Representations of Knowledge and Discretionary Decision-Making by Decision-Support Systems : the Case of Judicial Sentencing

Cyrus Tata, John N. Wilson and Neil Hutton.

University of Strathclyde

[email protected]
[email protected]
[email protected]

Acknowledgments

We are particularly grateful to Professor Alan Paterson for his sustained and dedicated commitment to the project which this article describes. We are also indebted to the contributions of all High Court Judges, but most especially to The Lord Justice-Clerk, The Rt. Hon. Lord Ross, The Hon. Lord Sutherland, The Hon. Lord Penrose and The Rt. Hon. Lord Murray, The Rt. Hon. The Lord McCluskey, as well as Mr. Eric Cumming and to the Scottish Office Home and Health Department for its funding of a study, part of which is discussed here. We are also grateful to Simon Halliday for his comments on an early draft of this article.

Contents

1. Introduction

2. International Reforms in Sentencing

3. Computer support to aid sentencing decision making.

4. How can the representation of knowledge enhance the utility of decision-support systems for sentencers?

4.5 Is a schematic-holistic approach open to modelling by a computer? (i) | (ii) | (iii)

References

Abstract

This article critically examines approaches to the production of systems of support for discretionary legal decision-making. It discusses a project to research and develop a Sentencing Information System for the High Court in Scotland and examines the wider theoretical implications of work to produce a system to support discretionary decision-making.

Briefly placing the Scottish development in the context of world-wide themes in sentencing reform, the article then focuses on attempts to produce systems of computer support for sentencing: both knowledge-based approaches and also database technology. It then briefly describes the background of the Scottish system and speculates on the present and future positions of the project.

Perhaps the most important question concerning systems of support for discretionary decision-making is their ability to impact on decision behaviour. We argue that although every case is unique in some sense, it is necessarily possible to compare cases and therefore to represent them as 'similar'.

How, then, should this 'similarity' be represented? Traditionally, representations of similarity have tended to be informed by 'the legal-analytical' paradigm. This privileges official criminal law offence categories as the starting point for representation and then 'adds in' further information to describe the case analytically. We argue that systems based on this paradigm may be limited in their representation of the decision process. We suggest that these limitations may be overcome by adopting an approach which tries to represent the informal schema of understanding which decision-makers employ and the holistic way in which they think about a case.

The Scottish project has possibly provided a glimpse of a more holistic and schematic approach to representing 'similarity'. However, further study may help to provide a more complete representation of the informal behavioural rules which govern discretionary decision-making.

Key words : discretionary legal decision-making, decision-support systems, judicial sentencing

---

This is a refereed article.

Date of publication: 31 March 1996

Citation: Tata, C , Wilson, J.N.& Hutton, N (1996) 'Representations of Knowledge and Discretionary Decision-Making by Decision-Support Systems: the Case of Judicial Sentencing'. 1996 (2) The Journal of Information, Law and Technology (JILT). <http://elj.warwick.ac.uk/elj/jilt/artifint/2tata/>. New citation as at 1/1/04: <http://www2.warwick.ac.uk/fac/soc/law/elj/jilt/1996_2/tata/>

---

Introduction

Is it possible to build a system of computer-support to aid the discretionary decision-making process? If so, how can that process be understood and that understanding be most appropriately represented? This article discusses the application of different conceptual approaches to information technology to the judicial sentencing process. It discusses a recent project to produce a prototype Sentencing Information System (SIS) for the High Court of Justiciary in Scotland and to study its feasibility. The article proceeds to discuss the results of the study and prospects for the future. Its prospects will depend crucially, among other things, on its reception by judges themselves. The article therefore examines how the conceptual content of computer programmes intended to support legal decision-making can increase the likelihood of their acceptance and usefulness to their Users.

Over the last twenty years, many Western jurisdictions have taken various steps to reform the sentencing process (Ashworth, 1992b, Tata, Hutton, Wilson and Paterson, 1995, Hutton and Tata 1995). The following section sets the Scottish High Court prototype in the context of international developments.

2 International Reforms in Sentencing

The primary aim of this wave of international reforms has been to reduce disparity and promote consistency in sentencing (Ashworth, 1992a). Consistency in sentencing requires that like cases are treated in a similar way and conversely that dissimilar cases receive different sentences (Hutton, Paterson, Tata, and Wilson, 1995a).

It has been the 'just deserts' approach to sentencing which has, in part, influenced most of the approaches to sentencing reform (Ashworth 1992a). This approach argues that sentence ought to be proportionate to the seriousness of the offence and not based on the character or past conduct of the offender (von Hirsch 1976, 1993). Reforms in Scandinavia have used a narrative form of guidelines to guide sentencers as to how the principle of desert should be properly applied (von Hirsch and Jareborg 1989, 1994). Arguably, the Criminal Justice Act (1991) adopted a similar approach for England and Wales. The US approach to sentencing reform was to construct numerical guidelines which specified a limited range of penalty for particular offence categories (Wilkins et al 1978). These were introduced into a number of state jurisdictions, the most well known being the Minnesota Guidelines (Tonry 1987). U.S.-wide Federal Guidelines were introduced in 1987. The relative inflexibility of numerical guidelines considerably reduces the element of judicial discretion in sentencing and they have thus not been popular with judges (see Tonry, 1987, 1992 and 1993; Freed 1992; De Benedictis 1993).

These U.S. guidelines were all formulated by a sentencing commission or committee appointed by the government for the purpose. Sentencing commissions have also been formed to provide advice for governments, e.g. in Canada, the state of Victoria in Australia and a number of US states (Ashworth 1992b). These have had a very mixed reception from governments. The Canadian guidelines have not been implemented while the Victorian Sentencing Act was passed in 1991.

