TY - JOUR AU - Lozada-Pimiento,, Nicolás AB - I. Introduction Automation is everywhere. It is an important part of our financial systems, social networks, and even the mechanisms we use to solve our legal disputes. While many professions associated with law have been identified as being safe from automation,1 this does not mean our industry is immune to technology. It means that, in the current state of technology, there are still vital jobs within the legal field for which machines need not apply.2 Every lawyer uses databases to review precedents, applicable laws, and rules of adjudicative processes. In a broad sense, as lawyers, we rely on artificial intelligence (AI) even when we ‘Google’ a relevant subject to research a case. Even these everyday examples illustrate the extensive impact that technology has on our profession and the subtle definition of an ‘intelligent system’. They also show how important it is for lawyers to understand how technology is affecting our field. In a broad sense, an intelligent system could be defined as any system that takes the information provided by the user and gives, in return, a processed output that should be useful to that user through an informatic process. This includes not only artificial intelligent systems but also general databases and even recording and time-tracking systems. A human-based system is the union of an intelligent system with a comfortable interface designed to suit the specific user for whom it is designed. Far from being superfluous, this classification entertains the complex process of translation between machine and human languages and logic frameworks, which is a practical and relevant limitation for AI applications. Among the above, knowledge-based systems rely on data sets and informatic algorithms to process and retrieve relevant information from the entire set. As a subset of knowledge-based systems, rule-based systems are simpler and deterministic systems that return logic statements based on the premises introduced by users and the logic connections and conclusions programmed in advance. Finally, decision support systems rely on probability and other mathematical tools to provide answers based on incomplete inputs or data sets. The operation of these categories of human-based systems can be compared to a mathematician that solves a simple multiplication (rule-based systems), studies a problem from different perspectives (knowledge-based systems), and gives an educated guess for a complex problem (decision-support systems). Neither of the above definitions aims to be precise but to provide non-expert readers with an intuitive framework to understand AI and similar systems. However, this definition is not far from the discussions that have taken place around AI from the early stages of its development.3 One area of law where intelligent systems can be used is dispute resolution, which can be defined as an organized procedure for adjudicating the uncertain rights and obligations among two or more parties. This implies that dispute resolution does not inherently depend on the intervention of a judge or any third party—rather, this procedure could be managed by the parties themselves. This could apply in direct negotiations, commercial transactions, and even legal counselling. However broad the above definitions are, implementing an intelligent system into dispute resolution procedures requires a certain degree of formality to ensure the process abides by the judicial order that defines and guarantees property rights and reciprocal obligations. This is particularly important for lawyers, but it should be noted that this can be said for every dispute resolution system—formal or informal. In this article, we will examine this approach and discuss the impact of applying technological and intelligent systems to dispute resolution. In reviewing these concepts, we will present cases in which systems have been applied to the law (such as computer programs, online services, and neural networks) and consider the potential limits of applying related technologies to the law. This article will (i) introduce the technological revolution that is shaping our world; (ii) describe some intelligent systems and analyse their applicability to dispute resolution; (iii) explain some of those applications from both a theoretical, abstract perspective and a practical, case-by-case perspective; (iv) examine the current limits of technology on the legal profession; and (v) present a series of conclusions about how intelligent systems can and should impact the law. 1. The four industrial revolutions Humans are technological animals. More than any other species, humans can shape their environment using the tools they invent and develop. Then, in time, they replace these items with new technologies. Here, we will focus on four large-scale and geographically widespread transition periods that began shaping our world, beginning in the 18th century, and have guided us through to the current ‘fourth industrial revolution’. The first industrial revolution was marked by replacing hand-production methods with machines and led to a major cultural, economic, and social transformation. Machines replaced workers and forever altered the nature of work in society. The negative outlooks towards technological revolutions date back to this period, and for many whose lives were upended by the changes, this apprehension was well founded. Indeed, not only