Abstract: Computational Law is that branch of legal informatics concerned with the mechanization of legal reasoning. While the idea of legal computation is not new, its prospects are better than ever due to a convergence of technological trends - including the growth of the Internet, the proliferation of embedded computer systems, and progress in knowledge representation and automated reasoning. In this paper, we examine the concept of Computational Law, we summarize its prospects and problems, and we examine its philosophical and legal implications.

Introduction

We live in a complex regulatory environment. As citizens, we are subject to governmental regulations from multiple jurisdictions - international, federal, state, and local. As members of organizations, we are subject to organizational policies and rules. As social beings, we are bound by contracts we make with others. As individuals, we are bound by personal rules of conduct.

The sheer number and size of regulations can be daunting. We may all agree on a few general principles; but, at the same time, we may disagree on how those principles apply in specific settings. In order to minimize ambiguities, regulators are often forced to create numerous regulations, to deal with special cases; or, viewed alternatively, they are forced to create very large regulations.

A recent article in the National Review made this case forcefully. "The Lord's Prayer is 66 words, the Gettysburg Address is 286 words, there are 1,322 words in the Declaration of Independence, but government regulations on the sale of cabbage total 26,911 words."

Complicating the situation is the complexity of these regulations. Even small regulations can be very complex. While this complexity can sometimes be mitigated by careful drafting, such care is not always possible due to time constraints; moreover, once regulations are created, complexity often increases as the regulations are changed and then changed again.

A simple example of the problem of complexity is the Michigan Lease Termination Clause shown below. This case was first highlighted in a paper [Sergot et al.] written to illustrate this very point.

"The University may terminate this lease when the Lessee, having made application and executed this lease in advance of enrollment, is not eligible to enroll or fails to enroll in the University or leaves the University at any time prior to the expiration of this lease, or for violation of any provisions of this lease, or for violation of any University regulation relative to resident Halls, or for health reasons, by providing the student with written notice of this termination 30 days prior to the effective date of termination; unless life, limb, or property would be jeopardized, the Lessee engages in the sales of purchase of controlled substances in violation of federal, state or local law, or the Lessee is no longer enrolled as a student, or the Lessee engages in the use or possession of firearms, explosives, inflammable liquids, fireworks, or other dangerous weapons within the building, or turns in a false alarm, in which cases a maximum of 24 hours notice would be sufficient."

The rule itself is actually fairly simple. However, there are many conditions; there are conditions that modify other conditions; and so forth. The upshot is a regulation that is difficult for most people to understand without a substantial amount of study.

A third problem with our regulatory environment is the fact that regulations are not always well coordinated, arising, as they do, in different settings for different purposes. Sometimes, there are gaps, leaving important cases uncovered. More often, regulations overlap other regulations and in some instances are inconsistent with each other.

As and example of the latter problem, consider the issue of medical marijuana. In 1996, "56 percent of California voters approved Proposition 215, allowing sick people to use marijuana for medical purposes when approved by a physician. In all, 35 states have approved similar legislation. The only problem is that the federal government still outlaws the use of marijuana, for any reason, which has created enormous legal headaches for sick individuals, doctors and law enforcement." [Ventura County Star]

In this case, there is a seemingly simple resolution - under United States law, federal regulations always trump state and local regulations. In other words, the state regulations are invalid. Nevertheless, such inconsistencies lead to confusion and violations and needless costs.

Besides, not all cases can be resolved so simply. While states and municipalities are not permitted to create regulations that conflict with those of the federal government, they can still enact regulations that conflict with those of other states and municipalities. Such conflicts cause no problems so long as they are applied solely within their home jurisdictions. However, problems can arise when individuals and organizations act across jurisdictional boundaries, such as in transportation, shipping, and so forth.

Problems like these make it difficult for affected individuals to find and comply with applicable regulations; they make it difficult for monitors to assess compliance; and they make it difficult for regulatory bodies to evaluate proposed regulations and changes to regulations. The result is frequent lack of compliance, inefficiency, and widespread disenchantment with the regulatory system.

Fortunately, the problems described above are not insurmountable. To the extent that these are information problems, we believe that they can be mitigated by information technology.