3 Computer support to aid sentencing decision making.

Computer systems which have been used to support sentencing reforms include knowledge-based systems, both case-based and rule-based approaches, and simpler database retrieval systems.

3.1 Knowledge-based approaches

Bench-Capon (1994) and Zelznikow and Hunter (1994) have argued in favour of the development of computerised representation of legal rules to assist decision-making. Zelznikow and Hunter (1994) take the reader through a brief tour of legal theory in relation to building intelligent information systems. They present a debate between 'Legal Positivists' who argue that law can be represented by "...a settled body of rules..." (p.63) and 'Legal Realists' who take "...a more extreme approach and reject the categorisation of law as fundamentally about rules" (p.53). Having acknowledged the centrality of the debate they introduce, the authors conclude that "...the majority of cases should be decided on something approaching a positivist approach." However, this conclusion seems to be more as a result of a pragmatic preference than of a theoretical argument, (other than to say that strong legal realism is "...an unlikely conclusion and one which many would not accept").

We would suggest that this positivistic rule-based approach to decision-support systems is founded on a restrictive view of the sentencing process as one which is fundamentally prescribed by formal rules.(1) There appears to be a presumption in the rule-based approach that 'rules' are formal legal rules, rather than behavioural rules (Hawkins 1992). This presumption ignores the inter-connection between social and legal processes during sentencing process. Reliance on such an approach may help to explain the difficulties which confront the design and application of current rule-based systems. (See for example, difficulties reported by Hassett (1993) in producing a 'Bail Advisor').

A hybrid approach involving both rule based and case based systems has been developed by Bain (1989). The programme begins with an empty case-library and a handful of heuristics for deciding sentences when no cases can be applied to a new situation. After only a few cases, however, it begins to retrieve 'remindings' of its own cases from memory and to modify the strategies associated with those cases to form new sentences.

Knowledge-based approaches (ii)

The idea of a case based reasoning system unsupported by heuristics has also been used as a basis for modelling the sentencing process. Murbach and Nonn (1991) report progress on a project to develop a sentencing support system for fraud cases in Canada. Their system provides information about penalties but also includes information on case factors not included in the categories of offence used in the penal code but agreed by judges to be relevant to sentencing. There is thus an attempt to include information which reflects judicial perceptions of seriousness in order to make the system more sensitive and thus more useful to sentencers. Computer technology has been used to assist these reforms and to encourage greater consistency in sentencing. ASSYST (Applied Sentencing systems) has been developed by the US Federal Court system so that criminal justice personnel could easily compute, record, archive and examine the implications of the US Sentencing Commission Guidelines (Simon and Gaes 1989, Simon, Gaes and Rhodes 1991).

Schild (1995) reports work in progress to develop a case-based advisory system for sentencing. The domain knowledge was elicited from a senior judge, and the system uses "hierarchical discrimination trees" in order to retrieve relevant information. Schild notes that '[i]t is obvious that the area of sentencing is associated with an enormous amount of both common-sense and domain knowledge.' However, a model which would include this knowledge was considered impracticable and so it was therefore decided to use only the domain knowledge without any additional 'common-sense knowledge'.(p232)

Knowledge-based approaches (iii)

A part-simple retrieval system, part-expert system approach to sentencing support has been reported by Bainbridge (1991). The system focuses on sentencing practice in magistrate courts in England and Wales and contains components covering sentencing law and penalty statistics. The sentencing law component is intended to assist the magistrate by checking that the chosen sentence complies with relevant sentencing law. This part of the system is arguably more like an expert-system than a simple retrieval system, although it only answers the question, 'Is this sentence legally competent?' rather than, 'What is the appropriate sentence for this case?'. When a judge has selected a legally competent sentence, it is then possible to consult the penalty information section of the system. This shows the distribution of penalties for the offence in the form of probability calculations. Information is only available for two statutory offences of theft and burglary and for only 600 cases from four magistrate courts.

Database technology has been used in a number of large scale information systems. Such systems, generally referred to as 'Sentencing Information System', have been implemented in a variety of jurisdictions and used in practical, day-to-day basis to support the sentencing process.

3.2 Sentencing Information Systems

A Sentencing Information System provides users with information about the range of penalties which have been passed by the court for similar cases in the past. The system allows the judge to enter certain information into the computer about the case which he is considering and the range and quantum of penalties passed by the court for similar cases is displayed.

Formally, a Sentencing Information System is descriptive rather than prescriptive. That is, it contains no guidance as to how a sentencer might use this information to help in making the sentencing decision in a particular case. A Sentencing Information System (SIS) can display the range of sentences for the particular combination of offence and offender characteristics selected. The sentencer will have no guidance as to what extent and in what direction the appropriate sentence for the case at hand should vary from the average. This decision is a matter for the discretionary judgement of the sentencer. However, the frequency distribution indicates the highest and lowest sentences previously passed for the type of case at hand. In a well trodden area it might be assumed that a sentencer would have to have good reasons for straying outside the upper and lower limits, although there are no formal reasons why a sentencer should not choose to do so nor does the SIS restrict the sentencer's choice in any way.

Jurisdictions in Canada and Australia have experimented with Sentencing Information Systems. These are briefly described below.

