Abstract
This research paper aims to explore, a concept once considered alien, the usage of artificial intelligence (“AI”) in arbitral proceedings. The sphere of arbitration has, to date, been deemed inherently conservative, where change and development have been slow. However, this paper aims to illustrate that the new wave of the technological revolution has now made it difficult for arbitration to stay far behind, or follow obsolete practices. Although, this is not without its challenges, and therefore, the author seeks to strike a balance between the advantages and disadvantages of AI in arbitration, without undermining its very essence. Resultantly, it is argued that its usage needs to be slowly phased in. The discipline referred to in this paper mainly pertains to the realm of International Commercial Arbitration.
Introduction
“Everything has been said, and one comes too late since Men, for more than seven thousand years, exist and think.”
Pierre Lalive muses that this famous sentence can be applied to the field of international commercial arbitration, where nearly everything seems to have been said. Admittedly, while perusal of prevailing texts on the subject matter seems to give the impression that much has been said about the potential involvement of AI in arbitration, it is noted that most of the literature is scattered and speculative.
The use of AI has immense importance in our day-to-day living. It is used to filter spam emails, write newspaper articles, and provide medical diagnoses. More relatable to readers may be the algorithms used by social media applications such as Facebook and Instagram, which display content and advertisement that is personalised and caters to each individual user according to their preferences. More specifically, within the legal sphere and in international commercial arbitration, though in its primitive stages, the usage of AI in arbitration has already commenced, as will be illustrated in this paper, and is no more the far-off prospect that was subjected to dismissal and scepticism. Furthermore, it provides many advantages for speedy, efficient, effective, and arguably fair hearings. AI in arbitration, in its current stages is not without implications, however. This is so because arbitration by AI can pose challenges relating to risk of bias, lack of empathy, unemotional and unreasoned awards.
This paper is divided into several parts. Firstly, Part I endeavours to espouse AI and arbitration by defining the two and explaining their potential interaction with one another. Furthermore, Part II focuses on various scenarios in which AI may be potentially involved in the process of arbitration; each scenario illustrating a greater degree of involvement of AI within the arbitral process. Subsequently, Part III addresses the implications that are likely to arise with this increased use of AI, owing to the unique characteristics of the process of arbitration as an alternative to traditional dispute resolution i.e., the court structure of every state. These implications are not without solutions, however, and the paper will attempt to reconcile them. In Part IV, the author will suggest reforms that can be instated in order to make the transition of AI into arbitration as smooth as possible. Finally, the author will argue that the use of AI in arbitral proceedings should be introduced gradually to allow both lawyers and parties opting for arbitration to become accustomed to this new development. And that there needs to be greater initiative by the states to regulate the data entered into the algorithms in order to ensure a fair and just arbitral process. Since the prevalent use of AI is seen in cases of International Commercial Arbitration, the essay will focus on this discipline only.
The Interaction between Arbitration and AI
AI was the term coined by John McCarthy, in 1956, who defined it as ‘making a machine behave in ways that would be called intelligent if a human were so behaving’. Furthermore, Kathleen Paisley and Edna Sussman’s definition serves as a useful guideline in illustrating how AI operates. They define it as a process where a large amount of data is combined with processing systems, allowing the software to “learn automatically from patterns or features in the data”. However, Paisley and Sussman concede that the term AI is often used loosely and encompasses many subjects including machine learning, and also natural language processing. Ultimately, they conclude that AI is the software’s ability to learn automatically from patterns or features in the data, thereby making it “intelligent”. The ability to develop its own ‘thinking patterns’ is a starting point in understanding the significance of AI, and why it would be pivotal to the development of more effective arbitrations.
The distinction between different types of AI models is illustrated by Jacob Turner, who fine tunes the difference between automated and autonomous systems. “Autonomous systems are those which can take decisions themselves without being explicitly programmed, whereas automated systems must follow a pre-determined set of instructions with no discretion as to how they are to be followed.” Therefore, the difference between automated and autonomous systems is the degree of human intervention in the process. For example, an automated car would not possess the same level of intelligence or independence as an autonomous car, which would not only be driverless, but would also have the ability to self-navigate and decide its destination and route. The fact that the autonomous system has the ability to make decisions proves to be crucial when it comes to establishing legal and ethical rules. Furthermore, the distinction between autonomous and automated systems is also important because other forms of technology are deterministic i.e., they execute pre-programmed instructions from a human. Autonomous systems will be discussed with relation to the great possible intervention of AI in arbitration i.e., the distinct possibility of a machine, or ‘robot’ arbitrator in the future. Automated systems, on the other hand, will be discussed in relation to scenarios (1) and (2) which may be termed as the introduction or ‘easing in’ of AI in arbitral proceedings.