did machines displace a vast number of low-skilled workers but societies also did not develop solutions to integrate these workers into the new economy.4 The second industrial revolution was marked by energy innovation, including the development of new sources such as natural gas, petroleum, and electricity. New means of transportation (including the automobile and airplane) and communication (including the radio and the telephone) were foundational causes of socio-economical change. Along with ancient precursors like the Roman Empire and the Silk Road, the cornerstone technologies born during this revolution were among the most influential factors that gave rise to the modern reality of globalization. While putting an end to some traditions that had been commonplace for millennia, this revolution introduced a new way of life across a truly global economy as trade, transportation, communication, and other activities advanced and changed at a pace never seen before. Alternative dispute resolution methods developed substantially during this period. In large part, the resulting population growth and new communication channels created the necessity for alternative methods to solve legal disputes.5 For the first time, we began to see how deeply technology could affect law and dispute resolution. Furthermore, centres of arbitration might be thought of as technological tools for dispute resolution. Indeed, without computers and technological systems, it would not be possible for arbitration centres to be the hubs for hundreds—or even thousands—of international and diverse legal disputes. And without globalization, such a need would not exist. These developments highlight how technological revolutions can disrupt and, at the same time, generate entirely new fields of human activity. Having the ability to adapt to an evolving economy is essential for any profession, including law. The third industrial revolution came in the late 20th century. It is characterized by computerization, renewable energy, and the emergence of ‘green technology’. For Jeremy Rifkin, economic transformations usually occur where there is a convergence between communication technologies and new mechanisms to extract energy.6 Computers replaced a diverse group of jobs and even entire economic sectors. While previous revolutions mostly displaced what we now called low-skilled workers, the third revolution began to menace expert and high-wage professions, including those in the financial services and accounting sectors. These professionals may be the future’s low-skill workers, as machines take over a substantial fraction of their activities. In terms of dispute resolution, research databases began to be created, including precedent and law databases. While most lawyers have participated in the construction of these databases, perhaps without knowing, the impact of this development has not yet fully manifested. The currently emerging fourth industrial revolution is raising the power of computers and automation to a much higher level and represents an entirely new way of organizing the means of production. Technology can enhance human abilities by increasing labour efficiency at a given task. Carrying out some complex functions may be out of reach for current AI systems, but computers are known for outperforming humans in math and related algorithmic processes. Furthermore, AI can learn from interactions with users and additional data to maximize efficiency. For example, some specialized law firms already use intelligent systems to administer contracts, communications, procedural matters, and related legal materials. While the bulk of the legal work done by intelligent systems thus far is not advanced, this is only the beginning. Nothing prevents the looming developments of the fourth revolution—just like the previous revolutions—from posing a major threat to even highly trained and specialized professionals, including lawyers. The notion that machines will replace their trained and gained skills is a common source of discomfort for specialized professionals.7 But the impact of AI is already visible in the worlds of financial analysis, medicine, and accounting. As in the legal field, securing work in these professions requires a significant investment of time, effort, and money. Because the fourth industrial revolution hit them first, it is no surprise that these professionals have a more acute perception of the threat represented by technology. Because of this, many have already begun to retrain themselves for higher-level work that requires a type of cognitive ability, emotional intelligence, or specialized knowledge that a machine cannot (yet) achieve. These professionals are learning that they need to be more and that they must future proof their skill set. And those in the legal field are starting to recognize that change is coming for them as well. II. Human-based systems Intelligent systems are those that attempt to imitate human-reasoning processes. Human-based systems, by contrast, are informatic systems, including software and hardware, that use artificial intelligence techniques that attempt to mimic the processes of human reasoning, learning, memory, and communication. Through the application of these techniques, machines can be thought of as brains, and brains can be thought of as machines. What does this mean? For example, to be able to make a decision, human brains require information that has been previously acquired by experiences and stored in memories. In a human-based system, there is a similar process required, and it is called a ‘knowledge-based system’. In the case of the human brain, it processes all of this acquired information largely depending on the electrical activation and signal transmission of neurons. Through a natural synapsis, neurons process the information and return signals that activate other body systems related to movement, awareness, emotions, and other biological functions. These processes are those imitated by neural networks, one of the most advanced and complex tools for AI systems. 