One step in this direction has already been taken. Today, the text of many legal documents (including cases, statutes, analysis) is available online. In some cases, the information is adorned with "semantic" tags / keywords to help in search. The good news is that these documents can be found using general search services, such as Google, or using services that specialize in legal information, e.g. those provided by companies like Westlaw and LexisNexis. Unfortunately, the quality of such search is limited; they often return too many documents and sometimes fail to find relevant information. Moreover, there is no automation; a specialist must still be there to read the documents and apply them to individual cases.

In this paper, we focus on the next step in legal informatics, in which legal information is encoded in a form that supports not just search but legal automation. The result is an extreme form of legal informatics known as Computational Law.

Computational Law

Computational Law is that branch of legal informatics concerned with the mechanization of legal reasoning. The practical goal of work in the field is the implementation and deployment of computer systems capable of doing useful legal calculations, such as compliance checking, legal planning, and so forth.

Intuit's Turbotax is an example of a rudimentary computational law system. Based on values supplied by its user, it automatically computes the user's tax obligations and fills in the appropriate tax forms. If asked, it can supply explanations for its results in the form of references to the relevant portions of the tax code.

Our position is that systems like Turbotax can be implemented for many areas other than tax preparation - in dealing with privacy and security matters, in intellectual property rights management, in enterprise management (e.g. constraints on travel, expenditure, reporting), in assessing compliance of plans with building codes (affected by local, county, state, and federal safety requirements), in electronic commerce (e.g. import/export restrictions on technology, drugs, and so forth), in labor law (e.g. occupational safety regulations and health care benefits, notably cases where state regulations interact with federal provisions), and so forth.

More interestingly, we believe that it is possible to build systems capable of legal reasoning in general. Like Turbotax, such systems would accept facts as inputs and produce legal conclusions as outputs. However, unlike Turbotax, where the tax laws are "built-in", they would also accept encodings of laws as inputs and would be able to use these encodings in drawing their conclusions.

One advantage of separating representation and reasoning in this way is that a single general legal reasoning system can be used multiple times, for different jurisdictions and for different combinations of jurisdictions.

The dual of this is also true. Once a set of regulations is encoded formally, it can be supplied as input to different legal reasoning engines for different purposes, e.g. to check compliance, to plan for compliance, to detect inconsistencies or redundancies, and so forth.

Computational Logic

Our approach to building legal reasoning systems of the sort described above is based on Computational Logic. There are three components to this approach - (1) the encoding of facts in the form of relational data, (2) the codification of regulations as sentences in formal logic, and (3) the use of mechanical reasoning techniques to derive consequences of laws and facts so represented. In what follows, we take a look at how this works.

Before a system can compute consequences in specific situations it must have the facts necessary to derive legal conclusions. In the database world, it is common to encode facts as rows of tables. While the tabular representation is the most popular and arguably most intuitive way to think about database relations, for some purposes a sentential representation is useful. In this representation, we write each fact as a sentence in a notation similar to that used in mathematics, and we define a database as a set of such sentences.

As an example, consider how we might encode some basic information about a small, hypothetical enterprise. John manages Kat and Ken, and Jill manages Mary and Mike. John and Ken are in room 22 and Jill and Mike are in room 24. John, Ken, and Mike are male; Jill, Kat, and Mary are female. We can express this data in mathematical form as shown below.

The language of formal logic extends this language for writing facts in two ways. First of all, there are variables, which allow us to refer to arbitrary entities in the domain of a database. Secondly, there are logical operators, which allow us to express relationships between facts. In what follows, we use words that begin with capital letters as variables, e.g. X, Y, Z, Age, R14. Words that begin with lower case letters or digits are assumed to refer to specific entities, e.g. john, jill, kat, ken. Our logical operators include & (and), | (or), ~ (not), and :- (is defined as or is implied by).

The most basic use of these representational extensions is to define new relations in terms of existing relations. The sentence below is an example. Here, we define the officemate relation in terms of the office relation. A person X is an officemate of a different person Y if both X and Y are assigned to office Z.

We can encode rules and regulations in similar fashion by writing rules that define the concept of illegality. As an example, consider how we might express the organizational regulation that no manager may have a direct subordinate as officemate. The Computational Logic sentence shown below expresses this fact using the vocabulary used above. It is illegal if there is a person X who manages person Y and is the officemate of Y.

Evaluating legal compliance with facts and logical sentences encoded in this way is quite simple. In order to prove that a conclusion holds, we find a rule in which the head matches the desired conclusion and then try to prove the conditions.