3.2.1 Canada - Doob and Park System

One of the earliest systems was designed by Doob and Park in Canada and it operated for six years in four provinces (British Columbia, Manitoba, Saskatchewan, Newfoundland) (Doob and Park, 1987). By 1990 only the Saskatchewan system was still in operation, in the latter stages, using only Court of Appeal information. Anthony Doob has explained why he thinks judges did not make sufficient use of the system. There are two main reasons. First, judges in Canada had little interest in information about current court practice. They are not accustomed to using information in this numerical form nor does their legal tradition give any weight to current sentencing practice. Second, such authority as exists in sentencing comes from the Court of Appeal. The Sentencing Information System carried no institutional authority (Doob 1990).

3.2.2 The British Columbia System

This system operated in British Columbia from 1987 to 1992 (Hogarth, 1988) but is no longer operating. A private communication with the IT director suggests there are two main reasons. First, there was insufficient judicial consultation and involvement, particularly in the early stages of the project. Judges therefore felt that the information provided by the system was not helpful to them. Second, the costs of the system, although not revealed in detail, were very high. it appears that the cost of data collection and of a very powerful main frame computer were particularly high. Schild (1995) has suggested that Hogarth's system may be subject to criticism because "...the statistical knowledge embodied...is based on a very small number of characteristics. This does not suffice to express the actual complexity of the sentencing process." He also criticises the structure of system : it quickly runs out of cases since there are no hierarchies between 'variables'.

3.2.3 The New South Wales Sentencing Information System

The development of this system began in 1988. It includes information on penalty statistics, sentencing law and information about appeal case judgements. The system is administered by the Judicial Commission of New South Wales. In addition to managing the SIS the commission is also responsible for judicial education and training and monitoring judicial conduct. The commission is chaired by the Chief Justice. One of the main functions of the Commission is to assist judges to achieve consistency in sentencing without placing any limits on the discretion that a court has in determining sentence.

Although there has been no systematic evaluation of the SIS, reports from the Judicial Commission indicate that the system has been well received by users and that there has been a steady growth (at least as measured by the number of 'log-ins') in the use made of the SIS (Potas 1991), (Chan 1991) (Spears 1993).

3.2.4 The Scottish Sentencing Information System (i)

The initiative for this project came from the Lord Justice-Clerk, the second most senior judge in Scotland, who had seen the New South Wales system demonstrated in Canada at a Conference of the Commonwealth of Learning and decided that it might be useful to have a similar system for the High Court of Justiciary. The Lord Justice Clerk approached the Law School at the University of Strathclyde where there was existing relevant expertise. The Scottish Office provided funding for a feasibility study which ran from June 1993 to January 1995. The operation and description of the taxonomic issues and how the resolution of such issues was approached is documented elsewhere (Hutton, Tata and Wilson 1995).

After due consideration of the report of the feasibility study, the Scottish Courts Administration is currently funding (from November 1995) the first phase in the implementation of the prototype. This will involve the delivery of the system to up to thirteen judges and on-going study by the judges of its utility so as to identify weaknesses and areas needing further development. This first phase of implementation will also begin conceptual work on an Appeal Court Judgements Database to be subsequently incorporated into the SIS.

3.2.4 The Scottish Sentencing Information System (ii)

In its White Paper (SOHHD 1994), the government welcomed the Sentencing Information System initiative and indicated that if it was workable and successful, similar systems might be considered for the lower courts. It would therefore appear that there is some government support for sentencing information systems. While the future of the SIS seems to enjoy generally strong judicial support there are perhaps two broad issues which will require further attention.

First, unlike the New South Wales SIS, an operational Scottish SIS may well need to enjoy direct institutional protection. There does not yet appear to be any long-term plan for the system to be run by a government body directly responsive to judicial needs. In New South Wales, on the other hand, the SIS is administered by the Judicial Commission of New South Wales. In Scotland, however, the possible absence of a government body with a direct interest in defending the system from criticism and resource pressure may leave it vulnerable.

3.2.4 The Scottish Sentencing Information System(iii)

Secondly, given this possible future vulnerability to resource pressures, it may be difficult to lobby for funding to improve and expand the system in response to judicial requests for more or different information. As with any computer system, the SIS will require to evolve and improve over time. The apparent ability of the New South Wales system to do this may help to explain its longevity and apparent success compared with its North American forebears.

One of the most intriguing questions which will need to be explored will be the extent to which the introduction of information technology will change the nature of sentencing practice. Although there may be no formal requirement for judges to consult the system and take note of its information, there may be reasons for expecting that judges may feel informally encouraged to use the system and indeed that it may significantly impact on judicial sentencing behaviour. (For further explanation of this point see Hutton, Tata and Wilson 1995).

4 How can the representations of knowledge enhance the utility of decision-support systems for sentencers?

Aside from questions of political support and the relationship between information systems and judicial culture, a key question must concern the conceptual character of a system intended to support decision-making so strongly characterised by formal discretion.

4.1 Sentencing as a comparative process

The Uniqueness of Individual Cases

As we saw earlier, systems to support sentencing decision-making tend to be based on the idea that a decision about a current case can be aided by information about previous decisions for 'similar' cases. Of course, the ability to produce a system which is meaningful to sentencing decision-makers must presuppose the ability of the system, and those who programme it, to identify cases as 'similar'. Is this possible ? It is a judicial axiom of sentencing that every case is unique and as such each must be judged on its own facts (Ashworth, Genders, Mansfield, Peay and Player 1984). Further, it is argued that because each case is unique it is therefore not possible to compare one case with another. We have argued elsewhere that this view is untenable (Hutton, Paterson, Tata and Wilson 1996a). Moreover, the illogicality of the view asserting that every cases is incomparable was exposed by Hood (1962) writing over twenty years ago :

'...[M]agistrates and judges frequently turn to precedent for their ruling and place particular value on their experience in sentencing. Now, if this experience is to be of value, then all cases cannot be unique, they must be comparable at least in some respects; and even if it is agreed that all cases are unique in some sense, this cannot be decisive in the practice of sentencing, for frequently decisions are reached with the aid of "experience".' (Hood (1962), p16,).

