CTC Session Articles
Education Session Article
This article was written in support of a presentation given at CTC7 in 2001.
What do we know about Judicial Decision Support Systems? - Issues Confronting People Contemplating a JDSS
By Cyrus Tata
The paper accompanies (rather than repeats) the presentation made by the author in which he will illustrate the paper with a presentation of the implemented Sentencing Information System for the senior judiciary in Scotland, UK.
Introduction
Computers are at once both commonplace and exciting. From their inception, their awe-inspiring technical speed and power led many to suppose that computers can replace, or, in some way replicate human reasoning. This paper examines this supposition and asks what we know and what we do not know about Judicial Decision Support Systems. The paper accompanies, (rather than repeats), the presentation made by the author in which he will illustrate the paper with a presentation of the implemented Sentencing Information System for the senior judiciary in Scotland. It is this research and development and other research into judicial discretion over the last ten years, which informs the lessons outlined here in this paper.
Parameters of the Paper : IT and Judicial Discretion
The focus of this paper is not the law of judicial decision support systems, but on the nature of the relationship between IT and judicial discretion. In particular, it will tackle practical issues:
- Approaches to the construction of JDSS and understanding of judicial discretion
- The interdependence of research & development
- Implementation & Institutionalisation of JDSS
At its most basic JDSS’s are information systems, which aim to assist and support the exercise of discretionary judicial decision-making. However, while computers are now commonplace furniture of the courtroom and there has been a drive throughout the western world to equip and train judges to use computers, there are still only very few examples in the world of successful, implemented judicial decision support systems. For example, in supporting substantive sentencing discretion (as opposed to simply clarifying the law), the only successful implemented systems are those in use in New South Wales Judicial Information Research System (JIRS) and the Sentencing Information System for the senior judiciary in Scotland, UK. These are rare exceptions to the overall stark conclusion: most concrete progress in IT for judges has been limited to communications and automation tasks. For example, many judges now use word processors, e-mail, the internet, and even document assembly packages, but very little progress in the implementation of JDSS’s. Why? While word-processing; looking up the internet etc are fairly non-contentious tasks, (the only real argument might be about whether this is a good use of judges’ time), JDSS’s beg far more fundamental questions about the nature of justice and decision-making. Most fundamentally, we need to think carefully about how to approach the construction of a JDSS.
1. Construction of a JDSS
If, as defined here, a JDSS is a system to support the exercise of discretionary judicial decision-making, then the decisions are not matters of simple automation, technicality, or, logical deduction. Necessarily, a JDSS must tackle substantial decisions about which the law affords the judge substantive interpretation. In so doing, some kind of understanding about how the process of how judges make decisions is inevitable. Although there is now a substantial body of research into the exercise of legal and judicial discretion[1], it is only fairly recently that designers of JDSS have begun to take this question seriously.
To-date, the field of decision support systems in law has been largely dominated by Artificial Intelligence scholars, who suppose a "holy grail" in which machines will simulate human thinking. The AI & Law field is comprised of computer scientists, mathematicians, cognitive psychologists, and logicians, who have tended to hold (often implicitly) a view of legal and judicial decision-making as a mechanical process (albeit sophisticated one) which is basically similar to other forms of decision-making like, for example, playing chess. They have understood legal discretionary decision processes as an essentially rational, analytical and deductive individual process. In this picture the decision-maker mechanically processes given discrete "facts" and arrives at an outcome according to ‘rules’ which can be best expressed by algebra. The flaws of the AI approach to legal discretion have been explained elsewhere, and a human-social approach to JDSS[2], but below a few key lessons are set out. To the experienced practitioner, some of these lessons might resonate intuitively, but they can be all too easily be forgotten in the haste to "build" a JDSS assured by the easy seduction that the computer will do something qualitatively distinct from the judge.
i. Judgement is indeterminate.
Making substantive legal judgements is not like playing chess, where the goal of winning is clear. Judges trying to do justice inevitably have to grapple with the cold reality that you cannot "win." There are competing virtues, competing means, various and disparate audiences (e.g.: general publics, media, parties, "court workgroups") to try to satisfy. Judicial discretion cannot be adequately understood by the application of mathematics, algebra, and algorithms. Justice is not something that can be "solved," but can only be pursued in hope.
ii. Judicial Decision Support Systems cannot find objective answers.