1. Expert systems Kenneth Quinn defines an expert system as an interactive computer program that asks questions that could be answered by a human expert. Based upon information from the user, the machine provides the same answer that the expert would if both the system and the human expert have been given the same information.8 Historically, they are the precursors of modern knowledge-based systems. In their structure, they are rule-based systems. An expert system consists of three parts: a knowledge basis an inference engine a user’s interface. While highly relevant for law practitioners and dispute resolution, expert systems are the most limited. They work with a set of predefined rules and strictly logical deductions. For this reason, these rule-based systems are commonly excluded from more strict AI definitions, which require a system to have the capacity to learn before considering it ‘intelligent’.9 Just like calculators cannot conduct operations not previously programmed on them, rule-based systems cannot learn. Still, we place them into our discussion of AI systems because, as with the calculator, they can be of great use for law practitioners. They might not be able to propose a coherent strategy to resolve a highly complex commercial relation. But they can respond to questions like: ‘If my counterparty breached the contract on April 19, how long do I have to recourse to courts?’ Depending on its programming (and assuming it can understand the above formulation of the question), an expert system might be able to search its database for a generic statute of limitations. It may also be able to ask the user for more information before formulating a response, such as location of the parties, the location where the contract was signed and executed, and the jurisdiction where the user expects to bring his claim. Expert systems are based on rules that relate to statements of facts, applicable sub-rules, and a derived conclusion. For lawyers, this is the so-called modus ponens, which states that if a conditional statement is accepted, a necessary conclusion follows (for example, ‘if p then q’). This is also the basic structure of laws and rules. As stated by Norberto Bobbio,10 a precept is composed of an abstract statement about a factual hypothesis and a consequence for the occurrence of those facts. Hans Kelsen11 also proposed a perspective of a self-containing, self-referring system of hypercritical rules, which we know as the pyramid of norms. Herbert Hart’s distinction and classification of rules,12 as well as others legal theories, quickly seems to follow the same pattern when analysed. From this perspective, the difficulty of applying AI systems to the field of law is not the adaptation of law practice to artificial intelligence standards. Quite the opposite, law is always aiming for the deductive standard used by basic—and more advanced—intelligent systems. The difficulty lies, for this matter, in the adaptation of facts to the highly specialized language of law. Lawyers and law-related professionals spend a significant portion of the time they use to resolve a case adapting the facts to a narrative accordant with the specialized language of law. This is an obstacle for the application of AI systems to law, as the expected impact on time saving and effort reduction might be diminished by the necessity of adapting the facts of every particular case to the precise language of law and then to the one of the AI system. To perform this task correctly, the law practitioner will actually do most of the work required to solve the case because once they have ‘subsumed’ the facts related to the rules (the translation process), they will likely already know the answer to the juridical problem that the AI system was supposed to provide. In other words, the more a task requires the reasoning process imitated by machines just to be dealt with, the less the machine will have to offer to the solution of such task. When lawyers say they do not have—or have inconsistent—task-solving patterns, they are expressing the fact that the previous steps to the logical formulation of a legal problem are, indeed, not ‘logical’ themselves or, at least, not in the formal sense we used this word above. Rule-based systems are the primary source of information used by more complex intelligent systems to add new ‘cases’ to their databases. For machines, as well as for lawyers, cases and case-based systems imply the use of previous experiences in an organized, adaptative, and pondered decision framework. The information fed to these intelligent systems, as well as the basic rules it still has to follow to examine the fed cases, is what determines its reliability and general performance. In short, while AI systems can be trained to learn from data instead of requiring precise instructions, they