Although in one sense every case is unique (because each individual, social situation and context are unique), it is possible, at least in principle, to compare cases and therefore to say that some cases are 'similar'. The critical question, then, must be: 'how is 'similarity to be understood?' Traditional academic understanding has largely been informed by 'legal analysis'. Attempts to construct systems of computer support for sentencing decision-making have tended to be rooted in a paradigm which is primarily legalistic in outlook and analytical in its thinking. We would argue that systems based on this paradigm offer at best only a partial understanding and therefore computerised representation of 'similarity'.

4.2 The Weakness of the Legal-Analytical Paradigm

What are the characteristics of the 'Legal-Analytical' paradigm as it attempts to represent 'similarity' and what limitation do they have? We address this question below by considering each aspect of 'legalism' and 'analysis' in turn.

4.2.1 Legalism (i)

In developing decision-support systems for judicial sentencing, the starting point for representations of 'similarity' has tended to be a legalistic one (Bainbridge 1991, Gruner 1991, Potas 1991, Doob 1990). Systems search for 'like' cases or representations of 'like' cases primarily in terms of official legal convictions. This is normally the starting point of the search for 'similarity'. However, in designing the Scottish SIS we became increasingly doubtful of the appropriateness of this approach. From our work with judges (Hutton, Paterson, Tata and Wilson 1996a) and previous research with sentencers (Ashworth et al 1984; Hogarth 1971; Hutton and Tata 1995) there was strong reason to doubt the view that judges think primarily in terms of classifications represented by official headline categories of the official criminal law 'headline' offence.

Previous research (Fitzmaurice and Pease 1986, Ashworth et al 1984, Hood 1962, Hood 1992, Hutton and Tata 1995), shows that sentencing is determined by far more than the legal definition of the offence(s). Judges consider the circumstances surrounding the commission of offences as important. This point is applicable to both common law and statutory jurisdiction.

4.2.1 Legalism (ii)

As in other countries where sentencing is conducted in a predominantly common law jurisdiction, the strict common law headline offence category often provides little information as to the seriousness of the offence from the perspective of sentencing. A sentencer must not only consider the common law conviction, but also the circumstances of the events surrounding the commission of the offence. The 'headline' offence(s) (for example, 'Robbery', 'Rape', 'Housebreaking' 'Theft' etc.), with which an offender is charged and convicted may be of limited relevance in deciding sentence. Very frequently, the common law headline conviction does not provide a sentencer with sufficient information about the circumstances surrounding the commission of the offence.

If common law offence classifications provide insufficient information about the circumstances of an offence and its attendant seriousness, it might be thought that statutory based offences provide sufficient precision for the consideration of sentence. After all, could the organisation of information not simply replicate all offences as they appear in Acts of Parliament? Aside from the practical considerations of operating such a system, there is a conceptual difficulty. While the creation of offences on the statute books are used to charge persons, they may often of limited assistance when judges consider sentence. It is not clear how to organise statutory offences. (For further elaboration of this point, see Hutton, Paterson, Tata and Wilson 1996).

Both our knowledge of the empirical literature on sentencing (Fitzmaurice and Pease 1986, Ashworth et al 1984, Wilkins et al 1978) and consultation with judges strongly suggested that the only sensible course of action would be to arrange offence information in a way which is relevant to the purpose of sentencing. This might consequently mean that the arrangements of offences would not necessarily reflect the divisions within the criminal law.

Thus the assumption that a system's taxonomy should begin with the official criminal law headline offence category is very doubtful. Rather, the single most important criterion which judges consider is not the official criminal offence category, but its relative seriousness (Wilkins et al 1978, Fitzmaurice and Pease 1986, Hutton and Tata 1995). Although decision-support systems for sentencing have recognised the need to incorporate information relating to offence seriousness, the organisation of that information has still tended to priviledge a legalistic starting point.

4.2.2 Analysis (i)

While representations of similarity have tended to assume a formal legalistic starting point, there has been a recognition that there needs to be some account taken of 'aggravating and mitigating factors' which describe the seriousness of the headline conviction. Typically, having selected the 'appropriate' criminal law classification the judge is then invited to add in standard aggravating or mitigating factors (Chan 1991; Hogarth 1988; Murbach and Nonn 1991). What is striking about this approach is the additive and analytical nature of the representation of similarity. We would suggest that this analytical process of abstracting additive 'factors' from the whole case is an artificial representation of similarity. The abstraction of independent factors which purport to describe seriousness denies the relational meaning of information in a case. It assumes that a case can be meaningfully fragmented into discrete and abstract individual factors which can be analysed as if they each have a power independent of each other over the decision process. However, 'aggravating and mitigating factors' only make practical sense to the sentencer in relation to each other and to the whole case (Shapland 1981).

