlobohm & LT McCarty, "EPS II: Estate Planning With  Prototypes" (1989) Proceedings Second International Conference on  Artificial Intelligence and Law 1.

  163  P Bratley, J Frimont, E Mackaay & D Poulin, "Coping with Change"  (1991) Proceedings Third International Conference on Artificial  Intelligence and Law 69.

  164  AvdL Gardner, "Representing Developiong Legal Doctrine" (1989)  Proceedings Second International Conference on Artificial  Intelligence and Law 16 at p.21.

  165  Ibidem p.19.

  166  A narrow definition of "information" is a common criticism of  modern expert systems; q.v. HL Dreyfus & SE Dreyfus, Mind over  Machine (Basil Blackwell: Oxford, 1986); T Roszak, "The Cult of  Information" (Pantheon: London, 1986); DR Hofstadter, Metamagical  Themas (Penguin Press: London, 1985); and T Winograd & F Flores,  Understanding Computers and Cognition: A New Foundation for Design  (Ablex: Norwood, 1986).

  167  RN Moles, "Logic Programming - An Assessment of Its Potential  for Artificial Intelligence Applications in Law" (1991) 2 Journal of  Law and Information Science 137 at p.144.

  168  C Biagioli, P Mariani & D Tiscornia, "ESPLEX: a Rule and  Conceptual Based Model for Representing Statutes" (1987) Proceedings  First International Conference on Artificial Intelligence and Law 

  240 at  p.241.

  169  C Smith & C Deedman, "The Application of Expert Systems  Technology to Case-Based Reasoning" (1987) Proceedings First  International Conference on Artificial Intelligence and Law

  84 at  p.87.

  170  RE Susskind, "Expert Systems in Law - Out of the Research  Laboratory and into the Marketplace" (1987) Proceedings First  International Conference on Artificial Intelligence and Law 1 at p.2.

  171  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.78.

  172  JC Smith & C Deedman, "The Application of Expert Systems  Technology to Case-Based Reasoning" (1987) Proceedings First  International Conference on Artificial Intelligence and Law

  84 at  p.85.

  173  Approximately 50 texts and 100 articles; q.v. RE Susskind,  "Expert Systems in Law - Out of the Research Laboratory and into the  Marketplace" (1987) Proceedings First International Conference on  Artificial Intelligence and Law 1 at p.2;  note the similarity to the  issue of epistemology of law.

  174  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.27.

  175  For instance, Dworkin's "principles"; q.v. RM Dworkin, Taking  Rights Seriously (Duckworth: London, 1977), RM Dworkin, A Matter of  Principle (Harvard University Press: London, 1985), and RM Dworkin,  Law's Empire (Fontana: London, 1986).

  176  RE Susskind, "Expert Systems in Law - Out of the Research  Laboratory and into the Marketplace" (1987) Proceedings First  International Conference on Artificial Intelligence and Law 1 at p.3.

  177  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.254.

  178  B Niblett, "Expert Systems for Lawyers" (1981) 29 Computers and  Law 2 at p.3.

  179  PJ Hayes, "On the Differences Between Psychology and Artificial  Intelligence" in M Yazdani & A Narayanan, Artificial Intelligence:  Human Effects (Ellis Horwood: London, 1984) p.158.

  180  DR Hofstader, Gvdel, Escher, Bach: An Eternal Golden Braid  (Harvester Press: New York, 1979) p.578.

  181  RA Kowalski, "Leading Law Students to Uncharted Waters and  Making them Think: Teaching Artificial Intelligence and Law" (1991) 2  Journal of Law and Information Science 185 at p.187 nt.5.

  182  D Brown, "The Third International Conference on Artificial  Intelligence and Law: Report and Comments" (1991) 2 Journal of Law  and Information Science 233 at p.238.

  183  B Niblett, "Expert Systems for Lawyers" (1981) 29 Computers and  Law 2 at p.3.

  184  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.7.