will always rely on training data from the system programmers, large, good quality data-sets, and a fundamental algorithm design. This implies that there may be a risk of bias and a practical limit for the application of AI systems to law. 2. Case-based systems As somewhat of a second-level system, case-based reasoning is a problem-solving process in which a new problem is solved by retrieving a similar situation and presenting a similar solution. Transfer learning occurs when, after gaining experience from learning how to solve a problem, a system uses this experience to improve performance and learning on other problems.13 A clear parallel between this type of AI system and the law comes in the form of precedent, which compels judges to rely on relevant historical cases to rule on present situations. Again, it is quite interesting that legal practitioners, particularly in the dispute resolution process, apply the same techniques that these AI systems apply. Case-based systems highly resemble Ronald Dworkin’s Hercules, which, as stated by the British jurist, would be capable of deciding the most difficult cases by relying on ‘infinite time and efforts’ to revise and compare all relevant cases, combining all of the different interpretations of this material, to decide a specific case (a hard case). Dworkin even proposed some variables and algorithms that his judge would follow to analyse case records, which could be a valuable source for algorithm designers interested in mimicking a judge’s task-solving processes.14 Applying law to technology is, as far as I know, an unexplored field. 3. Decision support systems Different from the others, a decision-support system is a specific type of system that helps organizations make decisions even in the presence of uncertainty, relying mostly on complex mathematical and engineering techniques, which this article does not aim to explain. They represent a step forward because their algorithms explicitly account for uncertainty and probability measures. With this capability, they are able to provide useful insights to users even when they lack all of the relevant information to determine an answer with 100 per cent accuracy (in cases where arriving at a fully factual answer may be impossible). 4. Machine learning Machine learning is a leading technique used within AI systems. Voice-recognition systems, such as Apple’s Siri and Amazon’s Alexa, Facebook’s facial-recognition system, and self-driving cars are prominent examples of machine learning. Artificial ‘neural networks’ are the structure underlying machine learning. While biological neurons constitute the basic building blocks of the human brain, neural networks are designed to mimic their electric activity and connection principles (synapses). View largeDownload slide View largeDownload slide Essentially, neural networks combine and expand upon the features of the previously discussed intelligent systems. They apply a self-built set of rules that are developed by analysing previous cases and can provide answers when uncertainty remains. As with the previous systems, neural networks are highly sensitive and dependent upon the quality and quantity of the information that was fed in to design and train the system. But they can take this information and combine it in unforeseen manners, which are not always comprehensible even for their programmers. They find logical connections without such connections being thought of and are capable even of finding non-logical patterns in data.15 While beyond the scope of this article, it is important to note that the nature of neural networks opens a virtually unforeseeable set of applications in law and related fields. The Ross legal research platform, perhaps the most famous example of AI applied to law, is based on deep learning and neural network techniques. III. Applying AI to dispute resolution Dispute resolution is necessary for all areas of human activity. While judicial intervention is the exception, rather than the rule, lawyers and law-related considerations are always a relevant perspective to adjudicate the rights in a dispute during resolution procedures. The dispute resolution mechanism can be used to determine uncertain rights and obligations in a wide range of areas, including disagreements among corporate partners over profit distributions, divorce proceedings, criminal inquiries, or even discussions among friends about where to have lunch. Dispute resolution requires, as mentioned above, an organized set of steps and phases that guarantee that uncertain rights and obligations are resolved. Indeed, uncertainty can only be settled through a certain, strict, and a somewhat rigid procedure. These procedures, in turn, require a registry, which is a proof of compliance with the required steps to resolve a dispute. This is because, as with any ownership right or obligation, there must be some record of the legal right.16 One of the primary characteristics of intelligent systems is the use of big data, something that is easily applicable to judicial decision-making processes that require extensive record keeping of the information produced during procedures to guarantee legitimacy. Though this record-keeping process is inefficient and time-consuming for law practitioners, it is a good source of information for intelligent systems. Thus, as they are fed more and more data, AI systems can provide relevant information for lawyers, judges, and clients during the resolution of their disputes. Rule-based systems, for example, can inform parties to the dispute about the resolution mechanisms available as well as the associated costs of these mechanisms. Precedent consultation systems are already widely used in legal offices and courts, and they are also available to the non-professional public. These systems can respond to questions such as ‘what has been said about this particular rule or concept in this particular jurisdiction and in this particular period of time?’ Case-based systems can predict the outcome of a dispute resolution process by comparing the information of a given case with the results of others within its database. Decision support systems can provide information on the risks associated with dispute resolution strategies. Neural networks are said to be capable of detecting patterns unforeseen by programmers and users.17 While this capability is sensitive to the information provided to the neural network during training, the network can detect the influence of extra-legal variables that lawyers might not be good at detecting, such as policy, political, and economic concerns or the risk of corruption among adjudicators. 1. Legal rules, proofs, and procedures A. Legal production Professionals dealing with the creation, application, or interpretation of legal content develop a highly specialized language to fully describe their field. As in medicine, law contains a vast number of terms necessary for (and only applicable to) law. Different from philosophy, art, religion, or other languages, legal terminology pretends to be clear even when it is not. Lawyers, even when using obscure and convoluted interpretations of facts and laws, elaborate their upon the logical correctness of their interpretation more than the majority of professions. And they keep track of that process. From a programming perspective, this level of detail can lead to the creation of a robust system for big data analysis, and law professionals are expected to profit from such innovations in the near future.18 B. A case for proofs Proofs are an essential part of both judicial procedures and dispute resolution systems. Principles like publicity, contradiction, procedural and probatory unity, celerity, immediacy, necessity, and solemnity of other acts of the judicial process (which are transferred by analogy to alternate mechanisms for dispute resolution) entail a significant production, collection, storage, processing, and interpretation of evidentiary material, documents, and technical reports. To some degree, they can all offer high-quality resources to implement big data and, therefore, help implement artificial intelligence systems. As stated above, dispute resolution aims to solve the uncertainty surrounding property rights and reciprocal obligations between two or more parties. And both judicial and extra-judicial mechanisms will depend on the information introduced to the dispute procedures. In the judicial process, this implies that judges can rely only on the information given by legal and useful proofs. The greater the certainty provided by the proof, the greater the impact of automation can be in the dispute resolution process. There is something of a paradox regarding the development of proof regimes derived from technological advances today. As communication and information systems developed, it became unnecessary to keep certain formalities around proofs and judicial procedures. While convenient for interested parties (which, for example, can rely on proofs other than written documents for their contractual relations), this is inconvenient for the application of AI systems to dispute resolution systems, as these systems require not only abundant, but also standardized, information. C. Judicial procedures Even if it is our purpose to insist on a variety of dispute resolution systems, it must be recognized that judicial resources and courts are an essential part of dispute resolution. An important share of commercial and civil conflicts is resolved through State or private courts. Administrative disputes between private parties and the State, as well as criminal matters, are necessarily resolved with the intervention of judges (or administrative authorities with judicial powers) and must follow certain judicial procedures. On the one hand, judges are holders of a non-limited amount of data. It is interesting to note that such voluminous legal content production has even been used by economists to feed theories and predict the outcomes of judicial decisions. Economic analysis shows the existence and influence of hidden extra-legal variables on dispute resolution.19 On the other hand, the case for automation is strong for judicial procedures that rely on strongly technical and/or standard materials and proofs. Consider paternity procedures—cases that can be resolved by DNA tests. If a test is applied and presented as evidence, there is no way to doubt the result, as this is a highly certain and precise proof applied to the principal and only object to be proved: the paternity relation. As stated above, the higher the certainty given by a proof, the stronger the possibility to automatize the process. Therefore, if applying this scientific proof, paternity procedures can be totally automatized. A similar, more-complex example is technical arbitration. This is a dispute resolution process dependent on the decision of a technical, not a judicial, expert, which requires an appreciation of the facts that relies on structured, scientific knowledge. This is a hierarchical process and, therefore, could be an automatable process. Procedures for costs, exequaturs, and custom procedures can be thought of in the same way. In general, every dispute depending exclusively on a technical pronouncement of a third party to the procedure, or the exclusive application of a technical procedure, has the potential to involve automation. These types of cases reduce the uncertainty and appreciation margin of proofs required to reach a final, valid, and reliable resolution. 