Nowadays, AI is a unique phenomenon that is being widely discussed, especially in the legal sector. An example of the dynamic and novel independent action that may be taken by AI is illustrated by the 2017 victory of AlphaGo—a machine learning system- against the masters of the game Go. The interesting aspect of the defeat was the manner in which the program defeated the Go champion. Essentially, AlphaGo came up with a new technique of playing the game, which no human in history had ever done. This was hailed as a revolutionary development within the realm of AI. The game dates back around 3000 years, and is widely accepted as the most challenging strategy game that exists. Children, especially in South Korea and China, are sent to private schools specifically to learn how to play the game at an expert level. The taxing and strenuous nature of this game is further illustrated by the fact that mastering it takes years of playing for several hours daily. This ability of AI to take independent action is not merely limited to games but extends to every sector, principally the legal sector.
Coming to arbitration, according to Nigel Blackaby and Constantine Partasides, it is essentially a “simple method” of resolving disputes. In arbitration, each party submits their case to the decision-maker or makers, whose judgment they are prepared to trust, known as the “arbitrator”. The arbitrator considers the facts, hears the arguments of the opposing parties, peruses and applies the applicable laws, and ultimately reaches to a conclusive decision, known as the “award”. The award is final and binding on the parties—and it is final and binding because the parties have agreed that it should be, rather than because of the coercive power of any state. Arbitration, therefore, is an effective way of obtaining a final and binding decision on a dispute, or series of disputes, without reference to a court of law (although, due to the national laws and international treaties such as the Convention on the Recognition and Enforcement of Foreign Arbitral Awards, arbitral awards are enforceable by a court of law if the losing party fails to implement them voluntarily).
Previously, AI was considered outside the purview of dispute resolution, more notably in arbitration. This may be attributed to the reluctance of the arbitral community to introduce new procedures for the fear that it could lead to challenges in public courts at the enforcement stage. This is further elaborated upon by Lucas Bento, who argues that the reason for this hesitance is the utilisation of judgment in advocacy, which machines (previously) lacked. It is submitted that these reservations seem outdated in today’s time with the advent of the New York Convention and its pro-enforcement bias. The pro-enforcement bias now imposes a duty on state courts to give maximum effect to the award, subject to the reservations encapsulated under Article V of the New York Convention. Furthermore, the plight of Covid-19 has left the world scrambling to adapt to new routines and new work regimes. AI in arbitration proves to be a worthy addition especially in light of the fact that more and more of the arbitral process is becoming virtualised and digital; the next logical step would most certainly be the incorporation of AI based technology into the process. Therefore, the words of David Gauke seem to have aged very well since his speech at the AI Summit, London in 2019: “Technological revolution is no longer on the horizon—it is here.” This technological revolution has increased technology usage in almost every profession, and the legal sector has proven to be no exception. The use of AI in arbitral proceedings would resultantly present “synergistic opportunities”.
The Practicalities
I. Scenario I: Present Day
The use of technology in the arbitral community has proven to be slow but incremental, more so evidenced by Nappert and Cohen’s observation that technology is still being used in a relatively simplistic manner, which is neither novel nor does it break new ground. They list examples of the ways in which technology is being used in arbitral proceedings, for example, communications between the parties are electronic, and arguments in hearings are transcribed. Admittedly, these concepts do not seem to be either revolutionary or radical. Currently, therefore, AI in arbitration is in primitive stages. It is limited to and dependent upon, the quality of the data processed and the algorithm applied. A potential illustration of the same is that data would be fed into the machine system, which would then generate useful information for the parties wishing to resort to arbitration. The “useful information” may be, for example, the provision of a database.
