I. Introduction

With the Covid-19 pandemic forcing all sectors to adopt flexible working styles, there has never been a time more significant than 2020 to understand the wonders that technology has to offer. Technology has kept businesses, offices, and all other entities up and running through the year. This applies to the arbitration sector as well. Worldwide lockdowns forced mostly all international arbitral institutions to adhere to remote forms of dispute resolution raising numerous questions as to the risks and benefits of the system.

Taking this discussion up a notch, there are sections that call for a gradual replacement of the human factor in arbitrations to be replaced by a solely machine led regime. The desired result being a more efficient, well-calibrated, speedy system of arbitration devoid of biases and prejudices that very often are said to plague in-person arbitrations. In the present article, the author seeks to test the effectiveness of an Artificial Intelligence ("AI") led arbitration regime against the traditional one in a bid to test the effectiveness of such a system against a traditional one.

AI, is the science of making a computer smart, in a way that the machine is able to imitate human intelligence while performing tasks assigned to it.1

It is generally argued that AI may be able to play quite a significant role in arbitrations. AI systems are believed to be more effective and exact in nature. For instance, in a 2016 study of the predictability of European Court of Human Right's ("ECHR") decisions by an AI system, it was found to be 79% accurate.2 In a similar exercise involving decisions of the United States Supreme Court, the system was found to be 70.2% accurate.3 So, does this mean that an AI based system with maybe a little more calibration and feeding of appropriate data sets can be as effective as a human arbitrator?

III. Analysis of an AI-led arbitration system

An AI based arbitration may prove quite instrumental in providing assistance in a traditional arbitration system. However, there may arise certain key concerns.

Replacement of Human Counsels by AI Counsels

A counsel in an arbitration has a variety of important tasks to perform, some of which include the preparation of statements of claim and defence, filing and preparing necessary applications, arguing before the tribunal and cross-examining witnesses. A much-debated question that arises while discussing AI based arbitrations is whether the system shall do away with the involvement of humans in the arbitration sector. While we shall come to that question later, a related but intriguing question that arises herein is whether human counsels in an arbitration may be replaced by an AI counsel. While no such technology has been developed as of yet, some key functions that an AI counsel should be able to perform shall include the ability to process voluminous documents and files, acquiring know-how of the law and procedure in the sector under which the dispute has arisen, necessary legal acumen to connect the facts with the legal position in order to defend the client, and must also possess the ability to argue and cross-examine witnesses.

Analysing the above stated functions, while an AI counsel would be much more efficient in analysing and reading voluminous documents, and acquiring know-how of the law and procedure, the system's ability to perform the rest of the functions remains doubtful. The primary reason behind this is the system's lack of creativity, skill and legal acumen that is essential for connecting the facts to the law in question and present these in a manner that is convincing enough for a tribunal. This skill is what legal counsels take years to develop and it is hard to say if any technology can be capable of imitating the intricate functioning of a human mind so closely.

However, the need for AI counsels is eliminated in a fully AI based arbitration.

AI Arbitrator

The first stage in any arbitration is the appointment of arbitrator(s) by the parties. An arbitration where the human arbitrator is replaced by an AI arbitrator may prove both advantageous and disadvantageous in some respects. In several disputes, parties spend a lot of valuable time going back and forth in appointing an arbitrator. An AI arbitration would save parties a lot of time and hassle in this respect. However, there are several disadvantages that a party may have to face due to a machine being the arbitrator.

The first and primary disadvantage is with respect to the decision-making power of an AI arbitrator. The second stage of an arbitration is the submission of claims and defence by both sides. The question that arises is whether an AI arbitrator would be smart enough to process both sides of the story, connect the factual position to the law, and then render a decision. This is a pertinent question because a machine, even an AI machine is only as smart as the data that has been fed into the system. For instance, in the 2016 ECHR experiment stated above, the data fed into the system involved factual positions written by the judges themselves, thus making the data inherently biased. This could have been the reason for the accuracy of the system.

Statements of claims and defences are inherently biased to each side. To render a proper decision, the AI arbitrator would have to be trained to overlook the biased submissions and connect the factual position to the law stated by parties in their submissions. However, let us assume at this stage that an AI is capable of doing so.

Decision making capabilities in an arbitration cannot only be limited to the technical questions because arbitrations may involve several issues that would require an arbitrator (whether human or AI) to use individual judgment and skill, for instance, questions of bona fides, or the capacity of a party to pay, or the ability to judge the facial expressions and body language of a witness during cross examination would require objective judgment and assessment of the facts, circumstances and surrounding issues.

This brings us to the third stage, i.e., production of documents and taking evidence. Parties may cooperate with each other so as to feed all the relevant documents and evidence into the system and the AI arbitrator may be able to deliver a decision on the basis of these. However, a situation may arise, where one party files an application for discovery of documents, or seeks an order for adverse inference against the other party. The difficulty that an AI system might face with such narrow issues is the lack of adequate data in arbitrations. Awards being mostly confidential, finding adequate number of datasets for each issue that may arise in an arbitration would be a humongous task. Given that most arbitrations are quite complex in nature, and involve varied issues, narrowing down on issues for which datasets would have to be fed into the system would itself be a task quite beyond imagination.4

The fourth sub-stage, i.e., affording the right of oral hearing to parties is quite problematic as well since an AI system at present is not advanced enough to hear parties in the sense that a human arbitrator does and analyse these oral arguments in arriving at a decision. This would render an AI system quite ineffective in cases where a document-based arbitration would not be effective in covering the entire spectrum of the dispute.

