=Paper=
{{Paper
|id=Vol-482/paper-1
|storemode=property
|title=Intelligent Evaluation of Traffic Offender Records
|pdfUrl=https://ceur-ws.org/Vol-482/Paper_1.pdf
|volume=Vol-482
|authors=Uri J. Schild,Ruth Kannai
|dblpUrl=https://dblp.org/rec/conf/icail/SchildK09
}}
==Intelligent Evaluation of Traffic Offender Records==
https://ceur-ws.org/Vol-482/Paper_1.pdf
Intelligent Evaluation of Traffic Offender Records
Uri J. Schild1and Ruth Kannai2
1
Department of Computer Science, Bar Ilan University,
Ramat Gan 52900, Israel, schild@cs.biu.ac.il
2
Faculty of law, Bar Ilan University,
Ramat Gan 52900, Israel, kannair@mail.biu.ac.il
Abstract: This paper describes an intelligent computer system giving
decision support in the area of sentencing of traffic law offenders. The system
evaluates the previous record of a traffic offender, and suggests how to
consider that record when passing sentence in a new traffic case.
Keywords: intelligent evaluation, intelligent decision support system (DSS),
sentencing, traffic law offenders.
1. Introduction
Previous work by us considered the intelligent evaluation of an offender's previous
record in the general area of criminal law [1, 2]. The object of that work was to
develop an intelligent decision support system (DSS) to help judges (and perhaps other
parties in the legal system) to evaluate the previous, general criminal record of an
offender, i.e., a person that had been found guilty of some offence. Such an evaluation
would be of help to the judge about to pass sentence on the offender. No other work
has been carried out on this particular subject.
During that work we considered the possibility of doing similar work on
traffic offenders. Intuitively a DSS for this domain might have a different form, as the
issues to consider are different than in the general criminal area, but then, perhaps not.
Another question that presented itself was to which extent there is a connection
between an offender's general criminal record and his traffic offence record. This
paper describes the results of our work on the new DSS for evaluating a traffic
offender's previous record.
The purpose of the system is not to suggest any kind of sentence for the
offence at hand, but to evaluate the offender's previous record, and suggest the weight
this record should be given in the sentence in the present case.
2. Background
When the judge is about to pass sentence, he can in theory take many factors into
account. In practice he will consider only some of these, namely those that have been
salient in the case at hand. These factors will then have an aggravating or mitigating
1
U. Schild and R. Kannai
influence on the sentence. One of the factors a judge will often consider is the
offender's previous record. It is believed by many that the record is of importance and
should carry weight. Thus features like the increase or decrease in the severity of past
offences and the time-intervals between consecutive offences ought to bear influence
on the sentence in the present case.
What happens in practice in the Israeli courts (and presumably in courts all
over the world) is the following scenario: After an accused has been pronounced
guilty, the prosecutor hands the judge the "sheet", i.e., the record of previous
convictions. This record is a hardcopy printout of the entire record stored in the central
Israeli police computer relating to the offender.
There is a practical problem with the previous record: The record is often
quite extensive, containing a long list of past offences, which may all be of the same
type but often include related types of crimes, or even entirely different types of
crimes. The record may also span a considerable number of years. The judge can have
great difficulty in acquiring a clear picture of the situation, and he must necessarily
devote a lot of time to the interpretation of the record. This time is often not available,
and the sentence may therefore not reflect the facts embedded in the past record.
What has been described so far holds for general criminal cases and for traffic
offences. There are, however, also some important differences:
1. Traffic offences are usually considered less serious than general criminal offences.
The public believes that everybody could be involved and found guilty of a traffic
offence, not just professional criminals.
2. The sentences handed out in traffic cases are usually much lighter. Traffic offences
only very seldom lead to custodial sentences. The customary sentences are monetary
(fines and reparation) and driving disqualification. Often the sentences are deferred
(suspended), being applied only in the case of repeated offences within a certain period
of time.