For example, Dispute Resolution Data is a US start-up that provides a global database pertaining to international commercial arbitration and mediation. The information covers industry type, claim amount, location, cost, duration, and outcomes i.e., whether the case is settled, withdrawn, or a final award is issued, etc. The data is collated from a number of renowned arbitral institutions such as the International Chamber of Commerce (ICC), the International Centre for Dispute Resolution (ICDR) and the Centre for Effective Dispute Resolution (CEDR). Another example is Arbitrator Intelligence, a non-profit initiative that aims to increase access to information about arbitrators and their decision making through post-award questionnaires sent to participants. This initiative also maintains the confidentiality aspect of arbitration fully protected and allows there to be a reliable body of data without the need for the decisions to be published. These examples illustrate the fact that the current situation in arbitration is, for now, merely limited to the provision of information by using AI.
II. Scenario II: Increased usage
In the AI in Legal Services Summit, Sir William Blair narrated an interesting anecdote where one is greeted by a robot in the China International Court in Shenzhen- the Chinese equivalent of Silicon Valley. The sole function of the robot is to direct visitors how to reach their desired floor. If one imagines a more refined situation than the one above, in this scenario, the arbitral world, having fully accepted that AI may be used as a tool for the provision of information, the usage must increase a step further. The automated system, in this scenario would act as an “instructor”. Once again, the instructions are delivered through the datasets and the algorithms that are created for the system, and the system delivers instruction as output. The use of automated systems in arbitration in Scenario II would be the same as in Scenario I; instructional, rather than a method of decision making.
An example of the same would be introduction of AI systems which uses Natural Language Processing (“NLP”), as this would significantly reduce the tasks of a lawyer engaging in arbitration. NLP involves the usage of a special type of software that has the ability to read “natural language” i.e., the ordinary human language. The attribute of NLP lies in its ability to contextualise the language, and resultantly provide accurate results for analysis of legal texts. The utility of NLP is increasingly necessary in arbitration, where lawyers are paid by the hour and clients are reluctant to pay for mundane administrative work. Consequently, there has been a surge in law firms scrambling to make their practices as efficient as possible. NLP could be used to read a contract and identify the existence of an arbitration clause, deciphering the lex arbitri and the lex loci arbitri, the last two being the bone of contention in many cases where both parties fail to specify the same.
Furthermore, the task of translating documents takes up an enormous amount of time in international commercial arbitration, as it is notorious for its huge number of documents and bundles. Often, parties will speak differing languages, or the arbitral agreement itself is in a different language, or the place where the enforcement of the award is sought may again require a translation of the documents into the official language of that state. Allowing an automated system to translate hundreds and thousands of documents would lead to greater efficacy and an overall reduction in the time of the average arbitral proceeding.
Admittedly, NLP has yet to reach the stage where the translation of documents can be done at the precise accuracy that is required of legal texts, especially in arbitration where the arbitration clause—if not drafted properly—can be open to many interpretations. NLP service providers warn users with a caveat that their services may not be perfect for word-to-word precision of the translation and that some words may be lost in translation. This holds true for most languages as contextual analysis is necessary, and a literal translation does not always provide the desired outcome. For example, in recent international arbitration proceedings in which the author was involved, the Parties to the arbitration stipulated that the arbitrator be well-versed with the Urdu language. This was due to the fact that many of the contractual documents, written submissions and evidence were in Urdu. The Urdu language is heavily influenced by the Hindi, Arabic, and Persian languages; many words are derivations of these languages. An unsophisticated NLP processor would resultantly have difficulty in ascertaining the legal context of the words. Therefore, for translation of legal documents, especially those in multiple languages, NLP needs a higher level of sophistication. It is uncertain when this level of sophistication would be attainable, however one can expect that it will become better and more influential for businesses in times to come.
Moreover, AI at this stage could also be used to draft boilerplate terms of the award. Generally, arbitrators are put to task to extract information for the award such as the information regarding parties, procedural history, facts, and details about the arbitration clause which takes a considerable amount of time. The length of time taken to render an award would be arguably shortened if AI software was able to extract the relevant content from the voluminous documents effectively with the help of an algorithm.
III. Scenario III: The Robot Arbitrator
Perusal of journal articles and blogs even two to three years old illustrate the dismissive approach taken with regard to the possibility of introduction of non-human or mechanic arbitrators. For example, Ibrahim Shehata writes that the possibility of having robotic arbitrators is a discussion of the “unknown unknown” and the arbitration community would be better off focusing its efforts upon the “known knowns”. One cannot help but disagree. Indeed, Cohen and Nappert, even then ahead of the curve, urged us to comprehend that “this is not the idle speculation of science fiction”.
