AI use in IA: Potential use and misuse

Introduction

The potential applications of generative artificial intelligence (AI) in disputes has attracted the interest of stakeholders across the legal sector. The conduct of international arbitration, with its focus on party autonomy and procedural flexibility, will increasingly need to grapple with the use (and misuse) of generative AI. In this article, we examine three areas that are likely to be impacted.

Arbitrator selection

Arbitrator selection is a critical issue for parties and institutions. The search process usually involves preparing a shortlist of candidates from pre-existing databases and contacts/ recommendations. There is a natural tendency for the focus to be on the more experienced and high-profile candidates.

Generative AI tools are likely to add another dimension to the arbitrator selection process. In particular, machine learning tools can be used to conduct analysis of potential candidates and their inclinations toward certain legal theories in greater depth, at a lower cost and more efficiently. For example, AI could be used to create reports both from existing arbitrator databases and ‘unstructured’ data sources which can currently only be analyzed through manual review by experienced lawyers. This would include biographical information on the internet, academic articles and (in the case of judges) judicial decisions.

International arbitration has established networks which often may influence the visibility and ultimate appointment of arbitrators. The use of AI tools in the arbitrator selection process may also be helpful in expanding and diversifying the pool of candidate arbitrators, by highlighting candidates who may not be favored by a selection process driven by existing databases, experience and word of mouth.

However, any use of generative AI in arbitrator selection underlines the need for it to be done in a way which is impactful and accurate. It is critical that source data is up to date and of sufficient quality and size. It is also important that machine learning algorithms are trained with an appropriate selection system, due to the risk of potential bias resulting from the way AI is programmed. Indeed, if the algorithm is not programmed with potential bias in mind, it risks reinforcing existing tendencies in arbitrator selection, exacerbating underrepresentation and wrongly maintaining or even exacerbating the existing high barriers to entry for prospective arbitrators.

Document production and the evidence process

Generative AI is likely to have a significant impact on evidence and disclosure.

Document production in international arbitration has become increasingly complex and contentious over recent decades; it is often the most expensive and intrusive step in the proceedings. This, along with it being common to adopt witness statement/ expert report processes akin to common law litigation, has significantly contributed to the perception that the time and cost of arbitration can be disproportionate and wasteful. Generative AI tools have the potential to ease some of the pressures and potentially allow for a more efficient process.

In relation to production requests, this could include AI-powered assistance in preparing requests (such as by reference to analysis of the pleadings), the possible grounds for objections and also helping the Tribunal resolve disputes, particularly for long and complex schedules of requests to produce. AI could also help bridge the gap between requests and the processes employed to search for, identify and filter documents. For example, next generation e-disclosure tools can conduct searches based on ‘natural language’ queries (that is, more akin to actual document requests, rather than the keyword searches and Boolean operators usually used to construct searches in document review databases).

There are many potential AI uses in relation to witness statements, expert reports and the hearing bundle. For example, preparing chronologies from the parties’ written submissions, summarizing witness statements and expert reports, identifying avenues for cross-examination and easing the process of creating the hearing bundle.

Translation is another key area given the multilingual nature of international arbitration and its impact on time and cost, both in relation to translation of documents and AI-assisted live translation of witness testimony.

While there is great potential for use of generative AI in document production and evidence, there are also reasons to be cautious, which include:

• The current lack of procedural definition around the acceptable use of generative AI. Few arbitral rules currently address the use of AI (and the IBA Rules on the Taking of Evidence in International Arbitration do not currently address it).
• The implications for arbitral confidentiality – a key advantage of arbitration for most parties – need to be properly understood and safeguards need to be instituted.
• The risk of sophisticated AI forgeries, particularly given tribunals’ information gathering powers tend to be narrower than courts. Nigeria v P&ID shows the significant challenges that can arise for tribunals to identify traditional forgeries, let alone AI-generated ‘deep fakes.’ This may mean tribunals requires a greater degree of technical assistance in relation to evidence.
• Loss of human oversight. In a post-COVID world, virtual arbitration hearings are becoming more common. An overreliance on AI may continue to erode the human element of arbitration. While AI could improve efficiencies, arbitration needs to remain flexible and responsive to the parties’ needs, and not become mechanical.

Generative AI is likely to have a significant impact on evidence and disclosure

Challenges to arbitral decisions

A key differentiator between litigation and arbitration is that arbitral decisions generally cannot be appealed on the merits. However, arbitration is no different to litigation in the sense that large sums of money turn on the outcome of decisions and therefore the potential upside to challenging decisions is considerable. At least two issues potentially arise in relation to AI.

First, the possibility of broadening the grounds to challenge an award. Failure to follow due process is one of the most common grounds to challenge arbitral decisions in national courts. By definition, where parties consent to the use of AI in the proceedings, it should be difficult to successfully challenge an award based on procedural irregularity arising from its use. However, for the party keen to delay or avoid an award, in this nascent stage of the use of AI, finding instances of non-compliant use of AI could be easy – for example, AI being used in a manner outside the scope of a procedural order, in a way that was not disclosed, in a way that generated some inaccuracy hitherto unidentified, and other scenarios. Parties could potentially mount public policy arguments too – for example, the use of AI that is not permitted under relevant national law or even arguments based on technological disparities.

Second, there is a possibility of more frequent challenges to awards. For the losing party in an arbitration, attempting to challenge an award is often not attractive on the grounds of cost – that is, throwing good money after bad. However, to the extent that generative AI can lower the barriers to challenging an award, namely by employing AI tools that can sift through the award, hearing transcripts and generate arguments at relatively low cost, then it could result in a higher number of challenges to awards – resulting in delays to the finality of disputes.

Conclusion

The potential use of generative AI in international arbitration is wide-ranging and has the potential for increasing efficiencies in arbitration. Parties, tribunals and institutions will be increasingly grappling with these issues over the coming years.

The authors would like to thank Mariana Plaza Cardenas for her assistance in the preparation of this article.