W(h)ither AI and Law?

June 30, 2004

W(h)ither AI and Law? 
Back in the heady days of AI and law (late 1980s/early 1990s), was one of those students who enthusiastically embarked upon a thesis in this area but, on its successful completion, I was a little less sanguine as regards the prospects for the practical employment of legal AI.[i] Consequently, my involvement in this field lapsed. However, research, over the past few years into the impact of general IT on legal practice has rekindled my interest in AI and law.

What follows is a personal account of the state-of-play in this sphere; starting with a look at current research, as reflected in ICAIL-03, moving onto AI and law in practice, then advocating AI’s application to address a specific socio-legal need, and ending with a view of how XML, a technology which is rapidly and pervasively being adopted, might affect AI and law in the near future.

Research – ICAIL-03
In June 2003 I attended the biennial (9th) International Conference on Artificial Intelligence and Law, ICAIL-03.[ii]

The ICAIL is recognised as the leading international gathering for researchers in AI and law, and thus can be regarded as indicative of current focuses; academic and commercial.

As such ICAIL-03 illustrated two trends. First, it demonstrated the shift of emphasis which marked the last decade, away from work which has the development of a system of practical use to lawyers as a goal, even a remote goal. Secondly, it reflected the current preponderance of contributors to the field of AI and law whose academic origin is other than law (eg logic, mathematics, philosophy, computer science).

These two factors are, of course, not unrelated. They reflect, on the one hand, the disillusionment with theory of many individuals who are interested in practical systems. Such individuals recognise the limits of what is achievable given current and foreseeable AI techniques and the fundamental characteristics of reasoning in most legal domains (eg open texture, indeterminacy, nonmonotonicity, rule-conflicts). Many of these workers have turned their attention to what can be achieved with existing, proven tools, rather than trying to devise new (better) computational models.

On the other hand the above factors demonstrate that for those who now dominate the field of AI and law, law itself is merely an interesting domain in which to apply their disciplines and test theories. Their primary goal is not that of improving the efficiency of legal practitioners, nor, to quote the words of an early researcher in the field,[iii] that of making scarce knowledge and expertise more widely available and easily accessible.

The conference was dominated by papers on the following themes:

Argumentation and Logical Models, through which logicians and computer scientists reinforced what has probably been the dominant theme since the early 1990s, namely, the never-ending attempts to produce formalisms (logical or mathematical models) of the law capable of accommodating its essential nature, which classical logic cannot – law’s open texture, defeasibility, etc, in other words the human, rhetorical, ‘real world’ character of law and legal argument as it is practised.

Ontologies ie generalised conceptual models of specific legal domains. The futility of attempting to build such models would seem apparent to most lawyers when considering how (other than for closed, precisely defined areas of law, of which there are few) such a model could be expected to encapsulate the inherent (and often intentional) openness and flexibility of legal principles and rules, and their mutation through interpretation and interaction with equally fluid economic, social, political and cultural factors.

Deontic Logics – these are logical models attempting to formally express the legal modalities such as permission, obligation, prohibition, and their interaction. The impression left by papers presented on this topic was that the more work that goes into trying to express legal modalities, the more apparent becomes the complexity of the issue, and the greater the need to accommodate exceptions and do even more fundamental research.

Two other themes (confined to part of the final day of ICAIL-03) did, however, offer speakers whose work is such that even a humble lawyer might see relevance in it.

Case Based Reasoning (CBR) Brüninghaus and Ashley’s paper on CBR systems shunned mathematical notation and logic symbols in favour of prose expressing arguments and counter arguments which we all might recognise as being employed by protagonists as a case develops. However, the aspect of this presentation of greatest interest was not the description of the researchers’ current project itself but the reiteration of the obstacles to the development of CBR systems, which are basically unchanged since first identified during the initial excitement surrounding CBR over a decade ago. Indeed Ashley, a pioneer in and still at the forefront of CBR, has previously acknowledged that the practical difficulties of building CBR systems mean that such systems could remain restricted to the research world.[iv] However, as pedagogical tools, which are now the focus of Ashley and others, CBR systems have had success.