  185  P Leith, "Clear Rules and Legal Expert Systems" in AA Martino &  F Socci (eds), Automated Analysis of Legal Texts (North-Holland:  Amsterdam, 1986) p.661; and P Leith, "Fundamental Errors in Legal  Logic Programming" (1986) 3  The Computer Journal 29.

  186  LT McCarty, "Some Requirements for a Computer-based Legal  Consultant" (Research Report: Rutgers University, 1980) at pp.2-3,  cited in RN Moles, Definition and Rule in Legal Theory: A  Reassessment of HLA Hart and the Positivist Tradition (Basil  Blackwell: Oxford, 1987) p.269; emphasis added.

  187  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987)  p.53.

  188  MA Boden, Artificial Intelligence and Natural Man (Basic Books:  New York, 1977), cited in D Partridge, "Social Implications of  Artificial Intelligence" Chapter Thirteen in M Yazdani (ed.),  Artificial Intelligence: Principles and  Applications (Chapman &  Hall: London,

  1986) at p.326.  See also D Partridge, Artificial  Intelligence: Applications in the Future of Software Engineering  (Ellis Horwood, Chichester, 1986).

  189  RN Moles, "Logic Programming: An Assessment of its Potential for  Artificial Intelligence Applications in Law" (1991) 2 Journal of Law  and Information Science 137 at p.161.

  190  TJM Bench-Capon, "Deep Models, Normative Reasoning and Legal  Expert Systems" (1989) Proceedings Second International Conference on  Artificial Intelligence and Law 37 at p.37.

  191  Glauoma diagnosis system.

  192  LT McCarty, "Intelligent Legal Information Systems: Problems and  Prospects" in CM Campbell (ed.), Data Processing and the Law (Sweet &  Maxwell: London, 1984) p.126.

  193  SS Weiner, "Reasoning About "Hard" Cases in Talmudic Law" (1987)  Proceedings First International Conference on Artificial Intelligence  and Law

  222 at p.223.

  194  MJ Sergot, HT Cory, P Hammond, RA Kowalski, F Kriwacek & F  Sadri, "Formalisation of the British Nationality Act" (1986) 2  Yearbook of Law Computers and Technology; and TJM Bench-Capon, GO  Robinson, TW Routen & MJ Sergot, "Logic Programming for Large Scale  Applications in Law" (1987) Proceedings First International  Conference on Artificial Intelligence and Law

  190.

  195  JC Smith & C Deedman, "The Application of Expert Systems  Technology to Case-Based Reasoning" (1987) Proceedings First  International Conference on Artificial Intelligence and Law

  84 at  p.85.

  196  RA Kowalski, Case-based Reasoning and the Deep Structure  Approach to Knowledge Representation (1991) Proceedings Third  International Conference on Artificial Intelligence and Law 21 at  p.22.

  197  LT McCarty & NS Sridharan, "The Representation of an Evolving  System of Legal Concepts II: Prototypes and Deformations" (1987)  Proceedings of the Seventh International Joint Conference on  Artificial Intelligence 246 at p.250.

  198  D Makinson, "How to Give it Up: A Survey of Some Formal Aspects  of the Logic of Theory Change" (1985) 62 Synthise 347.

  199  P Bratley, J Frimont, E Mackaay & D Poulin, "Coping with Change"  (1991) Proceedings Third International Conference on Artificial  Intelligence and Law 69 at p.74.

  200  See RM Dworkin, Law's Empire (Fontana: London, 1986) pp.250-254  on the difficulty in "compartmentalization" of the law.

  201  LT McCarty, "Some Requirements for a Computer-based Legal  Consultant" (Research Report: Rutgers University, 1980) cited in MJ  Sergot, "The Representation of Law in Computer Programs", Chapter One  in TJM Bench-Capon, Knowledge-Based Systems and Legal Applications  (Academic Press: London, 1991) at pp.46-47.