The impetus demands discussion of relevant social, legal, and philosophical implications on possible scenarios that involve the use of AI in arbitration, as perhaps we are not at all far away from such a possibility. After all, Kira Systems was established in 2011, but it is only in the past three years that the recent boom has resulted in many of the London firms using this software. Furthermore, in Canada, a robot mediator was used for the first time in the history of mediation. The online tool used algorithms in place of a human mediator and settled a three-month long dispute in less than an hour. This indicates the notion that autonomous AI systems will occupy a significant place in arbitration in the not-too-distant future. This is given more weightage by the fact that there is already talk of replacing tribunal secretaries with AI to assist with legal research and summarising legal submissions or evidence.
A machine arbitrator would eventually operate as an autonomous system as described by Jacob Turner as discussed above. He theorised that the AI system would be able to make decisions autonomously, through its own cognitive and analytical processes without external programming. The arbitral proceedings would therefore be conducted by a non-human arbitrator.
Such technological advancement would pose many questions and issues pertaining to legal theory. According to Daniel B. Rodriguez, we need to encourage legal philosophers to think about this new technology and the legal philosophy underpinning this technology thereby making these discussions fundamental to future jurisprudential precedent.
Implications
In general, the legal community is sceptic about the lack of human element in decision making which poses moral as well as ethical dilemmas. Moreover, one of the preliminary issues likely to arise is how comfortable parties would be with an automated system determining their liberty. It may be assumed that parties trust a human arbitrator because the arbitrator possesses emotional intelligence- i.e., the ability to sympathise, empathise, and rationalise. It is difficult to prophesise the conclusion reached by a machine arbitrator in case of a moral dilemma because, for the time being, empathy cannot be translated into an algorithmic code. However, in 2018, renowned international arbitrator Sophie Nappert, in her humorously titled speech “Disruption is the New Black,” opined that the ability of parties to choose the adjudicator of their disputes is an underlying and inherent thread of international commercial arbitration. Despite the possibility of a machine arbitrator being able to eliminate the risk of human error and unpredictability, the ability to select the decision makers in one’s own dispute is what makes arbitration appealing at a basic and emotional level. However, Nappert warned the audience that scientists and suppliers of algorithms “are currently warning litigation and arbitration users that human decision-making as we exercise it on a daily basis is no better than a lottery. In addition to being costly, time-consuming, and resource-depleting, it is unpredictable and inevitably subject to bias.” After all, one is aware of the famous saying that “justice is what the judge ate for breakfast”, which is implicit of the fact that decision making is influenced by the inherent biases, beliefs, and morals of the adjudicator.
Furthermore, since arbitration is heavily dependent on party autonomy parties would need convincing that the option to choose a machine arbitrator would result in creating more just and fair decisions, and awards that are less likely to be challenged. And while the individualistic nature of commercial arbitration proves it difficult to conceive why parties would be on board if it does not directly affect their case, there would be a need to implement initiatives by arbitral institutions to encourage the same. This could be done by carrying out ‘test arbitrations’ or asking for volunteer parties to be the pioneers of change. Additionally, compilation of compelling research and data in favour of the AI system would need to be shown to parties in order to incentivise the use of AI in their arbitral proceedings. The incremental growth could therefore prove beneficial to the sphere of international commercial arbitration in the longer run.
However, the problem that arises subsequently with test arbitrations is the fact that the confidentiality of the arbitral process, more specifically the arbitral award, is one of the hallmarks of arbitration; and arguably why parties prefer arbitration over litigation. Additionally, international commercial arbitration awards are generally not published owing to the obligation of confidentiality upon arbitral institutions. Therefore, a foreseeable issue with test arbitrations is that parties would be reluctant to provide access to their awards. This poses a significant hindrance to the efficacy of the AI model as AI programs require access to data. This holds true especially for machine learning models, which are based on probabilistic inferences and are dependent on data for their operation. The higher the volume of sample data, the more accurate the model’s predictive value. However, Scherer urges not to despair as there are existing initiatives that publish commercial awards on a regular basis, typically in a redacted format. Furthermore, she states that in any event, even without publishing confidential awards, institutions could collect them and make them available for the purpose of building AI models.