Application Models While the aim of applying legal knowledge to a problem, or to achieve a particular outcome, tends to be the (more elusive) focus of most academic AI and law researchers, the large legal publishers seem to focus their research efforts on improving the more prosaic preceding task of accessing the information needed (or at least that which the lawyer believes is relevant).

These publishers already have widely used, practical IR (information retrieval) systems in place, and they hold vast resources of the raw materials of legal IR, ie the legal texts themselves. They are also better able, both financially and staff-wise, to support research than most academic institutions. Consequently they are ideally placed to pioneer improvements in the recall and precision performance of legal IR systems.

A paper from Thomson-West indicated that one of their continuing initiatives in this respect (ie improving the performance of legal IR) is to improve retrieval precision by allowing legal researchers to categorise their ‘information need’ into one of a number of predefined high-level classes of queries. However, although the presenters claim success in their experiments for this approach, they admit to huge obstacles to developing a production system. (So don’t expect any great improvements in commercial legal IR systems just yet!)

The session on application models also held out some hope, through a paper by Hall, Hall & Zeleznikow, that one of the fundamental impediments to the development and adoption by firms of practical expert systems is at last being tackled, or at least (tentatively) explored. This is the lack of a method to evaluate such systems, which the researchers believe must exist if expert systems are to move out of the research laboratory and into the marketplace. They also acknowledge that this need was in fact first suggested by Susskind at the first ICAIL conference back in 1987.[v] One might conjecture that it is perhaps indicative of the AI and law community’s lack of interest in seeing useful systems in everyday use in the workplace that it has taken so long for serious research on this matter to commence.

While the aspiration of these researchers is undoubtedly commendable, one should not expect this work quickly to produce an actual method or system for evaluating any (proposed) expert system; the work described is very much at an early stage. However it is encouraging (building on the standard international methodology for software evaluation, ISO/IEC 14598), ongoing, and it was interesting to hear the presenter comment on the interest in the research being taken by the Victoria Bar Association and City of London firms.

Rule-based systems – success?
The foregoing commentary on artificial intelligence and law in the context of ICAIL-03 paints a disappointing picture for those of us interested in practical advances. However, there is an AI tool which has already proven itself in the field of law, being employed in practical systems by government bodies, commercial concerns, and academic researchers. This is the simple production rule,[vi] employed to make rule-based ‘expert’ or ‘decision support’ systems.

Rule-based systems are unsophisticated programs, albeit potentially very intricate ones, and their potential domains of application are severely limited. They cannot deal well with situations where the law is uncertain, nor areas in which the facts are debatable. However, in areas such as procedural and administrator law, where the expertise lies in negotiating the complexity of known laws, for which legal opinions on interpretation can, if necessary, be encapsulated (because not infinitely variable) then such systems can provide useful advice.

Proof of the success of expert systems in such domains, where the law is complex but not discretionary, is visible in the systems of tens of thousands of rules developed by SoftLaw for the Australian government. Equally, rule-based engines encapsulate the expertise employed in the Web-based advisory systems developed by City firms for in-house counsel; for example, Linklaters’ Blue Flag and Clifford Chance’s NextLaw services.[viii]

In fact, it would seem that the development of the Web has provided a fillip to the development of practical rule-based expert, or advisory, systems.

As the UK legal market has started to appreciate the potential for both generating additional revenue and attracting new business that the Web offers, roughly 200 sites offering on-line legal services have arisen.[ix] These services range across the legal market from consumer and small business targeted document assembly systems, commercially epitomised by the RapiDocs product, through those products such as Hotdocs used by larger firms such as Lovells and Wragges & Co,[xi] to the expertise behind international financial services compliance advice, targeted at in-house lawyers. For a web site to dispense online advice, or assist with a task (rather than just provide downloadable static forms) a rule-based system of some degree of sophistication must lie behind the service.