4.2.2 Analysis (ii)

As an illustration, let us take one issue which has received surprisingly little attention : cases where an offender is convicted of more than one charge. The legal-analytical approach would try to record each conviction separately, or, just one 'main' conviction. However, this does not appear to be they way that sentencers tend to think about cases (Hutton, Paterson, Tata, and Wilson 1995). In considering sentence, judges do not seem to think about each conviction in a multi-charge indictment in isolation from each other and from the circumstances of the commission of the offences. Rather, they tend to view the case as a whole incident or narrative of events (Alschuler 1991; Parton, Hansel, and Stratton 1991). Therefore, in attempting to reveal the process of the decision-making of sentencers, empirical research has used sentencing vignettes (Ashworth et al 1984, Corkery 1992) rather than necessarily relying on criminal law classifications which are necessarily limited in the relevant information which they can offer the sentencer.

While the legal-analytical way of representing 'similarity' of cases for the purposes of sentencing may be limited, is there a possible alternative conception ? Below we suggest a possible alternative.

4.3 The need for a more schematic and holistic representation of 'similarity' (i)

Judges frequently turn to the cumulative and evolved wisdom of judicial experience represented by precedent for their decisions and also place particular emphasis on their personal experience. This 'experience' helps judges to interpret and schematise the mass of information with which they are presented. In developing this argument, let us consider research into discretionary legal decision-making more generally.

It has been established by cognitive psychology that humans are limited processors of information. From his field experimental research into the psychology of decision-making in criminal sentencing, van Duyne (1987) found that despite the considerable discretion which it afforded, prosecutorial decision-making could be characterised as "...one dimensional: the Prosecutor selected out of the total information on the case only those aspects which were consistent with a particular 'dimension' (e.g. 'professional', 'social misfit' or rehabilitation) and fitted these into simple conceptual schema." (At p.147). Although the prosecutor may make a few changes, the decision process remains basically the same. Giller and Morris (1981) found that social workers use 'operational philosophies' (the means by which professional ideologies are mediated through the demands of practice). "Having located the moral character of the case, the social workers were able to respond with a repertoire of provisions which routinely met the case as portrayed. In this way, social work with offenders became ordered and rational and a work priority was established. Decisions were not 'made'; they emerged as natural logical, even inevitable, responses to the social worker's interpretation of the case. But these interpretations were part of a dialectic. What 'explained' the moral character of the case also provided evidence of its nature." [pp 79 - 80]. This finding is also confirmed by research into problem-solving by other professionals, such as doctors who have to make diagnoses. For example, Elstein et al (1978) found that doctors appeared to do a simpler job than they imagined, making use of simple conceptual schema. Van Duyne (1987) concludes that the results from his study cast some doubt on the claim that,

"...sentencing is a highly complex process solely directed to 'unique' decisions in 'unique' cases. Without suggesting that the job of judges is just a matter of quick guesswork and routine, gilded with pomp and ceremony, I would like to state that judicial decision making is comparable to other kinds of open problem solving (2) such as the grading of works of art by a viewing committee or the marking of an essay by a teacher. If sentencing is difficult, it is because of its uncertainty, not because of its complexity...In problem solving, [it is] necessary to restrict the information to one manageable dimension in order to avoid overload and uncertainty (Nesisser 1976). There is no reason to assume that judges and prosecutors are a complex subdivision of the human race deserving a theory of their own." (pp 151-152).

4.3 The need for a more schematic and holistic representation of 'similarity' (ii)

The supposed model that decisions are in practice taken through a legal-analytical process characterised by deductive linear reasoning is doubtful. Rather like other problem-solvers, lawyers and judges make use of experience to help them to schematise new cases. For example, Crombag, Wijkerslooth and van Serooskerken (1975) asked experienced legal problem-solvers to think aloud while solving a concrete problem. They found that,

"[t]he most striking result was that what was said while thinking aloud created a rather chaotic and unsystematic impression. Often a person seemed to have a solution, although a provisional one, at an early stage for which he subsequently tried to find supporting arguments. Moreover, during the reasoning process, the subject did not seem to complete one part after another, but rather to jump wildly back and forth" (p.169).

4.3 The need for a more schematic and holistic representation of 'similarity' (iii)

Crombag, Wijkerslooth and Serooskerken argue that skilled legal problem solvers are like other skilled problem-solvers in that they tend to have a provisional solution in place early and to work backwards. Curiously, however, Crombag, Wijkerslooth and Serooskerken describe these results as "disappointing"(p.169). Had they, perhaps, hoped that their experienced legal problem-solvers might demonstrate a more legal-analytical process displaying the supposed virtues of linear reasoning derived from first principles ?

There is perhaps further reason to suppose that judges may not be very different from other experienced problem-solvers in coming to a broad view and then finding reasons to justify it. The background experience of most judges in most adversarial law jurisdictions is a career as lawyer. One of the main problems which the lawyer has to solve routinely is to try to achieve a favourable result for his client, (or, as in the case of advocates for his solicitor's client). The instrumentally-rational craft of lawyering demands the location, understanding, interpretation and marshaling of information towards a specific goal. It may be reasonable to suppose that that the experience of solving problems in this practically efficient way continues to be useful to the lawyer who finds himself elevated to the bench. After all few, if any, common law jurisdictions require judges to begin the sentencing process from a process of pure legal-analysis. Rather, judges are encouraged to use their practical career experience in coming to a decision. As Ashworth et al (1984) observe from their interviews with judges: "Most judges described [the sentencing decision] as an instinctive process, using such terms as 'instinct', 'experience', 'hunch' and 'feeling'". The giving of publicly-declared reasons for a decision tend to be couched in legal analytical terms. Has this more to do with the necessity to fulfill a popularly-held conception of justice than a revelation of the stream of consciousness in the decision process? "If [official] reasons [for decisions] have a place in court, it is because they are defensible, not because they are true." (Fitzmaurice and Pease, p.45).