As an example of this sort of reasoning, consider the task of proving that there is an illegality in the data shown earlier, given the rule shown above. In this case, we begin with the "goal" of proving that something is illegal.

Using the definition of illegal shown above, we can reduce this goal to the problem of showing that there is is an X who manages Y and is the officemate of Y.

Using the definition of officemate, we can reduce this subgoal to problem of finding an X and a Y such that X manages Y, X is in office Z, and Y is in office Z as well.

All of the conjuncts in this reduced goal involve relations used in our dataset; so we can evaluate its truth by matching against the data shown above. By "binding" variables to specific constants in the database, we see, that it is possible to make all of these conditions true. Let X be john, Y be Ken, and Z be office 22.

Note that legal planning and the analysis of regulations are more difficult than compliance checking since multiple hypothetical possibilities must be considered. However, both can be automated using well-known extensions to the compliance checking technique described here.

Technical Challenges

While the language introduced in the last section is sufficient to express many types of regulations, there are others sorts of regulations that are more complicated. In his seminal article on the use of logic in representing law, Bob Kowalski identified a number of different shortcomings of this approach.

The examples in Kowalski's paper are all centered on the British Nationality act. He found that he could represent certain aspects of the act with ease. One of his positive examples is the following clause.

Using the language of the last section, Kowalski was able to render this definition in terms of properties such as a person's birthplace, birthdate, the time at which the Act commences, a person's parents and where they live. Hs definition is shown below.

Unfortunately, in examining the Act, Kowalski also came upon a number of conditions that were not readily representable in this form. Some clauses depend on "default conclusions" (e.g. "... unless the contrary is shown ..."). Some clauses require the representation of metalevel information, i.e. references within the law to other parts of the law (e.g. "for the purposes of subsection (1) ..."). Some clauses depend on people's beliefs about the facts (e.g. "... the Secretary of State is satisfied that ...). Some clauses use words that are not fully defined in the Act. For example, one clause uses the phrase "new-born infant" without further qualification. So what qualifies as a "new-born infant". One day ld, two days? What exactly is an "infant" and when is it no longer "new-born"?

The good news is that some of these problems have been addressed in the years since that article was written and can be handled by well-understood extensions to the techniques shown above. However, unlike the techniques described in the last section, these new techniques have not yet secured universal agreement. Moreover, we still have no way of dealing with the Open Texture problem.

A different sort of challenge to Computational Law stems from the fact that not all legal reasoning is deductive. Edwina Rissland notes that, "Law is not a matter of simply applying rules to facts via modus ponens" [Rissland], and when regarding the broad application of AI techniques to law, this is certainly true. The rules that apply to a real-world situation, as well as even the facts themselves, may be open to interpretation, and many legal decisions are made through case-based reasoning, bypassing explicit reasoning about laws and statutes. The general problem of open texture when interpreting rules, along with the parallel problem of running out of rules to apply when resolving terms, presents significant obstacles to implementable automated rule-based reasoning. Also, in many legal domains, the facts of a situation themselves may be unclear or incomplete: human intervention and interpretation is necessary to make these facts available to a legal reasoning system so that it can even apply the rules. This further adversely affects usability and any notion of correctness. To combat these shortcomings, some rule-based systems have been hybridized with case-based systems, or augmented with meta-rules or with nonmonotonic, defeasible reasoning techniques, in order to make them more suitable for general applications in law.

Philosophical Challenges

In addition to these technical challenges, there is an important philosophical challenge. When creating laws, law-makers can choose to formulate the same law in many different forms. We can think of two broad categories of these form of laws - one type that have come to be known as "rules" and another that we can call "standards". To illustrate, let's take the example of lawmakers writing a hypothetical speeding law. They might have multiple ways of going about drafting this law. They consider the "standard" version - It is illegal to drive safely, or the "Rule" version - It is illegal to drive more than 65 miles per hour. It may seem that these two versions of the same law are basically equivalent - a half-dozen of one, 6 of another. They both regulate the behavior of speeding. But it turns out that the particular form that is chosen can have a dramatic impact on how the law operates in practice. For our purposes - the mechanization of legal reasoning - it turns out that laws formulated as rules are easier to apply than standards.

Note that the words "rules" and "standards" have their own specific meaning here - it's unfortunate terminology but it has been widely adopted -- so try not to confuse them with previous ideas about these widely used terms.