3. The public believes that the previous record of traffic offences is of extreme
importance. The judges do not all agree, but they are under great pressure from the
media. It is a common belief that the previous record ought to have a dominant
influence in determining the sentence in the case at hand. The media is happy to
publish and point out whenever it is believed that some traffic offender with a large
number of previous offences gets off with what is considered too light a punishment.
4. The previous record of a traffic offender submitted in a traffic court exclusively
contains traffic offences. Only if the offender has a relevant general criminal record (or
perhaps in the case of a professional criminal) will a separate printout of the general
criminal record be submitted by the prosecutor.
5. The computer printout of an offender's previous record is very hard to read. It is
almost impossible to understand for the uninitiated. This of course is not of great
importance, as judges, prosecutors and defense lawyers become familiar with the
layout over time.
However, even an experienced judge does not have the time to go through,
say 100 previous offence records to see whether how the offender has behaved himself
in traffic after receiving previous suspended sentences.
2
Intelligent Evaluation of Traffic Offender Records
3. Our System: Presentation of Basic Data
From our description of the computer record in the previous section it is clear that the
first step in building a DSS must be to present the previous record in a clear manner.
This will serve two purposes: (1) It will enable legal practitioners to carry out a speedy
overview of the record, (2) It will enable them to proceed to the second step: An
intelligent analysis of the record. In order to carry out step (1) we spent a large effort
interviewing legal professionals involved in reading such records: Judges, lawyers and
police officers.
There is no Artificial Intelligence in this part of the system. Applying basic
principles of modern interface design [3] and after several iterations with the legal
experts, we have reached a way of presenting the previous record in a way that is
easily and speedily overseen.
Figure 0 in the appendix shows the original printout from the police
computer. One can imagine how difficult it would be even for a legal professional (a
judge, a prosecutor, a defence lawyer or a police officer) to survey such a record if it
contains, say, 100 items.
Our assumption is that a user should be able to become familiar with even an
extensive past record should take three seconds! Surveying details should take another
three seconds. Figure 1 gives brief overview of who the offender is, and what he has
done in the past (three seconds). Figure 2 shows what Figure 0 would look like in our
system 1 (perhaps another three seconds).The colour code enables the user to get an
immediate impression of the different types of offences
If the user has more time - one can imagine a lawyer preparing himself for the
present case, or a police officer wishing to estimate the dangerousness of somebody he
has stopped on the road - more information is available.
Figure 3 is a graph showing the sentences given in the past: Periods of
Disqualification and Fines. Sentences are often combined: Disqualification + Fine, etc.
It would be nice if one could present such a combined sentence in one graph. This is
impossible, one cannot compare apples and oranges, and one cannot say that 3 months
disqualification is more serious than, say, a NIS 10,000 (US$ 3,000) fine. So we
decided to show two graphs in the same screen.
The system interface was established by asking the experts a set of pre-
formulated questions. For example:
1. What is wrong, impractical and/or not user-friendly in the old police output?
2. What are you looking for and in which order?
3. Are there data you would like to see sorted in various orders (e.g. dates)?
We did not ask whether there was additional data the experts would like to
see, even though this seems to be an obvious question. As mentioned above, the
printout of the previous record today includes what is stored about the offender in the
police computer. Obtaining additional information would call for a major overhaul of
police procedure and perhaps the information systems of the entire justice
organisation. It would also raise questions of legality of what information the
1
Obviously all records, computer printouts and screens are in Hebrew. We hope to have them
translated (at least partially) before the workshop.
3
U. Schild and R. Kannai
government should be allowed to keep in its computers, and would certainly
necessitate new legislation.
The Knesset (Israel's parliament) is aware of such questions and problems. It
has formed an external committee (chaired by one of us - R. Kannai) to consider the
kind of questions raised above with respect to all kinds of offenders, traffic and
otherwise..
In the theory of expert systems it is well-known that different experts come
up with different answers [4]. Sometimes experts outright contradict each other. This
phenomenon was indeed observed by us with respect to the layout. The solution was
simple (but a bit tricky): We chose the answer that was proposed by the majority.