Therefore, a more schematic and holistic approach, (which the Scottish SIS has labelled as 'The Modified Approach'- see Hutton, Tata and Wilson 1995), attempts to classify different types of criminal conduct exclusively from the perspective of sentencing. In the shape of this more schematic-holistic approach there may be the ability for the SIS to more accurately capture the practical nature of legal problem-solving than linear analytical legal theory has supposed. The idea behind this approach is to attempt to derive classifications from the mental schema around which judges, like all decision-makers, tend to operate (van Duyne 1987). Thus, rather than being more simplistic than strict criminal law classifications, this approach may aggregate or split legal constructs.

4.4 Is the schematic-holistic approach simplistic? (i)

It might be objected that the notion of a computer system which retrieves cases on the basis of a very small number of selections is too crude. Since it does not rely on official criminal law categories as a necessary starting point, it might be argued that such an approach overly-simplifies the intricate knowledge represented in each case by the criminal law. The attempt to produce such a typological representation of information about offences may be thought to be overly simple since the very idea of 'types' of cases denies the delicate and subtle complexities of the 'knowledge' represented by the criminal law. However, to make such an argument is to lose sight of the fact that this officially defined 'knowledge', or 'reality' is itself constructed for a particular purpose: securing convictions and acquittals. All criminal convictions are themselves typologies (Sudnow, 1964). Categories of criminal convictions do not, and cannot, reflect the full precise occurrences of 'what happened' during a criminal incident or series of related incidents. Rather, the criminal conviction for which a person may be sentenced is necessarily the result of a simplified and normalised representation of events. (Shapland 1987, Curran and Chambers 1982, Giller and Morris 1981).

4.4 Is the schematic-holistic approach simplistic? (ii)

Let us take just one point in the criminal process as an example: the drafting of charges by the prosecution. Charges are drafted in a variety of ways which have the effect of continuing to standardise and normalise reports of relatively complex human behaviour. Charges are "...drafted in a standard way commonly referred to by [prosecuting lawyers] as a style...[and it is a]...process of assimilating prosecutorial norms and rendering the unfamiliar familiar..."(Moody and Tombs, 1982 pp. 53-54). Both Shapland (1981) and Ashworth et al (1984) suggest that what legal rhetoric refers to as the 'facts of the case' should not properly be regarded as 'facts' but as "...constructions of the circumstances forming the offence, stemming from the constructions made by the police (themselves very much negotiable and influenced by the individual views of detectives and the wishes of victims and witnesses) (Ericson 1981, Ericson 1982)." (Shapland 1987). This 'information throughput' model, reveals that the information with which a sentencer is presented has already been pragmatically negotiated and refracted through a series of legal and oragnisational lenses. It can never provide a notionally objective and factual account of human behaviour, but is necessarily a pragmatically constructed representation of it. The nature of criminal events and criminals may be infinitely unique but the nature of their legal representations from which a sentencer must make a decision are necessarily finite, typical and schematic. As an approach which is specifically designed for the purpose of classifying information from the exclusive perspective of sentencing, it may be a more sensitive taxonomic tool than that designed by the criminal law and the addition of abstracted 'factors'.

4.4 Is the schematic-holistic approach simplistic?(iii)

The Scottish SIS also employs a 'Principal Offence Approach'. This approaches the retrieval of information about a case involving more than one conviction by beginning with the most serious (or principal) offence and then adding in further information (including further convictions) about the offence-related circumstance of the case This has tended to be a normal way of approaching the building of an SIS (Hutton, Tata and Wilson 1995). Although this approach does not necessarily begin from a strictly legalistic starting point, it is fundamentally analytical in nature. Although it has may attempt to overcome the question of multiple convictions by positing a single offence as the most serious it then 'adds in' information characterising that offence. The Modified Approach on the other hand perhaps has begun to suggest that it is possible to represent the schematic and holistic process of discretionary decision-making The Scottish SIS project has only scratched the surface of the kind of questions which research needs to explore if it is to be the basis of decision-support systems. Indeed, if, as we have suggested, a schematic-holistic approach may fit more closely with the reality of discretionary (legal) decision-making than traditional legal analysis, then the development of the Scottish SIS will itself have to explore these research questions further.

4.5 Is a schematic-holistic approach open to modelling by a computer? (i)

Judges stress their treatment of cases as a 'whole' and the 'feel' for individual cases. There has been judicial resistance to the attempt to represent sentencing according to some mathematical model associated with the addition and subtraction of 'independent' 'factors'(Lovegrove 1989, Kapardis 1987, Wilkins et al 1978). Does this mean that the schematic-holistic approach which has been suggested is too informal, fluid and intangible to be modelled by a computer ?

4.5 Is a schematic-holistic approach open to modelling by a computer?(ii)

In trying to answer the question of whether computers can understand legal reasoning, Tito (1987) succinctly states that, "[c]omputers can only do what humans program them to do". However, Tito presents two views of how lawyers decide : "[Do lawyers]...apply rules to decide when two situations are similar within a legal context? Or, do lawyers just get a 'hunch' or have a flash of insight that alerts them to similarities?"[original emphasis retained](p.411) Tito presents the latter as "ruleless" and therefore not amenable to understanding by a computer because they are not understandable to humans. However, we would suggest that it is possible to argue and that the Scottish SIS project has begun to demonstrate that legal decision-making may be both felt to be based on an 'instinctive hunch' and amenable to rules. Like other legal-analysts, Tito's conception of 'rules' seems to ignore the possibility of non-legal rules of behaviour (Hawkins 1992). However, legal decision-making can be both strongly characterised by formal discretion and also produce decisions that are patterned, predictable and comprehensible (Baumgartner 1992). Thus, any model of sentencing decision-making on which decision-support system is based should attempt to seriously study informal behavioural rules. We have suggested that this may uncover a more holistic, schematic model rather than a legal-analytical one.