Secondly, it is important to note that the vast majority of national laws are silent on the appointment of a machine arbitrator. A notable exception is French law, where only a “natural person having full capacity to exercise their rights may act as an arbitrator”. This would raise obvious implications with the introduction of AI systems into arbitration. One would counter this with the argument that France is famous (and often notorious) for being one of the most arbitration-friendly jurisdictions in the world. It seems incomprehensible that this piece of legislation would not be interpreted in an arbitration-friendly manner by the courts or may even carry the possibility of revocation, keeping in line with its pro-arbitration image. Under English law however, as per the English Arbitration Act (“Arbitration Act”), the objective of arbitration is to obtain a “fair resolution of disputes by an impartial tribunal without unnecessary delay or expense”. The parties are free to agree on how their disputes are resolved, subject only to such safeguards as are necessary for the public interest. While there is no explicit permission nor prohibition of machine arbitrators, a foreseeable problem may be the issue of ‘public interest’ under English law. Although often construed narrowly, this may be an instance where the public interest exception to the review of an arbitral award by a machine arbitrator may be relevant. This is especially so what with the current situation being that there is a lack of explicit standards that algorithms abide by; algorithms are often designed in such a manner that, while the outcome is provided, the reasoning behind the same is not. This becomes even more problematic in light of the fact that there is no review of the merits of the arbitral award, save for the Section 69 exceptions under English law. Ultimately, this becomes an issue of democratic importance as it is difficult to imagine a scenario under which both parties would consent to the machine arbitrator delivering its award and not knowing how the arbitrator reached its conclusion. Moreover generally, the lack of reasoning in an award makes the award susceptible to challenge in many countries.
Additionally, arbitral institutions are also mainly silent on the use of technology in international arbitration. An exception to the same is the Hong Kong International Arbitration Centre (“HKIAC”)’s Administered Arbitration Rules (in force 1 November 2018). They are the first set of rules in Asia to mandate the use of technology by the arbitral tribunal in adopting “suitable procedures”. Technology is also mentioned in the ICDR Rules of 2014. However, given the contractual freedom in arbitration to choose the procedure to be followed, it is safe to presume that institutional frameworks will not operate as a bar. Article 19 of the United Nations Commission on International Trade Law (UNCITRAL) Model Law on International Commercial Arbitration explicitly provides that the parties are free to agree on the procedure to be followed.
Thirdly, a redressal must be made of what the author refers to as the two C’s: “Complacency” and “Lack of Confidence”. Professor Burkhard Schafer, in his discussion at the AI Summit, mused whether there is a possibility of lawyers losing the intuition and knowledge that they attain from reviewing bundles and bundles of documents and whether the ability to “fine tooth comb” a large volume of documents may be lost. This may pose a threat to the training of lawyers in international commercial arbitration, where trainees are acquainted with the process by initially reviewing, tying, and bundling volumes of documents. This may be countered by the fact that introduction of AI systems would result in trainee lawyers being acquainted with the actual dispute resolution proceedings ab initio, giving them greater opportunity and responsibility, and leaving painstaking administrative tasks to the AI systems. The ultimate prospect of greater efficiency would arguably outweigh all other considerations.
The other foreseeable problem is the prospect of lawyers lacking the confidence to challenge the decision of the machine arbitrator owing to blind trust in the algorithm. As there is the risk of the AI system giving an erroneous decision, this must ultimately be kept in check. The resolution lies in the retention of human control over the arbitral proceedings, at least initially. While the machine arbitrator may be used in arbitral proceedings, the ultimate decision ought to lie in the hands of the human arbitrators. An odd number of arbitrators, with the ratio of two humans to one machine, allows the AI system to be afforded the opportunity of giving a novel outcome. The human arbitrators may then choose to adopt the decision of the machine arbitrator, either in whole or in part, if they believe that the outcome reached by the autonomous system is better suited to the dispute than their own.
