The Spirit of St. Louis – An ICAIL-2001 report

Henry Prakken

Department of Information and Computing Sciences

Utrecht University

Utrecht, The Netherlands

The Eighth International Conference on AI and Law, which took place in St. Louis, USA, May 21-25, was a success in many respects. The historic city of St. Louis turned out to be a perfect place to host the conference, with some impressive sights (especially the beautifully constructed 200 meters tall arch, symbolising the ''Gateway to the west'') and with some very pleasant neighbourhoods. Fortunately, the weather, which in May can already be very hot and humid in St. Louis, cooperated with the organisation, providing pleasant temperatures and refreshing rain showers.

The conference chair, Ron Loui, did an excellent job. The impressive Washington University campus and Law School building provided the perfect academic ambiance, and the conference hotel, the elegant Chase Park Plaza, was perfectly located in St. Louis' 'showpiece neighborhood' Central West End, so that the participants could continue their discussions well into the night in one of the many nice bars and restaurants. A highlight at the first conference day was a tour of the historic Old Courthouse, the site of the famous Dredd Scott trial, which once sparked the American civil war. The tour was even more appreciated since the transportation from the campus to the courthouse, at high speed in an old open tram which would have been banned in any other civilised country, had nearly brought a premature end to the AI & Law field. The final organisational highlight was the wonderful conference banquet at the roof of the conference hotel, ending with a spontaneous jazz improvisation by Mark Lauritsen (piano) and Tom van Engers (trumpet), which sparked unsuspected dancing talents of others.

It was good to see that some downward trends of recent ICAIL editions were stopped or even reversed. The number of submissions grew from 41 to 47, and the number of participants from 79 to 82; although some old friends had chosen to stay home, many new friends had found the way to the conference for the first time. A small disappointment was the poor attendance of the four tutorials, but the two workshops, Monday’s 'Legal Knowledge Systems in Action' and Friday's 'AI and Legal Evidence' were very successful, with around 30 participants, interesting contributions and lively discussions.

The main program consisted of 24 refereed papers, 9 refereed research abstracts, three invited talks, several system demonstrations, and a lively panel discussion on 'AI & Law meets jurisprudence', in which two AI & Law researchers, Jaap Hage and Danielle Bourcier, sided with two legal theorists, Frederic Schauer and Clark Cunningham. Frederic Schauer, of Harvard University, also gave an impressive invited talk (in the appropriate setting of the Old Courthouse), eloquently discussing the ever present tension between the aim to draft clear and predictable laws and the aim to ensure justice in individual cases. The second invited talk was given by Benjamin Grosof of MIT, who sketched very encouraging prospects of applying rule-based technology to the specification of contracts and regulations in electronic commerce. The final invited talk was given by one of the leading researchers from our own field, Kevin Ashley, who give a very insightful overview of the application of models of legal argument in intelligent tutoring systems. The Donald Berman Award for the best student paper had an impressive winner in Juliano Maranhao from Brazil, with his paper 'Refinement. A tool to deal with inconsistencies'.

The ICAIL conferences, being the world’s premier gathering of AI & Law researchers and practitioners, provide the best opportunity for scientific 'trendwatching' in our field. I will now sketch what I think were the most important developments at ICAIL-2001. The reader is warned that these are just my own subjective impressions, which may well differ from those of any other participant.

The ICAIL conferences have always been the forum for, roughly, two kinds of researchers: those interested in how information technology can aid legal practice, and those interested in understanding legal reasoning by computational means. Naturally, the first kind of researcher focuses more on practical applications while the second kind of researcher is more interested in testing theories. Therefore, it comes as no surprise that the first research strand has resulted in far more practical applications, especially for sentencing, information retrieval and public administration.

