Computer Representation of the Law

ABSTRACT

For the purposes of this discussion there are three main types of law: defini tional, normative and case law; and there are a variety of computer representations of knowledge and problem-solving which can be applied to law. The social implica­ tions of representing the law by means of computer depend upon both the type of law and the technical means of computer implementation.

Definitional .I..e..H.. Much of human law is already written down in the form of fairly precise rul es, regula tions and def ini tions. There i s a separate legal tradi tion, represented by the work of Layman Allen in particular, which has long advocated the use of simplified forms of logic to represent law.

Recent work at Imperial College has confirmed many of Allen's expectations. We have succeeded in representing a significant portion of several British laws in Horn clause form and various of its extensions. Among other applications. we have been able to run a significant portion of the 1981 British Nationality Act, implemented in PROLOG on a small micro computer. The first subsection of the Act, for example, can be expressed in the conclusion-conditions form of a single Horn clause:

For every individual x, date y, individual z and section of the Act w,

x acquires Bri tish ci tizenship by section 1.1.a on date y if x is born in UK on date y and y is after the Act takes effect and x has a parent z and z is a British citizen by section w on date y.

Not all of the Act can be easily represented in such a simple form. We have expressions such as:

"x is a citizen if

and bis mother is a citizen or would have been a citizen had she been male."

This requires a representation of metalevel reasoning about states of knowledge in addi tion to object level reasoning about the world. Investiga­ tion of such problems of representing legislation and legal reasoning has bad a salutory effect on

COMPUTER REPRESENTATION ..QE .TilE ..LMi

Robert Kowalski, Marek Sergot

Department of Computing, Imperial College, London

our more general study of knowledge representation and pro bl em- sol ving.

Other appl ica tions of computa tional logic to the representation of definitional law at Imperial College include immigration law, social securi ty law, grants to industry and a company's pension regulations.

Our studies suggest that such applications of computational logic are ripe for practical exploi ta tion. Computa tional Logic can be used not only to assist the application of rules and regulations, but also to aid the process of determining the logical consequences of legislation before it is enacted. It can be used not only for rules which have legal authority but also for rules which are used by organizations to regulate their own internal affairs. Potential applications include the formalization and mechanization of tax law, company law, airline regulations and universi ty examination regulations.

The technical opportunities are significant and may result in the development of systems which have great social impact. How can we be sure that the computer representation of laws is accurate; and therefore that the conclusions they imply are correct?

On the other band. our experience suggests that precise formal representation of the rules actually clarifies and often simplifies them. This same point was made by Layman Allen as long ago as 1957. In the long term we may be so attracted by the benefits of formalization that the real question becomes: How can we be sure that the natural language sta tement of the law does justice to i ts legally binding formalization?

Normative ..l&.H.. Many law s are concerned w i th permission, prohibition and obligation. Tradi tionally this is an area where modal deontic logic has been thought to be necessary. Typical examples are parking regulations, criminal law and

the ten commandments.

We have only begun to scratch the surface in our investigation of such lawL However, our dissatisfaction with modal logics has forced us to consider how norms might be handled within the framework of classical computational logic.

It can be argued that to say that an action is prohibited is to say that some punishment or other sanction can be expected if the action is performed. Thus something like Prohibited(a)

in a modal logic

becomes

there exists p such that Expected-result(a p) and Undesirable(p) in a classical logic.

Whereas classical logic requires us to refer to sanctions explicitly, modal logic wraps them up within modal connectives. Classical logic allows and even encourages us to identify sanctions explicitly, e.g.:

"Parking in a no parking zone carries a fine of f5."

But it also allows us to existentially quantify sanctions if we don't know or want to say what they are, e.g.:

"Cheating on examinations is prohibited"

means there exists a punishment for cheating on exams.

We believe that such a treatment of deontic concepts is technically attractive. It increases expressive power and allows existing proof procedures for classical logic to be used for deriving logical consequences from assumptions. These arguments are the same as ones we have made elsewhere for treating notions of time and events explici tly wi thin classical logic rather than

impll citly within modal temporal logic.

Such a technical solution to the problems of representing and reasoning with norms may further our ability to implement practical systems. But encouraging explicit reference to sanctions may have undesirable social implications, whether or not computer implementation is invol ved. It may encourage us to weight up the relative costs and benefits of breaking the law. A parking fine is virtually no punishment for a wealthy businessma~ A person who is terminally ill or about to commit suicide has no punishment to fear, unless he believes in an after life.

~ .l&H. A significant proportion of law in English speaking countries is based upon previous cases which create a precedent for the futur~ Even where precedents have no legal authority, previous cases must be considered for the sake of consistency and fairness. The need to reason by cases is also buil t into the lower levels of much definitional and normative law: What does it mean, for example,

"to exercise reasonable effort", "not to have residence in the U.K.

as one's primary intention of marriage"?

Such notions can only be determined by developing relevant criteria and applying them flexibly over a period of time.

Attempts to formalize aase law in Artificial Intelligence have been attracted to its similarity with reasoning by analogy and with reasoning by

means of frames.

However there is an alternative - which is to generate general rules by induction from previous cases and apply them to new cases in the future. Rules so generated do not, of course, have 'the same legal authority as rules which are explicitly written down. They may be too general or too narrow, They may conflict with other rules generated from other cases. Belief revision is needed to discriminate between conflicting rules arising from different precedents.

A certain amount of support for such a treatment of precedent can be obtained by referring to the arguments which are given when a decision is made in a particular case. If the case is to be a precedent for the future then the justification for the decision is usually made in general terms, In other words, the justification itself contains the main ingredients of a general rule.

We would argue that treating case law by generating tentative general laws is not only technically preferable to reasoning by analogy and by frames but it is also socially more desirable. It has the consequence that general rules can be cited to justify individual decisions, Since we know what the rules are, they can be scrutinized and changed. Transformations which reason by analogy or frames are inscrutable and therefore beyond social criticis~

l&&äl~ Systems. The state of technology

is rapidly reaching a stage where the computer representdtion and processing of law will make legal expertise available to a much broader spectrum of the communi ty.

In general this is to be welcomed. It will give more people better legal and financial advice. It will help more people to understand the law, so

that they can make better, more informed decisions.

But there are dangers too. There are the obvious dangers: Who is responsible if the computer makes a mistake? Less obvious perhaps, and more insidious therefore: Will we bother to listen to the computer's explanation? Will we bother to understand the issues involved? Or will be come to

rely on the machine to do our thinking for us?