Fourthly, there is a tendency for arbitral institutions to grant full or absolute immunity to arbitrators. An example is the ICC Rules, where Article 41 of the Rules gives immunity to not only the arbitrators, but also extends this immunity to other employees of the ICC, save to the extent such limitation of liability is prohibited by applicable law. This problem also arises in a country like England, where the Arbitration Act allows for complete immunity of the arbitrator unless the act or omission is shown to have been in bad faith. This is problematic because if a human arbitrator repeatedly delivers erroneous judgments, it would render the arbitrator liable for removal from that case and not made a choice for further adjudication. On the other hand, accountability of a machine arbitrator presents a unique dilemma of to whom is responsibility attributed to? Does culpability lie with the creator of the algorithm, the person who entered, or “fed” the data in the algorithm, or the intrinsic self-learning processes of the machine itself? Furthermore, given the total immunity of the arbitrator, what is the consequence of an erroneous judgment? Charlie Morgan, proposes that in such a scenario, states must ensure that there is a system of reverting back to national courts. A logical question that arises as a result is whether this would mean de-characterising the very nature of international commercial arbitration, which limits access to courts (hence a form of ‘alternative’ dispute resolution)? Perhaps not. About domestic court involvement in the arbitral process, Julian Lew says: “National court involvement in international arbitration is a fact of life as prevalent as the weather.” In the four characteristics he lists of international commercial arbitration, he explicitly states that arbitration does not operate from the relinquishment of jurisdictional control by states. Lew, however, concedes that the access to the “autonomous domain” of international arbitration is obtained through contract between parties that lead to relinquishment of rights by national courts. This may be countered by the notion that this “relinquishment” of rights by the national courts is limited to certain situations where the role of courts is curtailed, such as competence-competence i.e., the ability of the arbitral tribunal to ascertain its own jurisdiction, the merits of the arbitral award, and the arbitral proceedings themselves. It may also be noted that the principle of competence-competence itself is applied differently in different states, and most national courts prefer to retain some sort of supervisory jurisdiction over the arbitral process.
Finally, it is both an attribute and a shortcoming of AI that the data that is being fed into the system is dependent on the type and quality of the data. The driver of change (or lack thereof), therefore is the data that is being produced. According to Duncan C. Card, the algorithm can be developed according to prevalent standards thus, it can be customised and “attacked”, making the arbitral system more transparent, neutral, and just. However, it is virtually impossible to cater to a situation where the system is completely neutral as the algorithm. This is because AI currently hinges on the data fed into the system; the manufacturer of the data being will ultimately incorporate some of their own biases into the system, regardless of how hard they try to maintain neutrality. Thus, AI, at least in its early stages has the ability to create the perfect storm. This holds especially true in the case of arbitration and arbitral bias. If the data fed into the system is one in which racial, cultural, and religious biases are predominant, the result is less likely to produce a just and fair conclusion. One of the most controversial examples of algorithmic bias was the system developed by the Durham Constabulary in the United Kingdom. The Durham Constabulary and computer science academics developed the Harm Risk Assessment Tool (“HART system”). This AI system was designed to predict whether suspects were at a low, moderate or high risk of committing crimes in the foreseeable future, and used 104,000 histories of people previously arrested and processed in Durham custody. In 2017, the HART system incorporated new data, the purpose of which was to reduce reliance on postcode predictors as an indicator of whether or not a previous suspect was likely to commit a crime based on their neighbourhood. A dataset called “Mosaic”, developed by a company called Experian was used. Mosaic was based on profiles of all 50 million adult residents in the UK, assembled using data gathered from public sources, including the internet. Mosaic used offensive stereotypes to brand people and defined categories, for instance, by reference to age group or ethnicity (e.g., “disconnected youth”, “dependent greys”, “large extended families in neighbourhoods with a strong South Asian tradition”, or “Asian heritage”). Due to outcry by the public and the NGO “Big Brother Watch”, Durham Police stated that it worked with Experian to improve its understanding of local communities, and in 2018, it stopped using Mosaic. This seemingly dystopian algorithm serves to illustrate the bias that may come with a robot arbitrator- in terms of nationality, race, caste, etc. “Intellectual corruption” as Karen Mills terms it, is already a prevalent issue in arbitration where the arbitrator’s opinions is tainted with prejudice or racial bias. This is especially so in cases where western arbitrators sit to adjudicate disputes between western and “Third World” parties. Mills argues that intellectual corruption may range from simple cultural misunderstanding through cultural bias to actual racism. A western arbitrator may pay greater credence to a western witness than to an Asian one. She urges against falling into this ethnocentric trap.
The question then to be posed is how realistic it would be to expect global standards for algorithms? Hurly, argues that the difference in cultures all around the world means that there is no way of unifying standards that would be one size fits all. The attractiveness of arbitration stems from its flexibility—i.e., parties can choose to apply law, no law, or a combination of both, or internationally recognised principles such as lex mercatoria, ex aquo et bono, or religious law etc. One would argue that the method chosen by the parties to resolve their dispute is ultimately reflective of the legal culture of the parties’ home states.