During the 1970s and 1980s even into the 1990s there was immense interest among Legal Formalists and enthusiastic practitioners in creating "Expert Systems" for judges and lawyers which would simulate judicial processes. On the basis of complex "decision-trees" of "rules" and "facts" these systems would tell the judge/lawyer the "correct" decision.[3] (e.g. Bainbridge). While a handful of AI scholars hold fast to the panacea of Expert Systems, the realization among much of the AI and law community that "Expert Systems" were unappealing to judges and lawyers, (in all but the most trivial of discretionary decisions), led them to try to moderate the tone of their expectations.
One intriguing example of the resilience of the belief that there is an objective answer to justice which IT can find is in the ambitious personal project of a Circuit Court Judge from Oregon: Michael Marcus. Frustrated by the instrumental futility (e.g.: "revolving door") of normal sentencing practice, Marcus believes in using IT to ensure that sentencing is more "effective."[4] This is a laudable goal. Marcus is no doubt correct to observe that judicial sentencing does not serve the public as well as it might. Yet, his call for "smart sentencing" on the basis of previous recidivism rates seems to ignore the complexity of research into "what works," and in particular the intractable difficulty in agreeing "effectiveness." Sadly, scientific research data does not give the clear and simple answers which one might hope for. Secondly, the provision of recidivism data can be used to justify a variety of penal policies rather than the broad de-carceration, which Marcus seems concerned to advance. In this way, his work (perhaps unwittingly) shares a crucial defect of the Expert Systems approach in supposing there to be a single ‘correct’ answer to what judges should do in individual and uncertain cases.
iii. Judging is not simply "rules plus case facts."
Discretionary legal decision-makers do not make their decisions by adding and subtracting discrete pieces of information. Necessarily, AI approaches have to reduce the exercise of discretion to algebra. Yet, cases cannot be understood, (as the AI approach supposes), as a composite of discrete "facts." Rather, as judges are at pains to explain, they make decisions about the whole case. The whole case is far greater and more meaningful than the addition of the sum of abstract parts of the case. In practice, case "Facts" are not necessarily simple or obvious. They are contingent on the interpretation of the case as a whole. Thus rather than trying to break cases down into abstract parts, it can be more profitable to classify different types of recognized whole cases.
iv. A JDSS cannot survive in the real world without broad judicial support.
Many judicial systems have been prototyped by AI researchers using one judge as the domain expert.[5] Of course, one judge is perhaps better than no judge for prototyping purposes, but to get beyond prototyping to a credible system for real world use, a JDSS has to enjoy the support of and "ownership" by the users. A JDSS for substantial discretion cannot be imposed on judges.
v. JDSS will only survive when they provide judges with both useful and "neutral" data.
Given the indeterminacy of law, facts and the ingenuity of discretionary decision-makers, judges will not accept systems to "support" their decision-making which they feel present them with partisan data. Of course, all data is in some sense non-neutral in the sense that certain things are presented and others are not etc. However, the important point is whether judges perceive data produced by a JDSS produces as "neutral." For judges to find the data both neutral and useful/informative necessitates research into routine judicial practice and the everyday character of the use of discretion. This research has traditionally been resisted by judiciaries world-wide, but which they must permit if they are to allow themselves to gain a JDSS useful to their needs.
2. The Absolute Mutual Interdependence of Research & Development
To design a JDSS which can enjoy judicial support necessitates careful research not so much of the technology but of judicial discretion and the field of interest. If it is to survive in daily use, any JDSS must offer data which is both useful yet perceived as neutral, the design of JDSS has to be researched carefully with judges. For example in the Sentencing Information System for Scotland’s High Court and Appeal Court judges now in use, easily the most challenging task has been trying to capture how judges think in the sentencing decision process. This cannot be garnered simply from existing official sources, but necessitates academics and judges working together to articulate what is normally tacit or coded in everyday short-hand communications[6]. Development of the software and research of judicial discretion must be inextricably interdependent.
