Artificial intelligence and law? A short introduction without lies or exaggerations

1. Introduction

We have surely come across someone who has pointed out that artificial intelligence (AI) is going to replace lawyers by putting some examples of systems in development or simply because they heard it. These days, the sale of smoke regarding innovation and AI is growing overwhelmingly and something needs to be done against it[1].

Before coming to believe statements like the ones above, it is necessary to clarify some previous questions, such as what is the relationship of AI with the Law and if it actually has it. Little has been said about this, but it is essential to do so to get into the pool.

In this work we will develop i) What is AI ?; ii) How is it related to the law ?; and finally, (iii) if they will replace us. Which when the reader reaches the final lines, can be answered if indeed an AI system will replace the work of the lawyer.

2. What the hell is AI?

Rich and Knight (1994) in their classic book on AI, stated, without going into theoretical depths, that AI studies “how to get machines to perform tasks that, for the moment, are better performed by human beings” ( p. 3). And indeed, AI is a branch derived from computational and cognitive sciences that studies and executes simulations of actions classified as "intelligent". The standard to identify which behaviors are intelligent and which are not, are the behaviors coming from subjects with natural intelligence: humans and animals, for example. For this reason, part of artificial intelligence focuses on emulating these actions through algorithms (rules) programmed through computer language.

Now, according to the level of emulation of human behaviors, AI can be classified as weak or strong. We speak of weak AI when it only emulates certain human actions such as playing chess, recognizing facial data or predicting certain situations with previous data. While we speak of strong AI when it emulates the total activity of the human being and not just a piece of its activity.

Today there are successful systems that emulate "pieces" of human activity and not a strong AI, or "general" as they are also called, that can emulate all the joint activities of the human being.

Following the scheme of Mill (2016), within what is known as AI the following research areas can be included:

  • Machine learning (machine learning)
  • Natural language processing
  • The design of expert systems
  • Artificial vision
  • Automatic planning
  • The robotic

And in fact, each of these areas interacts with the other in order to generate systems that emulate human activity. In the case of Law, which is what we care about here, there are areas of AI that should interest us more than others due to utility criteria. For example, in law you need more natural language processing than machine vision or robotics.

Having clarified this, let's see briefly what areas and what applications can be extracted from this dialog.

3. How are artificial intelligence and the law related?

3.1. The foundation: natural language processing

"The law is a linguistic phenomenon" (Guastini, 2001, p. 7). The Constitution, the codes and other laws are expressed in natural language. We seek information by reading and understanding what the laws say. We find syntactic and semantic similarities, etc.

In daily life, the legal operator is dedicated to looking for information applicable to the specific case at hand. For that they go to the latest jurisprudence issued by the Supreme Court, the Constitutional Court or the superior courts of each judicial district. We also read some books that can clarify the concept of an institution to better frame our syllogisms. After having found all the pertinent information for our case, we began to write and form arguments based and anchored in all that information collected, generating argumentative blocks.

Well, all these seemingly easy activities, we do by processing natural language. To review decisions we need to read them; to read we had to learn to formulate sentences; To formulate sentences, we had to learn to pronounce and understand words; To understand words, we had to learn the alphabet and vowels. While understanding language and communicating using it seems easy to us, we fail to understand its underlying reasons and the fact that processing language and creating it has cost our brain millions of evolutionary years since we set foot on earth, for both cognitive and social reasons.[2].

AI seeks to create systems that can emulate that language processing that for us is natural and daily. To do this, it has generated different methods and approaches which border on syntax, semantics, pragmatics and others. Although I will not stop talking about each type of approach, it can be said that the techniques to understand the language range from the identification of similarities of the signs, to the attempts to generate ontologies that allow the words to be understood semantically. When you say to a human "look at that dog", this immediately mentally represents the object to which the word dog refers (a vertebrate mammal with four legs with ears and barking) and although the systems have not yet reached such natural sophistication, they have engineered them to make systems "understand" words in their own way.   