  202  P Leith, "Logic, Formal Models and Legal Reasoning" (1984)  Jurimetrics Journal p.334 at p.356.

  203  NE Simmonds, "Between Positivism and Idealism" (1991) 50  Cambridge Law Journal 308 at pp.312-313.

  204  M Minsky, "A Framework for Representing Knowledge" in J  Haugeland (ed.), Mind Design (MIT Press: Cambridge, 1981) p.95 at  p.100.

  205  It is also ironic in light of Hart's alleged reliance on  Wittgenstein's linguistic philosophy; q.v. Cotterrell, The Politics  of Jurisprudence: A Critical Introduction to Legal Philosophy  (Butterworths: London, 1989) pp.89-90.

  206  J Vaux, "AI and Philosophy: Recreating Naive Epistemology"  Chapter Seven in KS Gill (ed.), Artificial Intelligence for Society  (John Wiley & Sons: London, 1986) p.76.; q.v. L Wittgenstein,  Philosophical Investigations (Basil Blackwell: London, 1953).

  207  RA Kowalski & MJ Sergot, "The Uses of Logical Models in Legal  Problem Solving" (1990) 3 Ratio Juris 201 at p.205.

  208  See HLA Hart, "Definition and Theory in Jurisprudence" (1954) 70  Law Quarterly Review 37.

  209  L Fuller, "Positivism and Fidelity to Law - A Reply to Professor  Hart" (1958) 71 Harvard Law Review 630 at p.666.

  210  Raz's approach to adjudication shares similar characteristics;  q.v. J Raz, "The Problem about the Nature of Law" (1983) 31  University of Western Ontatio Law Review 202 at pp.213-216.

  211  HLA Hart, The Concept of Law (Clarendon Press: Oxford, 1961)  p.126.

  212  KR Popper, Conjectures and Refutations (4th Ed.) (Routledge and  Kegan Paul: London, 1972) p.46.

  213  JW Harris, Law and Legal Science (Clarendon Press: Oxford, 1979)  p.166.

  214  M Polyani, Personal Knowledge - Towards a Post-Critical  Philosophy (Routledge and Kegan Paul: London, 1958).

  215  DC Berry, "The Problem of Implicit Knowledge" (1987) 4 Expert  Systems 144.

  216  DC Berry & A Hart, "The Way Forward" in DC Berry & A Hart (eds)  Expert Systems: Human Issues  (MIT: Cambridge, 1990) p.256.

  217  E Husserl, Cartesian Meditations (Martinus Nijhoff: The Hague,

  1960) pp.54-55.

  218  B MacLennan, "Logic for the New AI" in JH Fetzer (ed.), Aspects  of Artificial Intelligence (Kluwer: Dordrecht, 1988) at p.163.

  219  C Hempel, Philosophy of Natural Science (Prentice Hall: London,

  1966).

  220  A Narayanan, "Why AI Cannot be Wrong" Chapter Five in KS Gill  (ed.), Artificial Intelligence for Society (John Wiley & Sons:  London, 1986) at p.48.

  221  P Davies, "Living in a Non-Maaterial World - the New Scientific  Consciousness" (1991) The Australian  (9th October) pp.18-19 at p.19.

  222  RS Pound, "Mechanical Jurisprudence" (1908) 8 Columbia Law  Review 605.

  223  J Searle, "Minds, Brains and Programs" (1980) 3 Behavioural and  Brain Sciences 417.

  224  RE Susskind, Expert Systems in Law: A Jurisprudential Inquiry  (Clarendon Press: Oxford, 1987) p.241.

  225  RHS Tur, "Positivism, Principles, and Rules" in E Atwool (ed.),  Perspectives in Jurisprudence (University of Glascow Press: Glascow,

  1997)  at p.51.

  226  EL Rissland & DB Skalak, "Interpreting Statutory Predicates"  (1989) Proceedings Second International Conference on Artificial  Intelligence and Law 46 at p.46.

