This essay briefly describes SoftLaw Corporation's approach to the use of rule-based systems in public administration before considering the relevance to that approach of the types of criticisms raised by Moles in his paper on logic programming.FN1 The essay then considers other possible criticisms of the approach taken by SoftLaw Corporation.
SoftLaw Corporation proposes the use of rulebase technology in the administration of legislation to achieve accuracy and consistency in a complex or volatile policy environment.FN2 SoftLaw Corporation regards rulebase technology as peculiarly appropriate to the administration of legislation because it asserts that rulebase systems can reliably automate the structural logic of sets of rules such as those contained in legislation.FN3
SoftLaw Corporation considers that the administration of the application of rules or legislation involves two types of interpretative skill: one which deals with logical intricacy and one which deals with legal interpretation or judgment. An expressed assumption by SoftLaw Corporation is that the complexity of legislation largely derives from its logical intricacy.FN4
While SoftLaw Corporation believes that rulebase technology can reliably automate logic processes, it concedes that legal interpretation or judgment cannot be reliably automated, although it believes that this is a general human skill which it can aid by the provision of explanatory material, elaboration of the meaning of a phrase, summaries of precedents and so on.FN5 SoftLaw Corporation's avowed position is that by using rulebase technology, complex rules or legislation can be reliably, correctly and consistently applied by people who do not have a clue what the rules are or legislation is.FN6
Rulebase technology in this context involves creating a knowledgebase by modelling the rules in the legislation in coherent English sentences which more or less correspond with the sections or paragraphs in the source legislation. This correspondence between the rules and the source legislation is known as isomorphism. Isomorphism is an important methodology because it facilitates the maintenance and debugging of a knowledgebase.
The knowledgebase comprises a set of IF-THEN type rules using logical operators (eg AND, OR) to identify the conditions necessary to reach the logical conclusion. For example, the definition of "work" from sec 10(1) of the Copyright Act 1968 (Cth)FN7 could be modelled as follows:
IF (the work is a literary work OR the work is a dramatic work OR the work is a musical work OR the work is an artistic work) THEN the work is a work under sec 10(1) of the Copyright Act.
A computer program, known as an inference engine, is then responsible for obtaining the relevant facts from the user by using a grammar parser to construct intelligent questions from the rules in the knowledgebase and from the answers to any previous questions. The inference engine matches the facts which it obtains against the conditions to determine a logical conclusion (if any) or ask further intelligent questions. This matching methodology is known as "forward chaining". A related matching methodology, known as "backward chaining", starts with the conclusion and works backwards to find the conditions necessary to support that conclusion.
Moles would criticise the SoftLaw Corporation approach, based as it is on logic programming, on the grounds set out below.
Moles does not believe that the reformulation of the legislation to suit the particular rulebase technology necessarily retains the meaning of the original legislation despite the claim that isomorphism achieves this result.
While his criticism appears amply justified in the case of the Alvey Programme, it would not appear to apply to the same extent in relation to SoftLaw Corporation's approach. This is because SoftLaw Corporation does not manipulate the source legislation to anything like the same extent as was done in the Alvey Programme where the source material was:
Instead, the process used by SoftLaw Corporation involves reformulating the legislation into rules by examining every line of the legislation and isolating each individual case that must be addressed.FN9 These rules are made up of coherent, complete English sentences which anyone can read and which correspond more or less directly with the source legislation.FN10 A grammar parsing program automatically converts the sentences to questions for investigation, and to negations and other forms for reporting.FN11
The example of sec 1 of the United Kingdom Domestic Violence and Matrimonial Proceedings Act 1976 which Moles uses clearly demonstrates that the meaning of a particular section in a statute may depend on a section or sections in another statute. Moles argues that this interdependence of statutes raises fundamental difficulties in representing statutes as systems of rules.FN12 Especially so where the rule builders are not lawyers and therefore might not realise the significance of other statutes in the context of a particular issue.
There are two aspects to this criticism: (1) the interdependence of statutes which creates the possibility that an attempt to take into account every possible related statute would overwhelm the technology and, in any event, would not be practical and (2) the knowledge of this interdependence.
The first aspect is somewhat spurious. Even a lawyer expert in a particular field of law would probably not claim to be intimately aware of every possible piece of legislation or interpretation of it which could be relevant in any particular fact situation. Indeed, by practical necessity, the comprehensiveness of any lawyer's research of any particular issue is limited by the cost (in both time and money) and by its imagined relevance. That is, a lawyer consciously or otherwise exercises judgment in determining how much research is necessary on any particular issue. The developer of a rulebase system would generally make a similar judgment which, provided it was communicated to users of the system, should not adversely affect the expected performance of the system.