Despite this, the truth is that today there are systems that can "understand" natural language. Reaching this state has been the product of many years of research that is still in constant improvement. Making systems "understand" language has been the first step in creating other systems that can i) translate texts; ii) summarize and find patterns in large amounts of texts (books, articles, laws, etc.); (iii) e, identify and create arguments[3]. For this reason, advancement in natural language processing is the fundamental connection between AI and law.

Some types of systems applied to law: As a result of natural language processing techniques, support applications have been created that allow summarizing legal texts[4], look for previous decisions applicable to the specific case, or systems that give answers to certain legal questions, the famous “question-answering” type systems, famous for the appearance of a system from the company Ross Intelligence[5]

3.2 Machine learning (machine learning)

When AI is dedicated to machine learning, it is primarily dedicated to studying and emulating machine learning. To do this, they look again at the human being, but this time wondering how they learn. As I see it, and following some authors like Berberich, Krause or Natterer, we learn in two ways: through rules or through examples. To clarify this, let's put an example similar to the one posted by Berberich (2019).

Juan wants to learn how to bake a cake. To learn how to make a cake, Juan can follow the unique and exclusive steps of a recipe: put a tablespoon of butter, sift the flour before making it, etc. The other option is to go see how Pedro, his brother, prepares a cake for a couple of days and for Pedro to teach him the rules but also, based on his experience, tell him which ingredients to increase or decrease.

Artificial intelligence, figuratively, also "learns" in the same way. Through specific programmed rules or through “examples” that can be shown to them. We refer to this last method of showing “examples” when we talk about machine learning, which mainly consists of a kind of prediction. Let's go back to an example for clarity.

Let's imagine that I want the system to learn to differentiate apples from pears. The first thing I have to do is put a label on each one, "x" for apples and "z" for pears. I enter one hundred images of "x" and another hundred of "z" into the system. The system identifies the characteristics of "x" and those of "z", classifies them and differentiates them. This is called the training process. Once trained, new images of pears and apples are entered and I let the system sort them into "x" or "z". From the previous data that I entered and with which I trained you, the system can predict whether these new images entered belong to “x” or “z”. This type of learning is called supervised machine learning.

Although there are other types of learning system such as unsupervised or reinforcement, for the present purposes, it is enough to mention this type of learning.

Example applied to law: In the legal sector, supervised machine learning techniques have been used for “Discovery” tasks[6]. Discovery is an institution typical of common law countries where the parties request the adversary's evidence to solidify their respective defense or prosecution hypotheses. In this sense, in the USA, starting with the Monique Da Silva Moore case, the use of systems with machine learning that allow classifying evidentiary material has begun to be institutionalized. The example of the Da Silva Moore case consisted in a few words of a complaint about a series of "sexist" behaviors promoted by a global policy of a company (x). To corroborate this, the complainants asked to analyze an overwhelming number of emails that exceeded the millions where, it was alleged, they could find sexist treatment. The company and the law firm that advised him, took out expenses and were surprised that doing manual searches of each email was going to be very expensive and laborious. For this reason, they decided to hire the predictive coding systems service, which allowed them to train the system with certain documents and "teach" them what kind of documents were "relevant" and "not relevant" evidently for the established purpose. After being trained by the paralegals, the system was able to carry out this task.

4. Will they replace us?

From the examples mentioned above, we can infer that everything boils down to support systems in a certain activity. An intelligent jurisprudence search engine, an evidence classifier or a system that answers a legal question based on current legislation, does not replace the total work of the legal operator, it simply facilitates it.

Gone are many attempts to generate a Strong AI, and although there are still many people who have faith in this happening[7], the current course of things has turned their interests to implement collaborative systems, where the human and the machine coexist and can enhance their work.

On the other hand, although it is true that generating these systems presuppose a minimum possibility of replacement for those who carried out research work on jurisprudence or relevant probative material for the case such as in Discovery issues (the case of paralegals in American legal studies where interns, junior attorneys, and assistants were in charge of manually searching for relevant evidence material), there are a number of principles to keep in mind each time these systems are purchased.