2. Out of court As stated in the introduction, dispute resolution does not rely on third party intervention. Instead, the principal institution of the law is not trial; it is settlement out of court. Ronald Coase considered that, if transaction costs were zero, property rights would end up with those who could make more economically efficient use of them. When transaction costs rise, economically optimal states are lost.20 As stated by common wisdom, lawyers are, for economists, transaction costs. Parties prefer the mechanism with the lowest transaction cost to resolve their disputes, and private resolution generally will be cheaper than resorting to public institutions. Indeed, courts have been proven to be slow and inefficient in a variety of situations.21 On the other hand, distinct disputes require distinct mechanisms to be resolved. Boaventura de Sousa proposed the existence of an overlapping, yet independent and distinguished, set of rules to be applicable to different dispute resolution systems in different ‘scales’ of society.22 In his theory, courts are only to be used by one sector of society and for a specific range of matters. These examples show how broad the field of dispute resolution systems can be outside of the courts. Today, we can see AI being applied in three types of conflict resolution scenarios: Dispute prevention Dispute management Dispute resolution. 3. Dispute prevention A. War prevention Technology has prevented military conflicts between States. AI is used to model scenarios and to predict possible outcomes regarding different approaches to resolve any potential conflict. However, results have shown a high variance, and the results may have to be analysed by experts in this domain. This is a common source of uncertainty around AI systems, which usually are not capable of determining their own limitations. AI systems cannot say when they are useful and when they are not, and this implies a deep limitation for their reliability for non-experienced users.23 Still, this technique offers a different way to avoid armed conflict between countries and prevent further wars by presenting different alternatives and solutions before tensions boil over. Here, a decision support system is used to provide information on the possible, although uncertain, outcomes of conflict. While different mechanisms to solve conflicts might have different costs, these systems show the more relevant human costs as well as the general effects of war on society and the economy. B. Siarelis Siarelis (an AI-based system for societary disputes resolution) is an example of the application of dispute prevention systems in Colombia.24 Considering the more common conflicts in its area, Supersociedades, an administrative authority, developed this chat-bot that can field questions from users and respond with probable outcomes to their disputes. It has elements of both rule-based and case-based systems, and it also serves as a decision support system. By disposition of Law 1564, administrative authorities were given jurisdictional functions.25 As they adapted and prepared before taking these functions, they seem to have developed more efficient mechanisms on dispute resolution, often related to intelligent systems. Superfinanciera is responsible for overseeing the real-time tracker of Colombia’s financial market. This information is used to calculate the risk of default by the nation’s central bank, Banco de la República (BanRep), thus preventing financial institutions failure.26 Both institutions rely on AI systems to fulfill this task. Superindustria enables users to present their plaintiffs and cases entirely online. However, Supersociedades warns that Siarelis is a mechanism that acts as a guide for users to explore solutions with respect to possible societary disputes. Moreover, this system is not designed to be a substitute of a judicial decision or to predict those decisions. It is interesting that Supersociedades is forced to make such a disclaimer about a system ideally used to make predictions. The necessity of certainty around legal outcomes is an obvious limit to the practical application of developing technologies to law. 4. Dispute management A. Settify27 In Australia, a dispute management system is mandatory for family disputes, aside from domestic abuse. Using this mechanism instead of seeing a judge immediately aims to ensure that the legal problem gets resolved faster and at a lower cost. B. Negotiation and settlement Automated systems are widely used as an effective way to resolve conflicts in far less time and using less resources than the ordinary judicial ways. Using a digital method for dispute resolution also makes the judicial process simpler for the people who do not have access to justice as easily as the people that have the means to pay for their