Let us examine these two broad categories of legal forms in some more depth. Standards are characterized by having open-ended, abstract categories. For example, the term "unsafely". Rules, on the other hand, are characterized by having what are called "bright line" criteria that are easily compared to readily discernable data in the real world. So, an example of a bright line criteria is the 65 mile per hour speed limit. It's a bright line because 65.1 miles per hour is too fast, but 64.9 miles per hour is fine. Secondly, we can readily determine speed - a fact about the world - and compare it with the criteria. Contrast how easy this is with the task of determining whether a defendant is legally insane for the purposes of the insanity defense in criminal law.

The virtue of standards is that they endow legal authorities with discretion and are flexible, whereas rules are usually inflexible and only a proxy for what is really wanted. Lawmakers really want to prevent unsafe driving, but unfortunately they have to settle for a limit on speed. Because of their rigidity, rules can be over-inclusive and/or under-inclusive - it's a fundamental tradeoff. They prohibit some desirable conduct, such as safe driving over 65 miles per hour, and allow undesirable conduct - such as unsafe driving below 65 mph. The major benefits of rules are that they are inexpensive to administer, and they are often seen as more objective, and they are more determinate - for predicting compliance.

Practical Challenges

Finally, there are some pragmatic challenges. We must look at the laws that govern the use of computer systems that derive legal conclusions. There could be issues of unauthorized practice of law, and there are issues of liability for erroneous legal conclusions.

The first question is whether a computer system is permitted to draw legal conclusions. Is this considered the unauthorized practice of law - the same way in which a non-attorney cannot give legal advice? In one early case in Texas in 1999, the answer was that software drawing legal conclusions constituted unauthorized practice of law. However, the Texas legislature quickly stepped in and changed the law to permit these types of computer programs.

Since then, with the widespread use of programs like Turbotax, things seem to be looking up for legal software. The cases seem to distinguish between software written or used by non-lawyers, versus software written or used by lawyers. Both non-lawyers and lawyers can write and dispense advice based upon very mechanical legal analysis - the filling in of legal forms, addition, or wizards that just ask you questions. However, for advanced legal analysis or advice, it appears that non-lawyers can neither write these programs nor dispense advice based upon the results of these programs.

Next, the question is what is one's liability for relying on the legal advice of a computer system? This question hasn’t been answered fully, but we can look to other areas that heavily rely on computer systems, such as structural engineering or medicine. Much of the answer is context dependent – it depends upon how reliant the industry is on the system, how accurate the systems are, whether they are more or less accurate than humans. The standards for relying on computerized legal advice are likely to change as these systems become more widespread.

Finally, we might wonder about the liability of makers of computer systems for erroneous legal conclusions. Again, the answer is likely to be rooted in negligence law - whether or not the makers act with the same level of care as human professionals. Much of this is currently covered by disclaimers - it’s unclear how effective these will be in court.

Opportunities

Even without complete answers to these challenges, we believe that there is value in investigating Computational Law. There are domains in which explicit rules govern behavior. Just because we cannot treat all laws in this way does not mean we cannot treat any laws in this way.

In order to promote research in this area, the Stanford Law School and the Computer Science Department recently established a research center - The Stanford Center for Computers and Law, aka CodeX. CodeX is a multidisciplinary laboratory operated by Stanford University in association with affiliated organizations from industry, academia, and government. The staff of the center includes a core of full-time employees, together with faculty and students from Stanford and professionals from affiliated organizations.

The primary mission of the Center is to explore ways in which information technology can be used to enhance the quality and efficiency of our legal system while decreasing its cost. The explicit goal of the center is "legal technology" that empowers all parties in our legal system, not just the legal profession per se. Such technology should help affected individuals find, understand, and comply with regulations; it should help enforcement organizations monitor and/or enforce compliance; and it should help regulatory bodies analyze proposed regulations for cost, overlap, inconsistency, etc.

In furthering this mission, CodeX is exploring a variety of general legal technologies, including the following.

* Orgnet is a web repository for data about organizations, public companies and private companies, profits and non-profits, givernmental organizations, and so forth.

* Peoplenet is a web repository for information about public figures, e.g. office holders in the government and officers of publicly traded corporations.

* RegNet is a web repository for regulations in machine-processable form. A public resource usable both by human users via an appropriate user interface and usable by programs as a server (analogous to a domain name server). Our initial focus is on rules of governmental procedure, e.g. federal rules of civil procedure, permitting and licensing procedures, and so forth.