  227  MJ Detmold, "Law as Practical Reason" (1989) 48 Cambridge Law  Journal 436 at p.460.

  228  Ibidem p.439.

  229  L Fuller, "Positivism and Fidelity to Law - A Reply to Professor  Hart" (1958) 71 Harvard Law Review 630 at p.663; see also RS Summers,  "Professor Fuller on Morality and Law" in RS Summers (ed.), More  Essays on Legal Philosophy: General Assessment of Legal Philosophies  (Basil Blackwell: Oxford, 1971) at pp.117-119.

  230  F Schauer, Playing by the Rules: A Philosophical Examination of  Rule-based Decision-Making in Law and in Life (Clarendon Press:  Oxford, 1991) at pp.59-60.

  231  HLA Hart, The Concept of Law (Clarendon Press: Oxford, 1961)  p.56.

  232  H Williamson, "Some Implications of Acceptance of Law as  Rule  Structure" (1967) 3 Adelaide Law Review

  18 at pp.42-43.

  233  c.f. A Glass, "Interpretive Practices in Law and Literary  Criticism" (1991) 7 Australian Journal of Law & Society

  16.

  234  DN Herman, "Phenomonology, Structuralism, Hermeneutics, and  Legal Study: Applications of Contemporary Continental Though to Legal  Phenomena" (1982) 36 University of Miami Law Review 379.

  235  P Linzer, "Precise meaning and Open Texture in Legal Writing and  Reading" Chapter Two in C Walter (ed.), Computer Power and Legal  Language (Quorum: London, 1988).

  236  M Weait, "Swans Reflecting Elephants: Imagery and the Law"  (1992) 3 Law and Critique 59 at p.66.

  237  P Gabel & P Harris, "Building Power and breaking Images:  Critical Legal Theory and the Practice of Law" (1982-83) 11 Review of  Law & Social Change 369 at p.370.

  238  HL Dreyfus, "From Micro-Worlds to Knowledge Representation: AI  at an Impasse" in J Haugeland (ed.), Mind Design (MIT Press:  Cambridge, 1981) p.161 at p.170.

  239  DG Bobrow & T Winograd, "An Overview of KRL, A Knowledge  Representation Language" (1977) 1 Cognitive Science 3 at p.32.

  240  C Fried, "Sonnet LXV and the 'Black Ink' of the Framer's  Intention" (1987) 100 Harvard Law Review 751 at pp.757-758.

  241  J Weizenbaum, Computer Power and Human Reason: From Judgment to  Calculation (WH Freeman & Co: San Francisco, 1976) cited in D  Partridge, "Social Implications of Artificial Intelligence" Chapter  Thirteen in M Yazdani (ed.), Artificial Intelligence: Principles and   Applications (Chapman & Hall: London,

  1986) at pp.330-331.

  242  Contra note MA Boden, "AI and Human Freedom" in M Yazdani & A  Narayanan (eds), Artificial Intelligence: Human Effects (Ellis  Horwood: Chichester, 1984).

  243  C Tapper, "Lawyers and Machines" (1963) 26 Modern Law Review

  121.

  244  Ibidem p.128.

  245  c.f. TJM Bench-Capon, "Deep Models, Normative Reasoning and  Legal Expert Systems" (1989) Proceedings Second International  Conference on Artificial Intelligence and Law 37 at p.42.

  246  J Zeleznikow, "Building Intelligent Legal Tools - The IKBALS  Project" (1991) 2 Journal of Law and Information Science 165.

  247  e.g. DE Wolstenholme, "Amalgamating Regulation and Case-based  Advice Systems through Suggested Answers" (1989) Proceedings Second  International Conference on Artificial Intelligence and Law 63.

  248  c.f. R Wright, "The Cybernauts have Landed" (1991) Law Institute  Journal 490 at p.491.