4.5 Is a schematic-holistic approach open to modelling by a computer?(iii)

Thus, it may not be the idea itself of a computer model of decision-making based on some notion of 'rules' that presents a difficulty, but the normally assumed interpretation and representation of 'rules' as formal canon. Further research examining informal behavioural processes and patterns, and their inter-relationship with the formal context, is more likely to help to understand the 'rules' of discretionary decision-making.

References

Ashworth, A. (1987) 'Devising Sentencing Guidelines for England' in Wasik and Pease (eds) Sentencing Reform : Guidance or Guidelines ? (Manchester University Press, Manchester.)

Ashworth, A. (1992a) Sentencing and Criminal Justice, (Weidenfeld and Nicholson, London).

Ashworth, A. (1992b) 'Sentencing Reform Structures' in Tonry, M. (ed), Crime and Justice: A Review of Research volume 16 , (University of Chicago Press, Chicago and London.)

Ashworth, A., Genders, E., Mansfield, G., Peay, J., and Player, E., (1984) 'Sentencing in the Crown Court: Report of an Exploratory Study', (University of Oxford Centre for Criminological Research, Occasional Paper No 10.)

Ashworth, A., and Gibson, B. (1994) 'The Criminal Justice Act 1993 - Altering the Sentencing Framework' (1994) Criminal Law Review, 101.

Bain, W. (1989) 'Judge' in Christopher K. Reisbeck and Roger C. Schank (1989) ' Inside Case-based Reasoning ', (Lawrence Erblaum Associates, New Jersey).

Bainbridge, D. (1991) ' CASE: Computer Assisted Sentencing in Magistrates' Courts , (Paper presented to the BILETA Conference 1991.)

Bench-Capon, T.M.J. (1994) 'In Defence of Rule-based representations for Legal Knowledge-based systems', Law, Computers and Artifical Intelligence ,1, pp.15-28.

Baumgartner, M.P. (1992) 'The Myth of Discretion', Hawkins (1992)(ed)

Canadian Sentencing Commission (1987) Sentencing Reform: A Canadian Approach .

Chan, J. (1991) 'A Computerised Sentencing System for New South Wales Courts' Computer Law and Practice , 137

Corkery, J.M. (1992): 'The Use of Vignettes in Sentencing Studies of English Magistrates', Interantional Journal of the Sociology of Law , 20, pp253-270

Crombag, H.F., De Wijkerslooth, J.L., Van Tuyl Van Serooskerken, E.H. (1975): 'On Solving Legal Problems', Journal of Legal Education, Vol.27, pp 168-201.

Curran, J. and Chambers, G. (1982): ' Social Enquiry Reports in Scotland ', HMSO, Edinburgh.

Doob, A. (1990) Sentencing Aids: Final Report to the Donner Canadian Foundation , (Centre of Criminology University of Toronto: Toronto.)

Doob, A. and Park, N. (1987) 'Computerised Sentencing Information for Judges: An Aid to the Sentencing Process' Criminal Law Quarterly, vol 30, 54.

Ericson, R. (1981) Making Crime: A study of Detective Work , (Butterworth, Toronto.)

Ericson, R. (1982) Reproducing Order: A study of Police Patrol Work ,( University of Toronto Press, Toronto.)

Fitzmaurice, C. and Pease, K. (1986) The Psychology of Judicial Sentencing , (Manchester University Press, Manchester.)

Freed (1992) 'Federal Sentencing in the Wake of Guidelines : Unacceptable Limits on the discretion of Sentencers' Yale Law Journal, Vol 101, No. 8, 1681.

Giller, H. and Morris, A. (1981) ''What Type of Case if This?' Social Workers' Decisions about Children Who Offend' in Adler, M. and Asquith, S. (eds) (1981) Discretion and Welfare (Heinemenan Educational Books, London.)

Gruner, R. (1991) ' Sentencing Advisor: An Expert Computer System for Federal Sentencing Analyses ', Paper presented at BILETA Conference.

Hassett, P. (1993) 'Can Expert System Technology Contribute to Improved Bail Decisions?', International Journal of Law and Information Technology ,2.

Hawkins, K. (1992) 'The Use of Legal Discretion: Perspectives from Law and Social Science' in Hawkins, K. (1992)(ed) ' The Uses of Discretion', (Oxford Socio-Legal Studies Clarendon Press, Oxford.)

Hogarth, J. (1971) Sentencing as a Human Process , (University of Toronto Press in association with the Centre of Criminology University of Toronoto, Toronto)

Hogarth, J. (1988) Sentencing Database System: User's Guide , University of British Columbia, Vancouver.

Hutton, N. (1995) 'Sentencing, Rationality, and Computer Technology' Journal of Law and Society 1995, Vol. 22, No. 4.

Hutton N., Paterson A., Tata C., and Wilson J. (1995a) ' A Prototype Sentencing Information System for the High Court of Justiciary: Report of the Study of Feasibility ' (Scottish Office Home and Health Department, Central Research Unit: Edinburgh. Forthcoming)

Hutton, N. Paterson, A., Tata C., and Wilson, J. (1995b) ' Decision Support for Sentencing in a Common Law Jurisdiction ', The Fifth International Conference on Artificial Intelligence and Law: Proceedings of the Conference, Maryland.