* Legal Spreadsheets and legal web forms are applications of general technologies, like logical spreadsheets and smart forms, in which the logical constraints include regulations and contracts.

CodeX is also developing specific applications, such as the following.

* Calc. Project CALC aims to explore the application of computational law within the field of building design and construction. CALC will examine the degree to which existing laws governing the domain of building design can be modeled within computer systems and made to interact with systems currently used in the field. Existing building construction projects are covered by numerous laws and regulations, including local building codes, federal environmental rules, and accessibility laws such as the Americans With Disabilities Act. The project will examine whether computer systems can assist design professionals in knowing and complying with these rules. CALC will also explore legal theoretical problems related to the representation of laws in computer systems, and propose principles for selecting and creating such laws. CALC is a interdisciplinary effort involving researchers and research from the fields of law, computer science, and civil engineering.

* Stanford Intellectual Property Exchange (SIPX). An online intellectual property exchange, with robust commercial and non-commercial functionalities, which is equally accessible to individual content creators, large media companies, consumers, and others. The system, once deployed, will reduce legal transaction costs for intellectual property exchanges. It will obviate, or eliminate the need for live legal consultation for platform-based transactions. IPX is a literal "marketplace of ideas," and their myriad instantiations.

Conclusion

Computational Law is admittedly an ambitious approach to legal informatics. Although there are significant challenges to using the technology in all areas of the law, there are many areas where the technology is suitable even in its current form. Moreover, as the technology comes to be used, we believe that regulators may write find advantage in writing more rules and fewer standards, thus enlarging the range of applicability of the technology.

Given these opportunities, we believe it is worthwhile to pursue research, development, and deployment of computational logic systems. The ultimate goal is "legal technology" to benefit all parties in our legal system, not just the legal profession per se. Such technology should help affected individuals find, understand, and comply with regulations; it should help enforcement organizations monitor and/or enforce compliance; and it should help regulatory bodies analyze proposed regulations for cost, overlap, inconsistency, etc. Like the motto of CodeX, the result would be "legal empowerment through information technology".

In a sense, such progress is essential to the proper functioning of the law as a mechanism for achieving social good. One of the functions of the law is to help individuals predict the consequences of their actions. If we do not know what the law is, the law does not serve this function.

The Constitution of the United States, in both the fifth and the fourteenth amendments, mandates "due process" for its citizens. Part of due process is the concept of notice. Citizens must receive notice of applicable regulations before they can be charged with violations. Some legal scholars have argued that, when the law becomes so recondite that citizens are unable to understand it, then they have not received adequate notice and cannot be charged.

In a sense, Computational Law is the natural next step in a progression that began millenia ago. Around 1750 BC, Hammurabi had the laws of the land encoded in written form (literally cast in stone) so that citizens could know what was expected of them and what would happen if they violated those expectations. Since then, it has been the norm to encode rules in written form. However, with the proliferation of rules and regulations, just writing things down is not enough. In a way, Computational Law is the first bit of revolutionary progress in this regard since the days of Hammurabi. By making the law intelligible in the context of real situations, it helps to mitigate the growing complexity of the law and helps it to achieve its social purpose.

References

Buchanan, Bruce and Thomas E. Headrick. Some Speculation About Artificial Intelligence and Legal Reasoning. 23 Stanford Law Review 40 - 62 (1970).

Gardner, Anne: An Artificial Intelligence Approach to Legal Reasoning, M.I.T. Press, 1987.

McCarty, L.Thorne., Reflections on Taxman: An Experiment in Artificial Intelligence and Legal Reasoning, 90 Harvard Law Review 837 (1977).

Mehl, L.: Automation in the Legal World, Conference on Mechanisation of Thought Processes, Teddington, 1958.

Meldman, J. A. A structural model for computer-aided legal analysis. Rutgers Journal of Computer Law, 6, 1977.

Rissland, E. L., Ashley, K. D., and Loui, R. P.. AI and law: a fruitful synergy. Artif. Intel l., 150(1-2):1-15, 2003.

Sergot, M. J., F. Sadri, R.A. Kowalski, F. Kriwaczek, P. Hammond and H.T. Cory, The British Nationality Act as a logic program. In: Communications of the ACM, vol. 29, no. 5, pp. 370-386, May 1986.