  249  C Tapper, "Lawyers and Machines" (1963) 26 Modern Law Review 121  at p.126.

  250  q.v. PJ Ward, "Computerisation of Legal Material in Australia"  (1982) 1 Journal of Law and Information Science 162.

  251  J Bing, "The Text Retrieval System as a Conversation Partner" in  C Arnold (ed.) Yearbook of Law, Computers and Technology  (Butterworths: London, 1986) p.25.

  252  G Greenleaf, "Australian Approaches to Computerising Law -  Innovation and Integration" (1991) 65 Australian Law Journal 677.

  253  SJ Latham, "Beyond Boolean Logic: Probabilistic Approaches to  Text Retrieval" (1991) 22 The Law Librarian 157.

  254  J Bing, "Legal Text Retrieval Systems: The Unsatisfactory State  of the Art" (1986) 2 Journal of Law and Information Science 1 at  pp.16-17.

  255  RM Tong, CA Reid, GJ Crowe & PR Douglas, "Conceptual Legal  Document Retrieval Using the RUBRIC System" (1987) Proceedings First  International Conference on Artificial Intelligence and Law

  28; and  J Bing, "Designing Text Retrieval Systems for 'Conceptual Searching'"  (1987) Proceedings First International Conference on Artificial  Intelligence and Law

  43.

  256  J Kolodner, "Maintaining Organisation in a Dynamic Long-Term  Memory" (1983) 7 Cognitive Science;  CD Hafner, "Conceptual  Organisation of Case Law Knowledge Bases" (1987) Proceedings First  International Conference on Artificial Intelligence and Law 35.

  257  T Mitchell, "Learning and Problem Solving" (1983) Proceedings of  International Joint Conference on Artificial Intelligence.

  258  DE Rose & RK Belew, "Legal Information Retrieval: A Hybrid  Approach" (1989) Proceedings Second International Conference on  Artificial Intelligence and Law 138.

  259  CD Hafner, "Conceptual Organisation of Case Law Knowledge Bases"  (1987) Proceedings First International Conference on Artificial  Intelligence and Law 35.

  260  LT McCarty, "On the Role of Prototypes in Appellate Legal  Argument" (1991) Proceedings Third International Conference on  Artificial Intelligence and Law

  185 at p.186.

  261  TJM Bench-Capon & MJ Sergot, "Toward a Rule-Based Representation  of Open Texture in Law" Chapter Six in C Walter (ed.), Computer Power  and Legal Language (Quorum: London, 1988) at p.58.

  262   KD Ashley, "Toward a Computational Theory of Arguing with  Precedents: Accomodating Multiple Interpretations of Cases" (1989)  Proceedings Second International Conference on Artificial  Intelligence and Law 99; and KD Ashley & EL Rissland, "But See,  Accord: Generating 'Blue Book' Citations in HYPO" (1987) Proceedings  First International Conference on Artificial Intelligence and Law 

  67.

  263  EL Rissland, "Examples in Legal Reasoning: Legal Hypotheticals"  (1983) Proceedings Eighth International Joint Conference on  Artificial Intelligence 90;  EL Rissland & EM Soloway, "Overview of  an Example Generation System" (1980) Proceedings First Annual  National Conference on Artificial Intelligence; and EL Rissland, EM  Valcarce & KD Ashley, "Explaining and Arguing with Examples" (1984)  Proceedings National Conference on Artificial Intelligence.

  264  CC Marshall, "Representing the Structure of a Legal Argument"  (1989) Proceedings Second International  Conference on Artificial  Intelligence and Law 121.  On the structure of legal argument see S  Toulmin, The Uses of Argument (Cambridge University Press: Cambridge,

  1958); S Toulmin, RD Reike, & A Janik, An Introduction to Reasoning  (MacMillan Press: New York, 1979); and C Perelman, The Idea of  Justice and the Problem of Argument (Routledge & Kegan Paul: London,

  1963).