Perhaps the major practical success story of our field is the application of rule-based expert systems, especially to public administration tasks (recall e.g. Peter Johnson’s invited talk at ICAIL-99). At the present conference a new practical use of rule-based systems emerged, viz. for providing small-scale legal advice on the WEB. (e.g. Woodin and Stranieri et al. in the main conference, and Macrossan in the 'Legal Knowledge Systems in Action' workshop). Benjamin Grosof, in his invited talk, pointed at another new application area for small-scale legal rule-based systems, viz. the automation of ''law in the small'' in e-commerce applications, possibly combined with Legal XML, as also illustrated by Laurence Leff in his tutorial. Finally, an impressive report on large-scale applications for public administration was given by Tom van Engers, who leads an R & D group at the Dutch Tax Department of a scale most academic researchers can only dream of. He also showed that the large scale of his projects leads to several new research issues, such as the semi-automatic generation of knowledge bases from the legal sources with the help of natural-language parsers. In sum, it seems that legal rule-based expert systems are becoming a major practical success (as also stressed by Karl Branting in the previous issue of this newsletter).

For long it has been commonplace in AI & Law that rule-based systems, although useful for many applications, do not provide a suitable model of legal reasoning 'when the rules run out'. Many theoretical advances have been the result of this observation, and at ICAIL-2001 several important further contributions were made, for instance, on the topics of theory formation and coherence. However, the practical spin-offs of this research strand are still virtually absent. Should we be worried by this? Some will say no, arguing that understanding legal reasoning is a valid research goal in itself for our field. Others, however, fear that if a theoretical research strand leads to no practical spin-off at all, it will eventually die as a branch of AI. I share this fear, and since I am from the theoretical research strand myself, it of vital interest for me to find reasons for optimism when it comes to applications.

Fortunately, I have found such reasons at ICAIL-2001, and I heard several others saying that they were more optimistic at the end of the conference than at the beginning. My optimism, however, is not about traditional knowledge-based systems for legal argument (such as TAXMAN, HYPO, CABARET, and GREBE). Although these systems brought important theoretical advances, they require a large amount of formalised knowledge in order to scale up to practical applications; and the knowledge acquisition bottleneck seems just too hard for such systems, since the knowledge used in arguing when the rules run out is simply too diverse, imprecise and unreliable.

AI & Law researchers have investigated two ways to avoid this bottleneck. The first is to let the system automatically learn its knowledge from the natural-language sources (see e.g. Daniels & Rissland and Brueninghaus & Ashley at ICAIL-97). This approach was continued at ICAIL-2001 in several papers. Given the recent AI successes in natural-language understanding, I expect that this research strand will eventually lead to useful applications, especially when combined with information retrieval (in this respect it was encouraging that Westlaw's AI research group, led by Peter Jackson, sponsored the conference and contributed with several papers).

The second way to escape the knowledge-acquisition bottleneck is to let the user provide the knowledge. This approach (pioneered by Tom Gordon at ICAIL-93) leads to procedural-support systems, viz. systems that lack domain knowledge and thus cannot solve problems, but that instead ensure orderly and effective disputes and help the users to structure their arguments. Two recent developments make me optimistic that this research strand can be practically successful: the rise of online dispute resolution on the WEB (described at ICAIL-2001 by Arno Lodder) and recent commercial systems for knowledge management in litigation, such as CaseMap (www.casesoft.com) and Masterlist (presented at the 'Legal Knowledge Systems in Action' workshop). I think that our field has a lot to offer to these developments. However, I also think that to obtain practical success, more empirical research is needed into the ways lawyers structure their reasoning and knowledge, and into how arguments and disputes can best be visualised. At ICAIL-2001, Conrad & Dabney’s investigation of the cognitive structure of judicial opinions was a very interesting example of such empirical research.

Having expressed my optimism about practical spin-offs of the theoretical research strand, I want to conclude with some thoughts on the future place of the practical research strand at our conferences. There was a time when researchers interested in improving legal practice with IT almost automatically ended up in our field. However, times have changed, and these researchers now have many different topics to choose from, not all of which are counted as AI, such as knowledge management, electronic institutions, legal XML and online dispute resolution. In my opinion, the ICAIL conferences should not lose contact with these fields, but aim to be the premier forum not only for ‘traditional’ AI research, but also for new research on advanced IT and law. Fortunately, as indicated above, some such research was already reported at ICAIL-2001, and I hope that these trends will continue at ICAIL-2003 in the wonderful city of Edinburgh, where with no doubt John Zeleznikow will organise a great conference and Giovanni Sartor will take care of a great program.