It might not always be possible to reference actual documents describing specific AI related tools employed in law firms, accountancy practices and the like for, as Oskamp and Lauritsen observe, they will be kept out of the public eye for reasons of competition.[xii] However, it is reasonable to infer that the AI employed will be at least as sophisticated as the rule-based expert system described over a decade ago by Ernst & Young at ICAIL-91; a time when firms were less circumspect about parading their achievements.[xiii]

At a more general level one might speculate that, in addition to their deployment in client-facing advice systems, the scope for expert systems of practical use within the law office is also increasing. Susskind and others observe that rule-based expert systems are suited to administrative and procedural areas of law.[xiv] Wall comments on the increasingly administrative and procedural nature of the modern solicitor’s job.[xv] (The view that it is procedures, rather than substantive law, which occupies the majority of the time of practising lawyers is stated also by Leith and Hoey.)

Combining the ‘what are expert systems good for’ aspect with the ‘what is it that most lawyers actually do’ element would suggest rule-based systems being increasingly used internally by and for lawyers.[xvii] However, in this context perhaps we would more easily recognise them if discussing them in conjunction with more currently acceptable jargon such as ‘commoditised work product’ and ‘workflows’.

Social need and automation
While in the UK high profile, big budget projects attract most attention in the legal press, in the US and Australia rule-based systems have seen success at the other end of the legal market.

Within the US, the web-based Protection Order Advisory system described by Branting operates throughout Idaho and helps litigants in person both to understand the legal remedies available to them and to satisfy the substantive and procedural requirements for those remedies. In Victoria, Australia, the GetAid system has been developed to free lawyers from the extremely time-consuming administrative task of assessing an applicant’s eligibility for legal aid, and allow them to concentrate on the substantive task of dispensing advice.[xix]

Both Zeleznikow’s and Branting’s work has an inspirational social dimension to it, ie both systems enable more people to benefit from legal expertise than would otherwise be the case. Thus they seem to have made concrete steps towards the goal articulated by Susskind (and mentioned earlier) of making scant legal expertise and knowledge more accessible and available.

AI, Law and Utilitarianism in the UK
Although Zeleznikow’s and Branting’s contexts are Australia and the US, their underlying concern, that many people who might wish to benefit from professional legal advice do not obtain it, would appear to be increasingly manifesting itself in the UK.

For example, on the one hand a substantial number of private individuals have significant legal problems, but do not qualify for legal aid and consequently either represent themselves or forgo acting because they fear the prospect of high legal fees.[xx] (It should be noted that this substantial number could become much greater.) For those who do qualify for legal aid, its availability is becoming increasingly perilous because of the shortage of legal aid lawyers.The injustice of this situation is obvious.[xxiii] Additionally, the strain on the legal system itself which litigants in person impose is observed by both the above researchers as well as, specifically in the UK context, the likes of the Master of the Rolls Lord Phillips and Lord Justice Brooke.[xxiv]

Given all these issues, surely the UK government should be exploring the work of Branting and Zeleznikow, and other similar research, with the aim of providing systems which, while fulfilling a primary function of assisting the individual, could (incidentally) help protect the legal system.

This government has, after all, in recent years, expressed in consultation papers such as civil.justice and civil.justice.2000,[xxv] the desire to employ IT radically to improve the delivery of legal service for the majority of the population. However there is, to date, little evidence of this happening and the issue of exclusion from legal assistance is often being highlighted; for an example, see the recent survey of the Legal Services Commission.[xxvi]

On the horizon? Legal XML and rule-based (expert) systems

Analysts such as Gartner Group seem certain of the ascendancy of XML and its support across the software industry would seem to confirm the accuracy of that belief. More particularly, ‘legal XML’ initiatives developing across the world would suggest that XML will have a major impact in the field of law. This view is certainly evident in the bold statements of some commentators; for example, “Legal XML is a new family of standards under development that eventually will dominate the electronic exchange of legal data”.[xxx]

As regards the relevance of XML to AI and law, the prevalence of XML could greatly enhance the environment for, and likelihood of, development of rule-based systems because XML-formatted data can facilitate the operation of such systems.