What then invariably happened was that at the next iteration the experts found the
solution acceptable - also the ones who initially suggested other approaches.
4. Our System: The Intelligent Component
4.1 Preliminaries
In this section we shall deal with two issues: (1) The complexity of the
problem, (2) What kind of system to aim for.
4.1.1 The Complexity of the Problem
The intelligent component of the system aims at analysing the previous record in order
to determine the presence and extent of certain factors. These are the factors that
influence the decision of the judge in passing sentence in the case at hand.
It was clear to us at the beginning of the project, that a sizable amount of
specific domain knowledge would be necessary. The problem of how to evaluate an
offender’s previous record is far from trivial, even for humans. We shall give just a
few examples of the complexity of evaluating a previous record:
1. A person is about to be sentenced for speeding in an urban zone. His past
record shows a large number of convictions for parking offences. Should judges take
such past offences into account? (A case like this would come to court only in extreme
cases).
2. A person is about to be sentenced for speeding in an urban zone. He has but
one previous conviction, also for speeding in an urban zone. However, that previous
offence was ten years ago. How should that fact bear upon the decision by the judge?
This offender has possibly spent the previous nine out of ten years out of the country.
Is that information available to the judge?
3. A person is about to be sentenced for driving without a valid licence. His past
record shows no convictions for that particular offence, but several quite recent
convictions for speeding. How should a judge compare the offences (if at all).
4. A person has been found guilty of driving while his licence was suspended.
His past record shows no convictions for this offence, but he has several previous
convictions for reckless driving, having been involved in several accidents. Is this
situation somehow similar to the one in example 3?
4
Intelligent Evaluation of Traffic Offender Records
5. A person has been found guilty of reckless driving. He has been found guilty
in causing an accident where the other driver was killed. His past record shows that he
has several convictions for having neglected to renew his licence and pay the yearly
car-tax . How should that fact influence the sentence in the present case? (if at all).
6. Combinations of the above examples occur of course, and complicate
matters even further.
4.1.2 The System Architecture
Various system architectures have been used in the past to build DSS in the sentencing
domain. In principle we distinguish five kinds of systems: (i) Statistical Systems, (ii)
Model-Based Systems, (iii) Case-Based Systems, (iv) Neural Network based system
and (v) Rule-Based Systems.
(i) Statistical Sentencing Systems in the general criminal domain have been
built in the past [5], [6], [7], but are not in use (except, possibly, for one).
(ii) Model-Based Systems have been proposed, but not implemented.
(iii) A Case-Based Sentencing System like the one described in [8] and [9] is
appropriate for a court of appeal. The time span of an appeal case is measured in
weeks and months (perhaps even years). A judge at this level has the time to apply a
case-based system, convince himself that the retrieved case or cases are indeed
relevant, and include the conclusions of the system in his deliberation.
However, our present system is intended for a judge at the lowest level of the
judiciary. He often hears several cases a day, he has practically no time for
deliberation, and he must hand down his decision the moment counsel and witnesses
have had their say. It is therefore clear that a case-based sentencing system would be
of no use. The judge simply does not have the time to apply it.
(iv) A neural network based system. Such a system lacks transparency in the
sense that the user cannot see clearly how a certain recommendation by the system is
derived. Nevertheless, in some legal applications there is a definite place for this kind
of system. [10]
(v) A rule-based system is the classical kind of expert system. It uses a
knowledge-representation in rule-form and applies logical deduction to the rules. Such
a system can be appropriate in our case if:
1. It operates very fast, so the user (judge) receives a qualified answer to a query
practically without any waiting time.
2. The output is concentrated and summarised for the user to survey in a moment.
As we shall show below there is no problem in fulfilling both of these
conditions. The rule-based paradigm is therefore the appropriate choice for our system.
The system is a rule-based system written in Prolog, with the interface (shown in the
Appendix) in Visual Basic.