Hutton, N. and Tata, C. (1995) Patterns of Custodial Sentencing in the Sheriff Court , Scottish Office Home and Health Department Central Research Unit, Edinburgh, forthcoming.

Hutton N., Tata C., and Wilson J. (1995) 'Sentencing and Information Technology: Incidental Reform ?' International Journal of Law and Information Technology 1995 , Vol.2, No.3.

Jareborg, N. (1994): 'The Sweedish Sentencing Law', ' European Journal on Criminal Policy and Research ' Vol 2, No. 1.

Lovegrove, A. (1989) Judicial Decision Making, Sentencing Policy and Numerical Guidance, (Springer-Verlag, New York.)

Moody, S.R. and Tombs, J. (1982) Prosecution in the Public Interest , (Scottish Academic Press, Edinburgh.)

Murbach, R. and Nonn, E. (1991) ' Sentencing by Artificial Intelligence Tools: Some Possibilities and Limitations' Paper presented at The Joint Meeting of the Law and Society Association and the Research Committee of the Sociology of Law of the International Sociological Association Amsterdam 1991.

Parton, D. A., Hansel, M., Stratton, J. R. (1991): 'Measuring crime seriousness: lessons from the national survey of crime severity', The British Journal of Criminology , Vol. 31, pp 72-85.

Parker, H., Sumner, M., and Jarvis (1989) Unmasking the Magistrates , (Open University Press, Milton Keynes.)

Kenneth Pease, Judith Ireson, Stuart Billingham and Jennifer Thorpe (1977) 'The Development of a Scale of Offence Seriousness' in International Journal of Criminology and Penology , 5, pp 17-29

Potas, I. (1991) ' The Sentencing Information System of New South Wales: Promoting Consistency in Sentencing Through Computerisation ', Paper presented to The Commonwealth Magistrates and Judges Association Sydney 1991.

Schild, U.J. (1995) ' Intelligent Computer Systems for Criminal Sentencing ', The Fifth Inetrnational Conference on Artificial Intelligence and Law: Proceedings of the Conference, Maryland 1995, pp229-238.

Shapland, J. (1981), Between Conviction and Sentence: the Process of Mitigation , (Routledge & Kegan Paul.)

Shapland, J. (1987) ' Who Controls Sentencing ? Influences on the Sentencer ' in Donald C. Pennington and Sally Lloyd-Bostock (eds) (1987) The Psychology of Sentencing: Approaches to Consistency and Disparity , (Centre for Socio-Legal Studies, Oxford.)

Simon, E. and Gaes, G. (1989) ' ASSYST - Computer Support for Guideline Sentencing ', The Second International Conference on Artifical Intelligence and Law: Proceedings of the Conference, Vancouver 1989, pp195-200

Simon, E., Gaes, G., and Rhodes, W. (1991) ' ASSYST- The Design and Implementation of Computer Assisted Sentencing ', Federal Probation, vol 55 46.

Spears, D. (1993) Providing Computerised Sentencing Information to Judicial Officers: The New South Wales Experience , NSW Judicial Commission.

Sudnow, D. (1964) 'Normal crimes: sociological features of the penal code in a public defender office', ' Social Problems ' 12, pp 255-276.

Celeste Tito (1987) 'Artificial Intelligence: Can Computers Understand Why Two Legal Cases are Similar ?', Computer/Law Journal , Vol 7, pp 409-437

Tata, C., Hutton, N., Wilson, J., and Paterson, A. (1995a) 'Sentencing and Computerised Information Systems: the Process and Progress of the Scottish Prototype', Juridical Review , 4, forthcoming.

Tonry, M. (1987) ' Sentencing Guidelines and Sentencing Commissions: The Second Generation ' in Wasik and Pease (1987) (eds) Sentencing Reform: Guidance or Guidelines ? (Manchester University Press, Manchester.)

Tonry, M. (1992) ' Judges and Sentencing Policy - The American Experience ' in Munro and Wasik (eds)(1992) Sentencing, Judicial Discretion and Training , (Sweet and Maxwell: London.)

Tonry, M. (1993) 'The Failure of the U.S. Sentencing Commission's Guidelines' Crime and Delinquency 131.

Van Duyne, P. (1987) : ' Simple Decision Making ' in Pennington, D.C. and Lloyd-Bostock, S. (1987)(eds) : ' The Psychology of Sentencing: Approaches to Consistency and Disparity ', (Centre for Socio-Legal Studies, Oxford)

Victoria Attorney General's Department (1988) Victorian Sentencing Committee Report , Melbourne.

Von Hirsch, A. (1976) Doing Justice: The Choice of Punishments , (Hill and Wang: New York.)

Von Hirsch, A. (1993) Censure and Sanctions , (Clarendon Press, Oxford).

Von Hirsch, A. and Jareborg, N. (1989) 'Sweden's Sentencing Statute Enacted', Criminal Law Review , 275.

Leslie T. Wilkins, Jack M. Kress, Don Gottfredson, Joseph C. Calpin, Arthur M. Gelman (February 1978) ' Structuring Guidelines: Structuring Judicial Discretion. Report on the Feasibility Study ', U.S. Justice Department

Zeleznikow, J., and Hunter,D. (1994) Building Intelligent Legal Information Systems, Computer Law Series 13 , Kluwer Law and Taxation Publishers.

Footnotes

1 Hutton (1995) has argued that the sentencing decision is characteristically substantive and irrational; as opposed to formal and rational.

2 An 'open problem' is characterised as one where the problem-solver cannot objectively and irrefutably determine whether his or her solution is the correct one.