While the above discussion was only on the functional aspects of an arbitration, an AI arbitrator might not be legally feasible in several jurisdictions that may have requisites such as 'nationality' which is only applicable to natural persons at present. For an AI system to successfully function as an arbitrator, there has to be some form of legal status that needs to be afforded to such a system.5 Presently, no jurisdiction affords such status.

Due Process Concerns

While an arbitration is not guided by a rigid set of rules, it is guided by the rules of natural justice, equity and conscience, collectively known as due process. Several pertinent questions arise in an AI based arbitration. A possibility may arise where parties may assail the arbitral awards in instances where one party disputes the use of AI in the arbitration, or where one party or appointed counsel has the ability to access AI systems, while the other party does not. These are some situations that may need consideration.

Arbitral Award

The Award stage is when in a traditional arbitration, an arbitrator exercises his mind to analyse the factual position against the legal background. This is precisely what an AI led arbitration regime would do as well. However, there are two significant difficulties that an AI arbitration might face. Firstly, in a traditional arbitration, appointed arbitrators have years of experience leading to development of the requisite expertise and skills. An AI system does not possess that skill. And this might be crucial in an arbitration where understanding of the commercial impact of the award is necessary. Secondly, it is a requisite in various jurisdictions that arbitral awards carry proper reasoning to explain how and why the arbitrator arrived at a particular decision. An AI system would fail here as reasoning and explaining the reasons constitute a delicate human process and the system would not be able to achieve that. Further, several jurisdictions require arbitral awards to be signed or in writing, thereby excluding an AI arbitration.

Other key concerns

  1. Complexity of the dispute: AI arbitration may not be preferrable for disputes involving complicated issues of law and facts with larger stakes.
  2. Enforceability of Awards: AI Awards may not be enforceable in jurisdictions that require arbitral awards to subscribe to a particular form.
  3. Right to Appeal: It is hard to say whether an AI based arbitration would be able to afford the right of appeal to parties. In a traditional arbitration, an appeal is addressed to a higher authority. There being no such hierarchical categorization in an AI system, it seems difficult to have an AI based appeal scenario.

IV. AI based Arbitration under the Arbitration & Conciliation Act, 1996 ("Arbitration Act")

In its present form the Arbitration Act does not support an AI based arbitration regime due to the following reasons:

  • The term 'arbitral tribunal' under Section 2(1)(d) of the Arbitration Act has been defined as a sole arbitrator or a panel of arbitrators. However, under Section 11 (Appointment of Arbitrators), the terminology used such as 'nationality' can only be held to be applicable in the case of a natural person till the time AI and other computer systems are afforded legal status.
  • Under Section 18 (Equal Treatment of Parties) the arbitral tribunal may have the power to disallow a party from using AI where the other party disputes the same.
  • Coming to the stage of the award, Section 31 (Form and contents of arbitral award) requires the award to be in writing and signed by the members of the arbitral tribunal.
  • Sub-clause (3) of Section 31 mandates the arbitral award to state the reasons for the same.

V. Concluding Remarks

In every discussion that involves AI, a persistent question that occurs is whether AI will replace humans in the arbitration sector. And the answer to that is a solid, resounding no. At least AI in its present form is not advanced enough to be able to capture the arbitration sector. While involvement of AI in arbitrations may gradually be increased so as to provide assistance to humans, a complete AI driven arbitration may not be suitable for all types of commercial disputes due to the sheer complexity of some of these. Even if arbitrations were to be subsumed by AI, parties may find it difficult to bear the cost of expensive AI systems. This is where arbitral institutions may be able to play a significant role in bridging the gap between AI and parties by providing affordable AI services. In essence, it can be said that AI is still at a developing stage, and if sought to be utilised in its present form shall create more troubles than it solves for the arbitration sector.

Footnotes

1 Professor John McCarthy, 'What is AI/ Basic Questions', http://jmc.stanford.edu/artificial-intelligence/whatis-ai/index.html.

2 UCL, 'AI predicts outcomes of human rights trials', (October 24, 2016), https://www.ucl.ac.uk/news/2016/oct/ai-predicts-outcomes-human-rights-trials.

3 Communications of the ACM, 'Artificial Intelligence prevails at predicting Supreme Court decisions', (May 5, 2017), https://cacm.acm.org/news/216852-artificial-intelligence-prevails-at-predicting-supreme-courtdecisions/fulltext#:~:text=Artificial%20Intelligence%20Prevails%20at%20Predicting%20Supreme%20Court%20 Decisions,-By%20Science&text=The%20algorithm%20correctly%20forecast%2070.2,about%2066%25%20right%20in%20comparison.

4 In the case of Pyrrho Investments Ltd. v. MWB Property Ltd.4 , a United Kingdom court for the first time allowed e-discovery using predictive coding. However, predictive coding in the case was more synonymous to a review and sorting of documents by relevance rather than a traditional discovery process.

5 Horst Eidenmüller and Faidon Varesis, 'What is an Arbitration? Artificial Intelligence and the Vanishing Human Arbitrator' https://poseidon01.ssrn.com/delivery.php?ID=091071099092098099098104072093101068104013035023062 02101003111208802510212606501409309703010210604601611411608102308609111702612603408907801 21180830861021181260270480140020031220660310191240240250830040290010210261040100050720750 21085072031096112127&EXT=pdf&INDEX=TRUE

- 08 January 2021

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.