In the latest International Arbitration Report from Mealey’s, the publication poses the question: “Have AI tools contributed to efficiency in international arbitration, such as organization of discovery materials, review of expert witness reports or summaries of awards?” Haynes Boone attorneys Fiona Cain and Jack Spence provided a response in the article excerpted below:
Cain and Spence: As with every tool, AI has applications where it excels and also those where its limitations become very apparent.
The most established and successful use case (at least for now) is in document review, where tools on platforms like Disco and Relativity can train on a review corpus and a human reviewer’s decisions. The resulting custom model is then used to prioritise, showing the documents which are the most likely to be disclosable to a human reviewer for them to confirm the model’s results. This can greatly cut down the documents which are subject to human review (and therefore on both time and costs) in a manner which is defensible to, and normally accepted by, tribunals.
A developing field, where we have found increasing utility, is in running natural queries over large sets of documents. Uploading submissions, exhibits (and potentially disclosure) to a platform such as Harvey allows lawyers to make natural language queries which return more relevant extracts than traditional search tools would. Internally we have been fairly impressed by this functionality — although the outcome will depend on whether the questions are asked in a way that is understood by the platform and the results, as always, need to be double checked by a human lawyer!
By contrast, one area where the reality of AI is not living up to the hype (and much of the advertising from vendors!) is in legal research and the drafting of submissions. Large-language models can generate fluent prose, but they very frequently “hallucinate” sources — citing cases, statutes or procedural rules that do not exist. A growing number of judgments from around the world (although predominantly in the United States) have highlighted the issues that arise when lawyers fail to verify these citations. It seems likely that this practice is also found in international arbitration but shielded from view due to the confidentiality of the proceedings. To address this, guidelines on the use of AI in arbitration are being issued, including those by the American Arbitration Association and the Chartered Institute of Arbitrators.
Equally, when asked to reason, the model often defaults to arguments rooted in abstract notions of fairness or commercial common sense. While such considerations may resonate, they are rarely sufficient (at least in English-law governed proceedings, which still principally depend on precedent, statute and orthodox interpretative principles).
Read the full article from Mealey’s here.