[xxxi] one might reason that, if the documents containing the required input values were marked up in XML, then the efforts required to build and develop an (XML aware) expert system might be considered worthwhile, as the inputs could be automatically extracted from those documents.[xxxii] Avoiding manual data entry saves time, staff costs and avoids transcription errors. These would be benefits additional to those of the expert system per se to be considered when debating whether to build one.

Of course the underlying assumption here is that the documents in question have been marked up in XML. The above noted incentives for system development would not exist if it was thought inputs would still have to be manually transcribed. However, it is not unreasonable to speculate that government bodies or corporate clients might soon demand XML-formatted documents for certain matters (in order to reap full benefit from in-house systems they had developed), just as Word or PDF documents might be required today. Given Microsoft’s support for XML and the fact that Word 2003 is an XML editor, such speculation should not be regarded as fanciful. (Some IT commentators even suggest Microsoft sees XML as eventually supplanting its own proprietary Word format but this probably is fanciful.)

However, perhaps the most significant recent development in the context of ‘what XML can do for expert systems’ is that concerning the statute law database. As was recently announced on the SCL Web site, there now seems to be greater likelihood of free, public access to the database, over the Internet (possibly by the end of 2004), being achieved than there has been in recent years.

[xxxiv] plus the general acceptance that law which is primarily statute based is most suited to rule-based expert systems, together with the fact that XML facilitates the automatic processing of information in documents, one must conclude that rule-based systems using data from, and/or guiding one through, the legislation within the statute law database would seem an obvious area of development for government, law firms, software houses and others.

Where are we … practically?
The AI technology of practical use, the rule-based system, has changed little over the last 15 years.[xxxv] Using it, useful systems have been, and will continue to be, built.[xxxvi] However more sophisticated AI techniques and fundamental AI research are unlikely to yield practical results in the foreseeable future, if ever.

Tools to build decision support, or expert, systems are continually being developed and refined (for example, the open source inference engine JustReason[xxxvii]) and XML appears to hold out the prospect of more easily applying rule-based techniques to legal tasks. Each of these factors might provide an additional incentive to bodies which see possible benefit in building a system in a particular noncontentious and procedural or administrative area of law to go ahead and do so.

While the production rule has so far been the principal success as regards practical AI in the field of law, research, both from within the AI and law school (such as that discussed earlier) and out with it, should not be ignored. For example, from the latter it is reported that Glasgow University’s computing science department has developed software that tracks the development of arguments within documents and looks for material connected to the conclusions.[xxxviii] No doubt many readers could think of potential uses for such software in a legal context.

So, as ever, those who look forward to AI having a major impact on the law and transforming the work of the legal practitioner beyond recognition, or radically improving the service provided to the public, should not lose heart. The big technical breakthrough could be just around the corner … maybe.

We have the technology …
But history shows us that it is often socio-economic factors that prompt a radical change in the way things are done in society (rather than intellectual endeavour or technological evangelism).

The economic aspects of the legal aid crisis are that solicitors are leaving this type of work because they perceive it to be poorly paid, and that increases in pay rates have been repeatedly ruled out by the Department for Constitutional Affairs. The social dimension is the obligation of government to ensure proper legal service for all members of society.

If any body is positioned to bring about fundamental change, it is of course government, and this one has made much of its desire to improve access to justice, legal services, etc. through IT. Unfortunately the UK government has an inglorious record as regards IT projects, but the problem of ‘legal aid advice deserts’ will not go away, and rule-based systems have been successfully applied by governments elsewhere. Moreover, the documentation of fruitful projects in other countries to make scarce legal advice more widely available, specifically to the socially disadvantaged, is available for all to see.

Thus we already have the technology, and we certainly have the need. Surely the time is right for a greater deployment of AI in the law in this country.

Alastair Morrison works in IT at Strathclyde University where his main current focus is an institution-wide deployment of Exchange 2003. If you would like to discuss any aspect of this latest version of Microsoft’s leading e-mail, calendering and collaborative working application, or the University’s particular implementation of it, e-mail him at alastair.morrison@strath.ac.uk.