  265  KD Ashley & EL Rissland, "Toward Modelling Legal Argument" in AA  Martino & F Socci (eds), Automated Analysis of Legal Texts  (North-Holland: Amsterdam, 1986) at p.19; also KD Ashley, "Toward a  Computational Theory of Arguing with Precedents" (1989) Proceedings  Second International Conference on Artificial Intelligence and Law

  93.

  266  EL Rissland, "Learning How to Argue: Using Hypotheticals" (1984)  Proceedings First Annual Conference on Theoretical Issues in  Conceptual Information Processing; EL Rissland, "Argument Moves and  Hypotheticals" in C Walter (ed.), Computing Power and Legal Reasoning  (West Publishing: St Paul, 1985).

  267  RH Michaelson, "An Expert System for  Federal Tax Planning"  (1984) 1 Expert Systems 2.

  268  e.g. the Retirement Pension Forecast and Advice System (relying  on the Aion Development System shell) q.v. S Springel-Sinclair & G  Trevena, "The DHSS Retirement Pension Forecast and Advice System" in  P Duffin (ed.) Knowledge Based Systems: Applications in  Administrative Government (Ellis Horwood: Chichester, 1988).

  269  G De Jong, "Towards a Model of Conceptual Knowledge Acquisition  Through Directed Experimentation" (1983) Proceedings of International  Joint Conference on Artificial Intelligence.

  270  Legal Decision-making System; q.v. DA Waterman & MA Peterson,  "Rule-based Models of Legal Expertise" (1980)  Proceedings First  Annual National Conference on Artificial Inelligence 272; and DA  Waterman & MA Peterson, "Evaluating Civil Claims: An Expert Systems  Approach" (1984) Expert Systems 1.

  271  q.v. DA Waterman, RH Anderson, F Hayes-Roth, P Klahr, G Martins  & SJ Rosenschein, Design of a Rule-Oriented System for Implementing  Expertise (Rand Corporation: Santa Monica, 1979).

  272  System for Asbestos Litigation; q.v. DA Waterman, J Paul & MA  Peterson, "Expert Systems for Legal Decision Making" (1986) 4 Expert  Systems 212.

  273  G Greenleaf, "Australian Approaches to Computerising Law -  Innovation and Integration" (1991) 65 Australian Law Journal 677 at  p.679.

  274  SS Nagel & R Barczyk, "Can Computers Aid the Dispute Resolution  Process?" (1988) 71 Judicature 253.

  275  q.v. WM Bain, Toward a Model of Subjective Interpretation  (Department of Commerce Research Report: Yale University, 1984) cited  in MJ Sergot, "The Representation of Law in Computer Programs"  Chapter One in TJM Bench-Capon (ed.), Knowledge-Based Systems and  Legal Applications (Academic Press: London, 1991) p.16.

  276  S Torrance, "Breaking out of the Chinese Room" in M Yazdani  (ed.), Artificial Intelligence: Principles and  Applications (Chapman  & Hall: London,

  1986) at p.301.

  277  TW Bynum, "Artificial Intelligence, Biology, and Intentional  States" (1985) 16 Metaphilosophy 355.

  278  T Cuda, "Against Neural Chauvanism" (1985) 48 Philosophical  Studies 111.

  279  DE Rumelhart, JL McClelland and the PDP Research Group, Parallel  Distributed Processing: Explorations in the Microstructure of  Cognition (MIT Press: Cambridge, 1986).

  280  AL Tyree, "The Logic Programming Debate" (1992) 3 Journal of Law  and Information Science 1111 at p.115.

  281  q.v. RA Clarke, Knowledge-Based Expert Systems: Risk Factors and  Potentially Profitable Application Areas(Working paper: Department of  Commerce, Australian National University, 1988).

  282  M Aultman, "Technology and the End of Law" (1972) 17 American  Journal of Jurisprudence 46 at pp.49-52.