While the probable lack of knowledge of the interdependence of statutes was a legitimate criticism of the approach being taken by the Imperial College Group, none of whom had any legal qualifications or experience, it would not appear to be justified in SoftLaw Corporation's case. SoftLaw Corporation claims to have a unique level of expertise in the specialist field of the administration of legislation and policy rules,FN13 being specialists in administrative and government law and using legal analysts, rather than software engineers, to create the rulebase.FN14
Moles criticises the virtue that rulebase technology builders see in the separation of the knowledgebase from the inference engine which applies the rules in that knowledgebase on the ground that it overlooks the consequentialist view.FN15 Consequentialism requires an evaluation of any potential application of the law in a way which will have regard to its political, moral or social acceptability. To the extent that an application of the law is not regarded as acceptable, it will lead to a reformulation of the law so as to result in an acceptable application. The B & Q Retail caseFN16 which Moles cites is a perfect example of the application of consequentialism. Moles concludes by questioning the value of individualised rule-oriented systems which cannot explain such outcomes.
Although SoftLaw Corporation's approach does not rely solely on the rulebase to provide an answer, as its approach also incorporates additional materials (eg policies, guidelines, case law) to help a user to make informed judgments on particular issues, it would not appear that there is any overt possibility of rejecting an outcome reached by the system. On the other hand, there is the covert possibility that the answers to the questions which the system asks could be manipulated so as to achieve what might be considered a desired outcome and this is acknowledged by SoftLaw Corporation, although SoftLaw recommends monitoring to detect such misuses of the system.FN17
Following on from his criticism of rulebase technology not being able to account for the feedback and reformulation which consequentialism requires or allows, Moles raises the question whether working from the representation of the law to the particular application of it, involves any mechanism, inferential or otherwise.
This criticism appears especially potent because the use of logical modelling is said to depend on this separation of the representation of the law from the inference mechanisms which apply it for a particular purpose.
These criticisms involve the fundamental jurisprudential question of how law should be characterised. If law should be characterised as a set of determinative rules,FN18 then the logical modelling of those rules can be justified. If, however, law should be characterised as indeterminate as Moles argues, then it becomes difficult to justify the logical modelling of it as rules.
The SoftLaw Corporation approach must necessarily characterise law as a set of determinative rules because it relies on the logical modelling of those rules. This is perhaps more justifiable in the field of public administration or bureaucracy, in which SoftLaw Corporation specialises, as Moles himself has conceded:FN19
[...] bureaucrats work within a limited decision-making environment and depend on someone else to make high level policy decisions for the organisational framework within which they work. In this situation, one is not just modelling knowledge from the source documents, but using the interpretive strategy of the client organisation as the key to decoding their meaning.
Moles notes the statements made by some engaged in logic programming that they model the unambiguous interpretation of legislation. He goes on to explain that the way in which they determine this unambiguous meaning is by conducting surveys to settle any ambiguity. He criticises this approach for determining the meaning of words as being inappropriate because he would rather know how the judge would interpret the meaning of a relevant statutory provision than the outcome of a survey of other people who do not have responsibility for determining the matter. FN20
In the approach taken by the SoftLaw Corporation, it would appear that the unambiguous interpretation used to model the legislative rules is that of the client organisation. To the extent that it is this organisation which has responsibility for determining the matter (subject to any administrative appeal by the person seeking the determination), Moles' criticism does not apply.
Moles criticises the proposition that the atomistic quality of legal rules is a virtue of isomorphism it is regarded as a virtue because it allows the correction of an error in, or an amendment of, the legislation by simply changing the one piece in the knowledgebase which corresponds with the piece in the source legislation. The essence of his criticism is that treating rules as atomistic entities cannot be reconciled with the fact that legislation is interdependent and that context is all important. However, the logic programmers must support the proposition because otherwise the discovery of an error, or the making of an amendment, would mean that the rulebase would have to be redone each time with the consequence that the problem of change would become insurmountable.FN21 Indeed, the whole purpose of isomorphism is to enable the easy maintenance and debugging of a rulebase.
SoftLaw Corporation's approach to rulebase technology is to use isomorphism to model legislation so that there is a more or less direct correspondence between the individual rules and the individual units of legislation.FN22 However, Moles' criticism of isomorphism -- that a change in one rule which is itself part of the context in which other rules occur will give rise to a change in the meaning of those other rules as well -- would not appear to be valid. This is because isomorphism is intended to facilitate that result. Each individual rule is examined by the inference engine in the context of the other rules, so that a change to one rule may well affect those other rules. The point is though, that you do not have to explicitly change those other rules. The change is isolated by virtue of isomorphism to one or more rules, but the effect of the change will apply to the whole rulebase to the extent that any other rule uses any of the rules which have been changed.