Recommendations and Looking Ahead
Firstly, states ought to invest their resources into discerning how algorithms work in practice and how they come up with conclusions. The focus needs to shift from the implications of AI to the actual technology itself. AI acquainted lawyers are virtually unheard of and are often treated paradoxically. Indeed, one author has fated them to be as “rare as vegan butchers”. The introduction of AI, therefore, would require an overhaul of the current legal regime, where lawyers are familiar with the mechanics of AI. While arguably this is an expensive task, the initiative by the United Kingdom’s Ministry of Justice to provide £2 million to support lawtech is proof of the fact that states are now willing to delve deeper into the exploration of this area. Whether or not these funds will be used for arbitration is a different matter as it does not neatly fit within the conservative and traditional legal model. However, with England being one of the friendliest arbitration states in the world, even if the £2 million is not to be utilised on the discovery of AI in arbitration, it is only a matter of time that resources will be allocated towards this task.
Furthermore, regulation of AI in arbitration is essential, and can be done at two levels. In Scenario II, for example, there would be a need for regulation of the algorithm itself. It would have to be certified, accredited, and the credibility of the algorithm would ultimately be dependent upon the reputation of the algorithm in the market. It is suggested that just as the parties to the arbitration are allowed to agree upon the choice of law to govern their dispute, this inherent consent-based system of dispute resolution would allow for the parties to choose different types of algorithms to govern their dispute. The Law Society Technology and Law Policy Commission’s recommendation of keeping a national register of algorithms that are logged into the book would be a welcome proposal, with a separate section for the algorithms to be used in arbitration. Moreover, in the case of a machine arbitrator, the decision of the machine arbitrator is already susceptible to check and balance owing to the fact that there are two other arbitrators on the panel. It is their prerogative whether or not to accept the decision of the machine. Interestingly enough, Nappert and Cohen have suggested the converse; they suggest that AI could be used to correct the biases of a human arbitrator and one could compare the decision of the human arbitrator against the decision of the robot to cross check the impartiality and independence of the arbitrator. This is indicative of the utility of the machine arbitration for different purposes i.e., to render an award, or to correct or cross check the outcome of the human arbitrator.
Lastly, the seminal text on international commercial arbitration is undeniably the New York Convention of 1958. At the time the Convention was drafted, the possibility of AI in arbitration would not have been even envisioned by the drafters. Indeed, even now, the full scope of AI into arbitration cannot be envisaged. However, the next decade will inevitably witness a change in the way arbitration once operated and the rise of AI demands for its regulation, especially in arbitration. It is suggested that arbitral institutions and states alike draft guidelines and rules for regulation of AI systems. UNCITRAL could serve as the forerunner of the same, with the introduction of a new protocol, or as a schedule within the existing Model Law. The draft ought to set out the definition of AI in arbitration, list accredited and certified algorithms, and address problems of accountability and consequences. A useful starting point may be the European Union’s High-Level Group Expert Report published in 2019 on guidelines for “Trustworthy AI”. According to the Guidelines, “Trustworthy” AI should be: “1) Lawful—respecting all applicable laws and regulations; 2) Ethical—respecting ethical principles and values; and 3) Robust—both from a technical perspective while taking into account its social environment.” Guideline (1) is especially relevant for arbitration. It is of utmost importance to the enforceability of the arbitral award that it is rooted in the mandatory laws of the place where the award is rendered and the place where the enforcement of the award is sought. As a result, the arbitral community must therefore ensure that the AI systems being used do not conflict with mandatory laws. The upcoming years will be a testament to legislative restraints and different approaches by different states on the regulation of such. This may even result in a different sort of regulatory arbitrage, where more relaxed rules on AI may increase in popularity of parties choosing the state as their seat of arbitration.
The Feasibility of AI in Pakistan
The Recognition and Enforcement (Arbitration Agreements and Foreign Arbitration Awards) Act 2011 (“REA”) is the current legislative scheme in Pakistan with regards to the incorporation of the New York Convention into the domestic legal sphere. REA itself is sparse and limited, with ten sections that mainly aim to set out the fundamental powers of the courts in relation to arbitral proceedings, and lay down grounds relating to recognition and enforcement of a foreign award, whereas the Schedule to this Act fully incorporates the New York Convention into its domestic regime. Therefore, while it is uncertain how AI would fit into the existing legal framework of Pakistan. Perhaps cue can be taken from the New York Convention itself, which, by virtue of being attached as a Schedule with REA, directly forms part of Pakistani law. As stated above, the New York Convention was brought into force more than sixty years ago when at that time, the use of AI could not possibly have been envisaged. Resultantly, where the formalities of the Convention have been complied with, it seems unlikely that the award would be rendered unenforceable under Pakistani law due to the involvement of AI in the arbitral proceedings in any form.