These principles presuppose, among other things, measuring the impact of the AI ​​system on the tasks that will be replaced and analyzing whether their acquisition generates more benefits than losses. And here the losses not only refer to the economic question, but also to human material. In very general lines, the question that must be asked and to which it must be answered is simple: If I acquire an AI system that automates certain work, do I put at risk the job of someone else who performed that task? If the answer is yes, the manager must relocate the person to an area not yet automated. However, the purchase of a system not only automates but also generates new jobs that need new profiles[8]. For example, in the aforementioned case of classifying tests, it is obviously necessary to have personnel who supervise the work of the system and train it.


This paper is very short to explain the whole world of AI and its uses in Law. Every day that passes, new applications and new projects begin to be generated in the world and following the pattern takes time and dedication. In this work I have tried to explain in a simple way what AI is, where it is related to the Law and if there will be replacement.

While what I have mentioned are not the only connections and I have omitted some that are even more stimulating, such as the modeling of legal reasoning or the prediction of court decisions, I consider that the two techniques explained: natural language processing and machine learning , are the essential techniques to understand any relationship between AI and Law, since without them I cannot imagine any considerable help system.

Regarding job replacement, although there are areas that will be partially affected, it is the obligation of employers, and their policies, to take into account the impacts and seek solutions to a possible replacement. Job relocation is part of the answer. However, incorporating systems also enables new types of work that may be incorporated, enabling unconventional jobs for which new professional profiles will be required.

*The opinions expressed in this article are those of the author and do not necessarily reflect the views of the administrators of The Crypto Legal blog or the Lawgic Tec association.

[1] Read the hilarious blog of José María de la Jara published in Interfaze about the sale of smoke and its relationship with the legal practice, called "Jurassic lawyers and how to deal with them":

[2] Here it is important to emphasize that both language and argumentation not only have biological bases that allowed their development, but also of a social nature. The social nature of language and argumentation is directly related to cooperation, the order of beliefs, and avoiding conflict. On this, see Santibañez (2018).

[3] Here it is necessary to emphasize that the creation and understanding of argumentative blocks is still in relatively early development. As a result, the most important forums on AI in the world have been a central topic of debate.

[4] Here I would like to mention the recent beta application on the Web designed by Elen Irazabal called "Resume {}" which under a friendly interface allows you to summarize texts.

[5] However, it should be clarified that ideas and project of question-answering systems have already been tested for a long time. A project by the Italian jurist Elio Famelli from the 80s in Italy is proof of this.

[6] See a comprehensive review of this case in Solar Cayón (2019).

[7] In a survey conducted at an AI conference in 2006, AI experts were asked the following question: “When will computers be able to simulate every aspect of human intelligence? 41% answered that in more than 50 years, another 41% answered never and the others simply did not have an opinion.

[8] On the new types of work and the need to have new profiles as a consequence of the acquisition of AI systems, I have written a forthcoming article in the next issue of the IUS INKARRI Magazine of the Ricardo Palma University.


Berberich, N. (2019). Algorithmen. Über die Kunst, Computer zu. In K. Kersting, C. Lampert, & C. Rothkopf, Wie Maschinen Lernen. Künstliche Intelligenz verständlich erklärt (pp. 11-20). Springer.

Guastini, R. (2001). Il Diritto eat linguaggio. Lezioni. Giapichelli Editore.

Mill, M. (2016). Artificial Intelligence in Law: The State of Play 2016. Thomson Reuters Legal Executive Institute, 1-6.

Rich, E., & Knight, K. (1994). Artificial intelligence (Second ed.). Madrid: McGraw-Hill.

Santibáñez, C. (2018). Origin and function of the argument. Steps towards an evolutionary and cognitive explanation. Lima: Palestra.

Solar Cayón, JI (2019). Legal Artificial Intelligence. The impact of technological innovation on the practice of Law and the legal services market. Navarra: Thomson Reuters Aranzadi.


- Advertising Notice-
Gabriel Uscamayta
Bachelor of Law from the Andean University of Cusco. Scholar of the University of Genoa. Member of the Competition and Intellectual Property Area at PRESTON + legal advisors. Co-founder of the School of Legal Culture, EMPATIA Lab and LAWnely.


1,954Happy fans


*All fields are required
es Spanish