Moles doubts that one could encode the meaning of individual statutory rules without taking into account reported decisions on those rules. In particular, he criticises the assumption made by the Imperial College Group that recently enacted legislation was free of the influence of case law and points out that in the specific case with which they were concerned, it was of particular importance because the subject matter of the legislation had a long history in prior legislation.FN23
SoftLaw Corporation's approach to rulebase technology is not susceptible to this criticism because it does not rely on the legislation to the exclusion of other relevant material including case law and legislative history. For example, SoftLaw Corporation's rulebase solution for the Military Compensation and Rehabilitation Service (MCRS) comprised a rulebase that modelled the legislation, authoritative commentary on each issue raised by the legislation, the authoritative guide to court and tribunal decisions on the application of the legislation as well as MCRS policy and procedural material.FN24
Moles criticises the stance taken by the Imperial College Group that a purely logical model of any legislation can be made without the help of anyone who knows even the most basic of the approaches taken to the problem of statutory interpretation. Moles does not believe that an expert's only contribution to such an endeavour would be to change the detail of the content of the rules, but would extend to advice on the method of interpretation being employed and the way in which the knowledge was structured. Indeed, he suggests that a legal expert would have advised the Imperial College Group to abandon their program.FN25
This criticism involves two different aspects: (1) that an expert legal adviser is required in the development of a rulebase; (2) that if an expert legal adviser were consulted, he would advise the abandonment of the endeavour.
Clearly, an expert legal adviser is required in the development of any rulebase. SoftLaw Corporation recognises this as its approach is to operate within areas in which they have substantial domain experience and expertise. In particular, SoftLaw Corporation claim to be legal specialists in administrative and government law including the drafting and administration of legislation.FN26
In relation to the second aspect, Moles' assertion that if an expert legal adviser were consulted his advice would be to abandon the development of a rulebase is rather far fetched. At least in SoftLaw Corporation's case, it would appear that its expert legal advisers have not given this advice. Indeed, examples of successful rulebase technology implementations for various Government departments can be found on SoftLaw Corporation's web site.FN27
Other criticisms which I would make of the SoftLaw Corporation approach to implementing rulebase technology are set out below.
SoftLaw Corporation argues that while rulebase technology is a seriously deskilling device, it should be accepted in the same way as calculators and spreadsheets have been accepted in the past on the condition that if the input parameters are correct, then the result will generally be correct.FN28
Indeed, calculators and spreadsheets have now become ubiquitous but it is questionable whether that fact alone provides any serious justification for the resultant deskilling. Those of us who can still perform simple mental arithmetic are not infrequently presented with situations in which the operators of calculators and similar devices make mistakes. Those operators, lacking the skill of simple mental arithmetic are, as a result, generally not able to recognise that a mistake has been made. Is this desirable? While the consequences in this particular example of deskilling may not be all that significant, I would suggest that the consequences in the legal area may well be more significant.
SoftLaw Corporation's approach to rulebase technology suggests that while the logical intricacy of legislation can be automated, the classic legal interpretation of the words of that legislation cannot be automated.FN29 Yet SoftLaw Corporation also assert that rulebase technology makes it possible for a person who does not know a set of rules to apply those rules accurately. Where that person needs to exercise judgment, SoftLaw Corporation then asserts that this is a general human skill which can be aided by the provision of various explanatory material.FN30
I am not sure that anyone would seriously argue that statutory interpretation is a general human skill that could be accurately exercised by any member of the public or public service, especially a member who did not know the rules which provide the context for any such interpretation.
SoftLaw Corporation's approach to implementing rulebase technology in public administration involves working closely with the particular government organisation and using that organisation's policy documentation and interpretations of their legislation as explanatory information for use by staff when making judgments.
It must be recognised that there is at least a possibility that the influence of the government organisation in the creation of a rulebase in these circumstances could, intentionally or otherwise, result in a system which is biased in favour of that organisation's views on any issue which has the potential to be contentious. Indeed, SoftLaw Corporation encourages the provision of supporting material to departmental staff using rulebase technology so that for each legislative issue that can arise, there is clear guidance and assistance.FN31 Thus, one could perhaps be forgiven for thinking that while such implementations of rulebase technology may provide instant answers, they may not necessarily provide the correct answers.
Would any lawyer be prepared to trust an answer provided by rulebase technology? If not, then it would perhaps be somewhat difficult to argue that non-experts in the field of law should do so.