[i] For a definition (and brief history) of, and appraisal of the practical issues concerning, AI and law see, for example, Susskind, R. Transforming the Law Oxford University Press 2000 pp162-176.

[ii] ICAIL-2003 Proceedings of the 9th International Conference on Artificial Intelligence and Law June 24-28 2003 ACM Press.

[iii] Susskind, R. Transforming the Law Oxford University Press 2000 at p163.

[iv] Brüninghaus, S. and Ashley, K. D. Improving the Representation of Legal Case Texts with Information Extraction Methods ICAIL-2001 ACM Press at p 42.

[v] Susskind, R. E. Expert Systems in Law: Out of the Research Laboratory and into the Marketplace ICAIL-1987 ACM Press pp1-8.

[vi] Which in computer science equates to one of the basic programming structures, ie the IF … THEN … ELSE statement. For an explanation of ‘production rule’ and other ‘expert systems’ terminology see, for example www.expertise2go.com/webesie/tutorials/ESGloss.htm

[vii] www.softlaw.com.au

Zeleznikow, J. Using Web based legal decision support systems to improve access to justice Information & Communications Technology Law Vol 11, No 1, 2002, pp15-33, at p18.

[viii] www.blueflag.com and www.nextlaw.com

[ix] Christian, C. Online Strategies for Law Firms 2004 www.legaltechnology.com/casestudies/OnLineStrategies2004.doc.

[x] www.epoq.co.uk/ep/

[xi] www.hotdocs.co.uk

[xii] Oskamp, A. and Lauritsen, M. AI in Law Practice? So Far, Not Much AI & Law Vol 10, No 4, 2002 pp227-236 at p 227.

[xiii] Thumper – An Expert System for Stamp Duty ICAIL-1991 ACM Press pp 266-271.

[xiv] Susskind, R. Transforming the Law Oxford University Press 2000 p 164.

[xv] “it was realised how little of the modern solicitor’s job involves the substance of law and how much it now involves administrative and legal procedures: thus indicating the degree to which lawyering has become rationalised and has moved away from the traditional conceptualisation of legal professionalism, towards new types and levels of specialism.” Wall, D. S. Law has a Future but do lawyers? Legal Executive Journal, 1 February 2002 www.ilexjournal.com/special_features/article.asp?theid=407&themode=2

[xvi]Hoey, A. The Computerised Lawyer Springer-Verlag, 1998 (2nd revised edition), p 111.

[xvii] Corroboration of this thinking may be found in one of Susskind’s recent pronouncements – “Lawyers are likely to devolve huge swathes of routine work to computers in 2004” Hold on tight for a surge of new technology The Times, ‘Law’ supplement, Tuesday 13 January 2004 p 10.

Branting, L. K. Advisory Systems for Pro Se Litigants ICAIL-2001 ACM Press pp 139-146.

[xix] Zeleznikow, J. Using Web based legal decision support systems to improve access to justice Information & Communications Technology Law Vol 11, No 1, 2002, pp 15-33.

[xx] Widdison, R. Electronic Paths to Justice Journal of Information, Law and Technology December 2003 para. ‘1 Introduction’ http://elj.warwick.ac.uk/jilt/03-2/widdison.htm

[xxi] Apparently consideration is being given, in the draft criminal defence bill, to setting the threshold for legal aid at an income level as low as £27,500. – ‘Means test’ in criminal legal aid makes a return … but will it make a difference? LawZone Newswire, Issue No. 245 – 27 January 2004 http://www.thelawyer.com/cgi-bin/item.cgi?id=108252.

[xxii] The looming crisis in legal aid provision has been consistently observed over the past 2 years, by such diverse bodies as Citizens Advice, AdviseUK, the Law Society, the Public Accounts Committee. For recent examples see Legal Aid advice deserts – one-in-four CABx report legal aid ‘no go’ areas LawZone Newswire, Issue No. 247 – 10 February 2004 http://www.thelawyer.com/cgi-bin/item.cgi?id=108407; Law society warns over legal aid Financial Times, Monday October 13 2003 p 4; Community legal advice is in danger of drying up The Times, ‘Law’ supplement, 30 September 2003 pp 6-7.