own lawyer or at least for proper consultation. Here, again, decision support systems deliver relevant information on the costs of different mechanisms to solve a legal dispute, thus preventing them from moving to the courts and slow, expensive, and inefficient jurisdictional procedures. C. The Persuader28 This is a mediator that uses case-based reasoning for resolving conflicts in the labour domain. The system keeps track of the agreements found in past negotiations, and, once a new conflict arises, it looks for the most similar past situations. Analysing previous cases enables the system to consider local variables—agreements that might not be written in relevant rules or contracts but whose influence can be detected throughout the history of case resolutions. D. Split Up Split Up is a decision support system that provides advice on property distribution following a divorce. Regarding mandatory systems, like the one used in Australia, human judges are only a higher court for the AI adjudicator, thus reducing costs associated with access to State courts.29 E. Family winner Family Winner is a negotiation support system that extends game-theory principles to deliver advice on structuring a mediation and reaching an equitable outcome.30 5. Dispute resolution A. Online dispute resolution Online dispute resolution (ODR) is the mechanism that settles disputes through online interaction and communication between parties. ODR undertakes disputes that are partially or fully settled over the Internet, having been initiated digitally but with resources in the real world. New disputes are exponentially being created by the interconnection of the world and the inclusion of more individuals in the World Wide Web. For example, this year Colombia implemented a law for online arbitration. It is an online system that allows the parties to create a user name, submit their case, and have an arbitrator appointed quickly. It cuts the processing time from 20 years to two months just by moving everything online. B. Do Not Pay Do Not Pay is a chat-bot that can support the user in any number of difficult legal situations without the need to enlist the support of a costly human attorney.31 It is the home of the world’s first robot lawyer. Its selling point is that users can fight corporations, beat bureaucracy, or sue anyone at the press of a button. C. LISA LISA, another technology claiming to be the world’s first robot lawyer, uses AI to enable parties to create legally binding agreements with another party, helping the involved entities find a middle ground as quickly and cost effectively as possible.32 D. ROSS ROSS is a digital attorney that was built using IBM’s Watson artificial intelligence platform.33 This innovation understands natural language legal questions and provides expert answers instantly, along with other relevant information, cutting down on legal research time and energy. While highly relevant, ROSS and similar technologies are at the early stages of application of neural networks to law and dispute resolution. AI systems lack the capacity to easily detect connections between different sets of information. They are not multi-tasking systems.34 Therefore, while being able to provide recommendations and detect patterns on the data, ROSS is limited by the state of the art on its possibilities to provide, for example, the entire document production of a case. Complex, highly covariant activities are those safe from automation. 6. Decision-making A. E-judge The judicial decisions of the European Court of Human Rights have been predicted with 79 per cent accuracy using an AI method developed by researchers at University College London, the University of Sheffield, and the University of Pennsylvania.35 This method is the first to predict the outcomes of a major international court by automatically analysing case text using a machine learning algorithm. There are not, strictly speaking, decision systems relying solely on AI. As stated below, this is both a limit and a concern related to artificial intelligence and law. IV. The limits of AI On the practical side, it has been pointed out that a lawyer’s task-solving process is not easily traceable, as lawyers do not have any standardize procedures to solve judicial tasks such as drafting a lawsuit, preparing for a hearing, or proposing legal strategies for a particular case. The application of informatic technologies is highly dependent and sensitive to the detection of protocols or decision trees and is, therefore, limited to those tasks that follow a certain pattern or order of execution (like those previously discussed). Below, we will discuss some of these limits. 1. Hidden variables A system that portends to be self-referent and self-contained might face some challenges. Positive analysis, from an economic perspective, has been identified as a circular argumentation used to justify political decisions as an objectively economic advantage.36 Self-referent systems are at least of limited use. Law is permanently permeated by other disciplines and sciences as well as by practical, social, and economic variables. These variables, often hidden from legal documents, are highly relevant for intelligent systems, and their absence represents a practical limitation for these systems. Richard Posner, a US judge, considers that judicial processes entertain the addition of a degree of uncertainty to the rights discussed in the legal process.37 c1=pc0,0