Yet it seems to be so tempting for officials and judges to ignore the lesson that a JDSS is only useful because of the research into judicial daily practice and information needs which underpins the apparent simplicity of a JDSS. For example, this author’s experience of demonstrating the successful Sentencing Information System (in use in Scotland) to judges, officials and others from various countries suggests that there is a strong temptation to see research as incidental to the main "product" (i.e. the software). Curiously, this temptation is strong precisely because the system is seen to be intuitive and successful. This is paradoxical: after all, it is the research and close tandem with development which makes the system seem so intuitively meaningful to judges; yet the fact that a system enjoys judicial support and seems neutral to judges, easily leads to the erroneous assumption that a JDSS is therefore mainly a technical matter. This is a grave mistake: no JDSS (in the sense of dealing with matters of substantive judicial discretion) will receive any widespread long-term judicial support without careful research into the everyday reality of judging. Development of perceived useful software and neutral data depends on such research.
3. Implementation and Institutionalisation
Implementation of a JDSS, in the technical sense, is one thing. Institutionalisation, (in the sense of a JDSS becoming an institutionally authoritative tool), is quite another. Even where there is a sense of ownership of a JDSS, judges in common law countries are likely to display a resolute ambivalence to this question of whether a judge-supported JDSS should be institutionally authoritative. In other words, on the one hand Appeal Courts may support the use of JDSS, but arguing that a JDSS should be authoritative begs further more fundamental questions.[7]
Conclusions
Creating a JDSS which is used on a daily basis and enjoys broad judicial support is an ambitious task. By its nature a JDSS cannot simply support automotive or technical decisions, but rather substantive judicial discretion. Far from being a simple technical task, the fundamental challenges lie in developing a system which judges recognize as both useful and producing neutral data. This cannot be achieved by resorting to attempts to replicate legal judgement through the medium of algebra and mathematics processes. Research into the everyday practice of judgement-processes is more likely to provide the indispensable basis of system, which judges feel is relatively natural to them.
[1] See for example the recent, seminal K.Hawkins (1992) The Uses of Discretion Clarendon Oxford University Press
[2] See P.Leith (1998) "The Judge and the Computer: How Best Decision Support?" Artificial Intelligence & Law International Journal 1998 pp203-230; see also C.Tata (1998) "The Application of ‘Rules’ and Judicial Intelligence to Systems Supporting Discretionary Judicial Decision-Making" Artificial Intelligence & Law International Journal 1998 pp203-230.
[3] See for example W. Bain (1989) "Judge" in C.K. Reisback and R.C. Schank (eds) (1989) Insider Case-Based Reasoning, New Jersey: Lawrence Erblaum Associates; D. Bainbridge (1991) "Case Computer Assisted Sentencing in Magistrates’ Courts" Paper presented to BILETA Conference 1991; P.Hassett (1993) "Can Expert System Technology Contribute to Bail Decisions?" International Journal of Law & Information Technology 2
[4] See Marcus website for his project: www.smartsentencing.com
[5] See for example, Computer Scientist Uri Schild’s use of Judge Talgam as Schild’s single "expert" on sentencing. International Review of Law, Computers & Technology Vol 14 (3).
[6] C.Tata "Conceptions and Representations of the Sentencing Decision Process" Journal of Law & Society Vol.27 (3) pp 395-420
[7] For explanation of this point see C.Tata (2000) "Resolute Ambivalence: Why Judiciaries Do not Intsitutionalize Tehir Decision Support Systems" International Review of Law, Computers & Technology Vol 14 (3) pp297-317
Cyrus Tata
Co-Director Centre for Sentencing Research
Associate Professor
Law School
Strathclyde University
Glasgow, Scotland, G4 ORQ
United Kingdom
Ph: + 44 +(0)141 548 3274/4863
Fax : + 44 +(0)141 553 1546
www.law.strath.ac.uk/CSR
E-mail: cyrus.tata@strath.ac.uk
Biographical Information
This biographical information may date from as far back as 2001. Please keep in mind that it may no longer be accurate.
Cyrus Tata
Co-Director Centre for Sentencing Research
Associate Professor
Law School
Strathclyde University
Glasgow, Scotland, G4 ORQ
United Kingdom