For one example of the human consequences, see Legal Aid advice deserts – one-in-four CABx report legal aid ‘no go’ areas LawZone Newswire, Issue No. 247 – 10 February 2004 www.thelawyer.com/cgi-bin/item.cgi?id=108407.

[xxiv] Civil justice system overrun with ‘hopeless cases’ 20 January 2004 www.thelawyer.com/cgi-bin/item.cgi?id=108168&d=11&h=24&f=23.

[xxv] civil.justice – Resolving and Avoiding Disputes in the Information Age September 1998


civil.justice.2000 – A vision of the Civil Justice System in the Information Age June 2000


[xxvi] “. one in five people . ‘don’t understand their basic rights or know how to seek help’ ” Legal aid and social exclusion LawZone Newswire, Issue No. 248 – 17 February 2004 http://www.thelawyer.com/cgi-bin/item.cgi?id=108492

[xxvii] Extensible Markup Language (XML) is a metalanguage, a language for describing other languages. With XML one (or a profession) can develop one’s own markup language and use it to describe the parts of a document. XML allows a program to understand the meaning of pieces of text that would otherwise be meaningless strings of characters. For more information, in Google type define:XML

Eguchi, G. and Leff, L. L. Rule-based XML AI & Law Vol 10, No 4, 2002 pp 283-294 at p 284.

e.g. An XML Schema for the Legal Profession version 2.1 www.legaltechnology.com/casestudies/LSSA_XML_Schema_White_Paper.doc

[xxx] Mountain, D. XML E-Contracts: Documents that Describe Themselves International Journal of Law and Information Technology 11(3) Autumn 2003 pp 274-285 at p 275.

[xxxi] Eguchi, G. and Leff, L. L. Rule-based XML AI & Law Vol 10, No 4, 2002 pp 283-294 at p 286.

[xxxii] In fact, automatic extraction of data from XML encoded documents is already being implemented in real (not research), marketed systems. The software company Real Decisions’ new smart-DOCS product allows data from PISCES (the XML standard for real estate transactions) files to be imported directly into pre-formatted templates such as new leases, valuation reports or investment particulars.

See Document Assembly – Real Decisions To Roll Out Smart-Docs Legal Technology Insider Newswire – Issue.138 – 05.02.2004

The government is certainly already promoting XML in the legal sphere. For example in Developing Common Standards – Metadata Scheme for Websites in the Legal and Advice Sectors it is stated “There should be a metadata scheme covering the legal and advice sectors .” and that “. website owners should be responsible for the creation of metadata using XML as the standard for metadata definition.” www.dca.gov.uk/consult/meta/metaresp.htm

Statute Law Database Latest 6 January 2004 http://www.scl.org/services/default.asp?p=135&c=-999&ctID=10&cID=1200000295.

[xxxv] A useful review of practical AI and law success of that period can be found in Oskamp, A. and Lauritsen, M. AI in Law Practice? So Far, Not Much AI & Law Vol 10, No 4, 2002 pp 227-236.

For producing working rule-based systems, the considerations and guidelines set out by Susskind, as early as 1990, for the successful development and operation of expert systems, are as relevant as ever.

Susskind, R.E. Artificial Intelligence, Expert Systems and the Law: A European Appraisal in Kaspersen, H.W.K. and Oskamp, A. (Eds.) Amongst Friends in Computers and Law Kluwer, Deventer 1990 pp248-253 reiterated in Transforming the Law Oxford University Press 2000 pp 175-6.

[xxxvii] www.justsys.com.au/

[xxxviii] In conclusion Guardian Unlimited Online Thursday August 21, 2003 www.guardian.co.uk/online/story/0,3605,1022668,00.html

See, for example, the recent interview with David Lammy (of the DCA) Legal aid: keeping the faith, updating the service The Times, ‘Law’ supplement, 27 January 2004, p 3.