However, an interesting situation may arise where the Arbitration and Conciliation Bill—which was brought into the Senate of Pakistan as the “Arbitration and Conciliation Act 2015”—is brought into force. The Bill itself is loosely premised on UNCITRAL Model Law on International Commercial Arbitration 2006. UNCITRAL operates as a framework and guideline for states who wish to adopt arbitration friendly legislation into their own national law. Throughout the Bill, the ‘nationality’ of an arbitrator is referred to, one example being Section 11 of the Bill, which states that a person of “any nationality” may be appointed as an arbitrator. One’s interest is indeed piqued when considering whether or not an autonomous “robot” arbitrator would be of a single nationality, or whether it would have any nationality at all, considering the fact that it is not a natural person. Therefore, were this Bill to come into force, there may be implications for the future of the arbitration regime in Pakistan, with the rise of AI in the legal sphere globally. It must be submitted, however, that the Bill itself was introduced in 2016, and five years down the line it is yet to be passed. As a result, REA remains the operative legislative regime regarding international commercial arbitration, which does not seem to explicitly accept, reject or even envision the possibility of AI in arbitration in Pakistan.
Conclusion
These are crucial times for the evolution of arbitration from a conservative sphere into one that is in line with today’s legal world. Recently developments in the field have been slow; however, the advent of Covid-19, and with it, the new wave of technology in the legal sector will ensure that arbitration catches up. The use of AI in arbitration would make the process smoother, speedier, and arguably more just. It is, however, necessary that the use of AI be phased in gradually and incrementally. Therefore, the implications need to be dealt with in order for the process to be streamlined and regulated. It remains the duty of the states to ensure that they are prepared to face the upcoming challenges and come up with adequate responses to that effect.
Furthermore, as far as the outcomes of AI assisted arbitration is concerned, the determinants of its success would ultimately be the quality of data input and the accuracy of the algorithm itself. Therefore, attention is merited for the quality control of the data input and an examination of the construct of the algorithms.
Lastly, it must be cautioned that the introduction of these new technological processes does not aim to devalue or undermine the conventional nature of the arbitral process but rather, aims to supplement it by enhancing its capabilities with the appropriate use of technology. AI in arbitration is coming into itself as an essential assistive tool, which has tremendous potential for further development and application.
Appendix
The ‘AI in Legal Services Summit was held in London on 4th June 2019. It covered a variety of debates ranging from the ethics of AI to the use of AI in Law to the impact on a lawyer’s job with the introduction of AI. The speakers were from all over the world. More information can be found here on their website:
<http://www.cityandfinancialconferences.com/events/the-artificial-intelligence-in-legal-services-summit/event-summary-a2f772e53254474ea5d85a621086a4f8.aspx>
List of Speakers (in alphabetical order of last name)
• Sir William Blair, Professor of Financial Law and Ethics at the Centre for Commercial Law Studies, QMUL and Associate Member at 3VB, Keynote Panel I ‘Algorithms in the justice System – current state of play and the future outlook’.
• Duncan C. Card, Senior Partner, Co-Chair Technology Law Practice, Bennett Jones LLP, Keynote Panel II ‘Wider societal impacts of Al and its future.’
• The Rt Hon. David Gauke MP, Lord Chancellor and Secretary of State for Justice.
• Noel Hurly, VP Strategy, AR, Interactive debate ‘The ethics of algorithms’.
• Dr Hannah Knox, Associate Professor of Anthropology, UCL, Keynote Panel II ‘Wider societal impacts of AI and its future’.
• Charlie Morgan, Senior Associate Dispute Resolution at Herbert Smith Freehills, Keynote panel I ‘Algorithms in the justice System – current state of play and the future outlook’.
• Daniel B. Rodriguez, Professor of Law, and Former Dean, Northwestern Pritzker School of Law, Keynote Panel III ‘Future skills and job requirements for the professions with a focus on law, banking and accountancy’.
• Professor Burkhard Schafer, Professor of Computational Legal Theory, The University of Edinburgh, Keynote Panel I ‘Algorithms in the justice System – current state of play and the future outlook’