5
U. Schild and R. Kannai
4.2 Deriving and Compiling the Domain (Expert) Knowledge
4.2.1 The Relevant Factors
Having decided on the architecture of the system, we approached the step of compiling
the domain knowledge. By this we mean the factors judges use to evaluate an
offender's previous record. This is of course where the intelligence is found. Two
questions came to mind before beginning interviews with the experts. The first
question was to which extent experts would agree among themselves about the factors.
The second question was to which extent the relevant factors were different for traffic
offences than for general criminal offences.
It appears that experts did not differ in their opinion of what these factors are
(or should be). This is both surprising and also a bit disappointing. As developers we
would have liked to cope with conflicting opinions.
The factors that judges considered relevant in the general criminal DSS were
as follows [1]:
1. Number of Previous Offences (Number of Adult Offences, Juvenile Offences)
2. Seriousness of Previous Sentences
3. Seriousness of Previous Offences
4. Similarity of Offences (Same type of offence, same law paragraph)
5. Frequency of Offences
6. New Offence Committed during Service of Previous Sentence
7. New Offence Committed during Cooling-off Period
The factors that traffic judges found relevant for traffic offences are as
follows:
1. Seriousness of previous offences
The offences are categorised as
(i) Serious offences:
Driving causing death, driving under influence of alcohol and/or drugs
Driving during period of disqualification (i.e. while licence is suspended)
(ii) Less serious offences (red light, speeding, etc.)
2. Similarity of previous offences
3. Seriousness of previous sentences:
Custodial, licence disqualification, deferred licence disqualification, fine, deferred
fine.
4. Driving causing accidents in the past:
Bodily damage, damage to property
5. Present offence committed during period of disqualification arising from a previous
traffic offence.
6. Present offence committed during period of deferred disqualification arising from a
previous traffic offence.
7. Frequency of offences
6
Intelligent Evaluation of Traffic Offender Records
4.2.2 The Analysis
The four classical approaches to punishment, Retribution, Deterrence, Prevention and
Rehabilitation form a classification of punishment commonly used by the judiciary
and by criminologists:
“We have thought it necessary not only to analyse the facts, but to apply to
those facts the classical principles of sentencing. Those classical principles are
summed up in four words: retribution, deterrence, prevention and rehabilitation. Any
Judge who comes to sentence ought always to have those four classical principles in
mind and to apply them to the facts of the case to see which of them has the greatest
importance in the case with which he is dealing” [Lawton L.J., in: Sargeant (1974) 60
Cr. App. Rep. 74 C.A. at pp.77-84].
We note that the traffic-factors from the previous section are quite similar to
the ones found for general criminal offences. This leads to the conclusion (confirmed
by our experts) that traffic judges apply the same approaches to traffic offenders.
However, we were somewhat surprised to find that one factor found relevant
for the general criminal DSS is not considered important: The total number of
offences. The reason could be that even a person with a great number of traffic
offences is not considered a professional criminal, neither by the public nor by the
judiciary.
In the first version of our prototype we simply gave ad hoc definitions of the
weight of the factors described above. However this is too simplistic a view of the
weighing of the factors against each other by a human.
There seems no particular reason to postulate complex interrelationships
among the factors resulting in a non-linear expression for the final result. However, the
computation of the individual weights had to been done in a more detailed and
intelligent manner, reflecting the views of the experts (judges). Thus, e.g., frequency
of offences is measured as a function of the type of offence.
The system analyses the record it obtains as input, determines the various
factors, and assigns them a weight according to the built-in rules derived from
interviewing the experts. Based on that computation the system issues a
recommendation to the judge of how to consider the previous record within the
framework of passing sentence in the case at hand. Figure 4 shows the intelligent
output of the system.
We have not been bothered by the fact that different experts assigned slightly
different weights to the factors. The contribution of the past record to the sentence in
the case at hand is never as great as the contribution of the offence at hand, so there
cannot be a great sensitivity in the choice of constants.
5. Conclusion
In the introduction we raised the question about the correlation between general
criminal offenders and traffic offenders. We have examined records of offenders who
committed both kinds of offences, and also searched the literature. A large number of
papers in the field of Criminology address this question, without reaching any definite
conclusions. It is therefore not surprising that we have not found any correlation.