Given that lawyers are trained to be professional doubting Thomases -- no proposition can stand unless it can be supported by some authority, argument or reasoning -- it seems to me that to trust the answer being provided by a rulebase system, a lawyer would need to understand how the system reached that answer. However, if the lawyer understood how the system reached the answer, presumably the lawyer could reach that answer without the aid of the system, in which case, what is the point of the system?
I concede that SoftLaw Corporation would probably argue that the point would be to enable someone who did not know the rules to reach the same answer, but I would question whether in fact someone who did not know the rules could reach the same answer. Would the non-expert recognise an answer which was, at least to an expert, plainly incorrect?
The approach of SoftLaw Corporation to the implementation of rulebase technology is susceptible to some of the jurisprudential criticisms raised by Moles which plague those who believe that law is a set of determinative rules which can be logically modelled. Further, there are other criticisms as outlined above which can be levelled at SoftLaw Corporation's approach.
However, I would not advocate, as Moles does, the abandonment of the application of rulebase technology to the legal domain. Its application is not the magic bullet that some of its adherents would claim, as I do not believe it can replace domain experts with those who do not have a clue what the relevant rules are. Nonetheless, rulebase technology would appear to have practical value in assisting, or providing guidance to, domain experts in reaching a conclusion or validating a conclusion which they have otherwise reached. Indeed, such a system could provide some guidance to a domain expert who is dealing with an otherwise unfamiliar area of law.
FN1 "Logic programming - An Assessment Of Its Potential For Artificial Intelligence Applications In Law", Journal of Law and Information Science (1991) vol 2 no 2; also at http://law.uniserve.edu.au/law/pub/compute/logic.
FN2 "Electronic Service Delivery: Achieving Accuracy and Consistency in Complex Transactions", Peter Johnson, 1998, http://www2.austlii.edu.au/cal/secure/Johnson1.html.
FN3 Ibid.
FN4 Ibid.
FN5 Ibid.
FN6 "SoftLaw's Distinctive Multi-Disciplinary Talents - Applying Complex Rules", http://www.softlaw.com.au/appr/appr_3.htm.
FN7 "work" means a literary, dramatic, musical or artistic work.
FN8 "Logic programming - An Assessment Of Its Potential For Artificial Intelligence Applications In Law", supra.
FN9 "Best Practice in the Use of Rule-Based Systems", Peter Johnson and Surendra Dayal, 1998, http://www2.austlii.edu.au/cal/secure/Johnson2.html.
FN10 "SoftLaw's Distinctive Multi-Disciplinary Talents The SoftLaw Solution", http://www.softlaw.com.au/appr/appr_3.htm.
FN11 "Expert Systems Collecting the Right Data", http://www.softlaw.com.au/capa/capa_2.htm.
FN12 "Logic programming - An Assessment Of Its Potential For Artificial Intelligence Applications In Law", supra.
FN13 "The SoftLaw Approach Domain expertise", http://www.softlaw.com.au/appr/appr_1.htm.
FN14 "SoftLaw's Distinctive Multi-Disciplinary Talents The SoftLaw Difference", http://www.softlaw.com.au/appr/appr_3.htm; "Expert Systems An Expert System for Administering Rules", http://www.softlaw.com.au/capa/capa_2.htm.
FN15 "Logic programming - An Assessment Of Its Potential For Artificial Intelligence Applications In Law", supra.
FN16 [1984] 2 All ER 332.
FN17 "Best Practice in the Use of Rule-Based Systems", supra.
FN18 This view was proposed by the legal theorist HLA Hart in "Positivism and the Separation of Law and Morality", (1958) 79 Harvard Law Review 593.
FN19 "There Is More To Life Than Logic", Journal of Law and Information Science (1992) vol 3 no 2 pp 188-218; also at http://law.uniserve.edu.au/law/lectures/contract_nk/part15.html.
FN20 "Logic programming - An Assessment Of Its Potential For Artificial Intelligence Applications In Law", supra.
FN21 Ibid.
FN22 Guest lecture by Peter Johnson of SoftLaw Corporation, UNSW, 14 April 1999.
FN23 "Logic programming - An Assessment Of Its Potential For Artificial Intelligence Applications In Law", supra.
FN24 "Electronic Service Delivery: Achieving Accuracy and Consistency in Complex Transactions", supra.
FN25 Ibid.
FN26 "SoftLaw's Distinctive Multi-Disciplinary Talents The SoftLaw Difference", supra.
FN27 http://www.softlaw.com.au/corp/corp_4.htm.
FN28 "Electronic Service Delivery: Achieving Accuracy and Consistency in Complex Transactions", supra.
FN29 Ibid.
FN30 Ibid.
FN31 Ibid.