Research in artificial intelligence and law goes back approximately 40 years. It remains largely based on formal logic, including non-monotonic logic, case-based reasoning, and logic programming. However, some researchers in and practitioners of law have argued in favor of quantitative approaches (e.g. probability) to account for uncertainties in legal arguments. Other researchers have pointed some of the shortcomings of the current artificial intelligence and law research, e.g. inability to take context into account. At the same time, machine learning has made huge inroads in many different fields and applications, and therefore, the question is whether machine learning has anything to offer to the theory, and, equally important, the practice of law. As a position paper, this is a preliminary study towards the exploration of a synergistic integration of current artificial intelligence approaches in law, with machine learning approaches. It puts forward the idea that formal, logic-based approaches, currently very popular the Artificial Intelligence & Law research, could benefit from an extension with a machine learning component, and discusses some ways in which machine learning could be integrated into these approaches.
When it comes to machine learning and law, there are two, quite unrelated, directions of study. On one hand, researchers are interested in legal issues raised by the research in machine learnig. For example, the symposium Machine Learning and the Law, held in conjunction with NIPS-20161, had as goal to ”explore the key themes of privacy, liability, transparency and fairness specifically as they relate to the legal treatment and regulation of algorithms and data. On the other hand, the second direction is that of actually use of machine learning in law research and practice.
Stimulated on one hand, by progresses, as well as by shortcomings of artificial intelligence approaches in law (as perceived by various researchers), and on the other hand, by the tremendous recent machine learning succeses in many different directions (not including law), this paper suggests that there is scope for using machine learning in law, and 1Annual meeting of the Neural Information Processing Society moreover, that a formal treatment may be used towards this end.
The paper is inspired by work on logic-based
formalization of legal reasoning,
As in many areas of research, ranging from science,
engineering, medicine, and social sciences including the
legal field, artificial intelligence has brought about
possibilities, which excited some, intrigued others. Pioneering work
done by Edwina Rissland and her students and
collaborators,
Formal logic is the approach of choice for artificial
intelligence and Law, as evidenced by a wealth of articles,
including those already mentioned, and others,
Moreover, analyzing the current results, Franklin
2According to Google’s Rob Craft We are currently at year zero
of the machine learning revolution.
7. ”Probabilistic (or default or non-monotonic) reasoning (including problems of priors, the weight of evidence and reference classes)” 8. ”Issues of the discrete versus the continuous”
Franklin discusses the use of fuzzy set based approaches,
to capture the nature of some concepts, or the similarity of
a case to precedents. As an example, he considers, the
concept ’vehicle’, in the ordinance ”No vehicles are allowed in
the park”, which obviously would refer first and foremost
to cars, less to motorcycles/bicycles, and even less to roller
skates. A fuzzy set of vehicles, defined by a membership
function µvehicle : U ! [0, 1], where U denotes a universe
of discourse of ’things’, would assign different degrees to
cars, motorcycles, bicycles, roller skates, for example,
respectively
µvehicle(v) =
8
><
:>
1
0.8
0.5
0.1
if v is a car
if v is a motorcycle
if v is a bicycle
if v is ”roller skates”
The ordinance has as goal prevention of accidents in the
park, and by consequence, the definition of the fuzzy set is
meant to reflect the common sense knowledge that cars can
cause serious accidents, motorcycles less, bicycles even less,
and so on. The actual assignment of membership degrees is
seen in
The issue of similarity is of utmost importance in legal
reasoning and to illustrate the difficulties in similarity
evaluation
In
e promotes M prod where ⇡Liv =) ⇧ means ”⇡ was pursuing his livelihood is a reason why ⇡ should have a legal remedy against ”.