7
U. Schild and R. Kannai
At this stage the system is undergoing testing by the experts under laboratory
conditions, not in the courtrooms. It is not clear to what extent the traffic judges in
Israel will actually use this system. We have in the past been involved in building DSS
for sentencing of various kinds. All were favourably received by the judiciary, legal
practitioners and the police. None of these systems are in actual use. This phenomenon
has also been observed by others [11]. This question will be the subject of our future
work.
6. Acknowledgements
We want to thank Advocate Hanan Mandel for his great help in carrying out this
project in a successful manner.
This work was graciously supported by The Ran Naor Foundation for the
advancement of road safety research (resh-gimmel 2006-009). We are deeply grateful
for this support, without which the research could not have been carried out.
7. References
1. Schild Uri J., Kannai Ruth, Intelligent Computer Evaluation of Offender’s Previous Record,
Proceedings of the 9th Int. Conference on AI and Law, ACM Press, 2003, 206-213.
2. Schild U.J., Kannai R, Computer Evaluation of Offender's Previous Record, Journal of
Artificial Intelligence and Law, 2005, vol. 13, 373-405
3. Shneiderman, B., Plaisant C., Designing the user interface: Strategies for effective human-
computer interaction (4th ed.) . Reading, MA: Addison-Wesley Publishing, 2005
4. Wellbank M., A Review of Knowledge Acquisition Techniques for Expert Systems. Ipswitch:
British Telecom, 1983.
5. Chan J. A Computerized Sentencing Information System for New South Wales Courts.
Computer Law and Practice,137-150 (1991).
6. Hutton N., Patterson A., Tata C. and Wilson J., Decision Support for Sentencing in a
Common Law Jurisdiction. Fifth International Conference on Artificial Intelligence and
Law (ICAIL-95), 1005, .pp. 89-95. Washington D.C.: ACM Press.
7. Hutton N., Tata C., Sentencing Reform by Self-Regulation: Present and Future Prospects of
the Sentencing Information System for Scotland's High Court Justiciary, Scottish Journal of
Criminology, 6, 2000, pp. 37-51.
8. Schild Uri J. (1998). Decisions Support for Criminal Sentencing, Artificial
Intelligence and Law, 6(4): pp. 151-202, Kluwer Publ
9. Hacohen-Kerner Y. and Schild U. J., The Judge's Apprentice, The New Review of Applied
Expert Systems, 1999, Vol. 5, pp.191-202.
10. Stranieri, A., Zeleznikow, J., Gawler, M. and Lewis, B., A Hybrid rule- neural approach for
the automation of legal reasoning in the discretionary domain of family law in Australia.
Artificial Intelligence and Law 7(2-3), 1999, 153-183.
11. Tata C., Resolute Ambivalence, Why Judiciaries Do Not Institutionalize Their Decision
Support Systems, International Review of Law, Computers & Technology, 14(3), 2000, pp
293-316.
8
Intelligent Evaluation of Traffic Offender Records
8. Figures
Fig. 0: Computer printout from Israeli Police computer of an offender's previous
record of traffic offences.
9
U. Schild and R. Kannai
Fig. 1: Short summary of previous record. Same colour-scheme as in Figure 2.
10
Intelligent Evaluation of Traffic Offender Records
Fig. 2: An offender's previous traffic record as it appears in our system (not on scale -
in actual system it appears as a full screen). The fields are coloured according to
different kind of traffic offences (red light, speeding, invalid licence, etc.).
11
U. Schild and R. Kannai
Fig. 3: Graph showing sentences over time. The upper graph shows sentences of
driving disqualification, and the lower graph shows fines (not on scale - in actual
system it appears as a full screen). The y-axis indicates months (for disqualification)
and sums in NIS (for fines).
12
Intelligent Evaluation of Traffic Offender Records
Fig. 4: List and pie-chart showing the relevant factors for weighing traffic offences,
summarizing the past record and computing a recommendation. Same color scheme as
in Figure 1 (not on scale - in actual system it appears as a full screen).
13