To formalize, following
Vi < Vj ; i 6= j Vi < Vi [ [ Vj j6=i The plaintiff caught the ball in the upper portion of his glove but was tackled and thrown to the ground by the crowd. The ball fell out and the defendant picked it up and put it in his pocket. The plaintiff sued for conversion. Holding: The plaintiff and defendant had equitable claims and could not prove their case either way. Reasoning: Although the plaintiff proved intent to possess the ball, he could not establish that he would have fully possessed the ball had he not been tackled by the crowd. If he could have established this, his pre-possessory interest would have constituted a qualified right to possession which can support a cause of action for conversion. Judgment: The ball was sold for $450,000 and the proceeds were divided equally. (1) (2)
Replacing [ in (2) by the maximum _ , and using ^ for
minimum, it follows that
Equality of values must also be specified as part of the
theory. Then, enlarging upon
⇣ Vi > min Vi, Vj6=i Vj ⌘ ⌘ d↵ 1 =) d↵ 2 =) . . . d↵ n =) e promotes V1 e promotes V2 e promotes Vn d↵ 1&↵ 2& . . . &↵ n =)
e promotes (3) (4) one can construct the following:
[min(V1, V2, . . . , Vn), max(V1, V2, . . . , Vn)]
Thus, considered together, rules (4) promote at least the smallest value, at most the largest value, and possibly values in between, i.e., those which lie in [min(V1, V2, . . . , Vn), max(V1, V2, . . . , Vn)]. All promoted values can be expressed as convex combinations of V1, V2, . . . , Vn, that is, d↵ 1& . . . &↵ n =)
e promotes w1V1 + · · · + wnVn (5) where wi 0, i = 1, . . . , n and w1 + · · · + wn = 1. For different values of wi, i = 1, · · · , n (5) can generate any subset of the set of values {Vi, i = 1, · · · , n}. Interpreted as a probability, wi = P rob( to promote Vi) can be obtained through a machine learning algorithm based on history of (similar) cases.
The mechanism outlined above has the effect of producing a continuum of legal values (even though to begin with, these form a discrete set), which in turn may lead to a continuum of possible decisions.
This section touches upon the issue of inference in legal
reasoning. It takes its cue from
Since complex arguments about inferences from evidence
rest on almost innumerable subjective judgments,
It can be claimed that achieving these purposes predict - replace - mimic - support falls into the machine learning realm, requiring machine learning algorithms of possibly different levels of sophistication.
From the point of view of a typical approach to machine
learning, data (usually, a lot) is needed to construct a
machine learning algorithm - a classifier, or a clustering
algorithm. Usually, such data is thought of as history on which
to base future predictions. The need to take into account
history is discussed in the conclusion section of
Using machine learning to analyze judges’ personalities and ruling tendencies helps tailor pleadings to their personalities. Machine learning helps analyze attorney personalities and use those to decide who writes what in a law firm, and evaluate a firm’s previous work and identify strength, weaknesses and faults.
This added dimension to legal theory and practice straddles several disciplines, including psychometry, representation of uncertainty (e.g., fuzzy logic to represent meanings of utterances, and similarity measures), probabilistic (point, interval valued or imprecise probabilities), all to be used in machine learning to build predictive algorithms of behavior.
As a recent example, with far reaching consequences, of behavior prediction, comes from the 2016 USA presidential elections: Cambridge Analytica5 used machine learning to specifically target independents and other voters disenchanted with the status quo, with messages that appealed to their personalities. A similar system that does the same - for judges, courts - could help build a litigation strategy, tailor language, and develop legal reasoning to the personality of the particular court.
5https://cambridgeanalytica.org/
We have discussed some preliminary ideas on the challenges/issues that law research faces, which could be approached from an machine learning point of view. This paper only hinted at these issues and possible solutions using machine learning. Much is to be done, including a very thorough understanding of quantitiative ideas in legal theory put forward by researchers in the legal profession.
The authors are grateful for the reviewers comments, some very enthusiastic, some very negative, on the first draft of this paper. All served to stimulate the authors’ thinking on how to further approach the use of machine learning in law, and are likely to inform their future study of this field.