Begin with the task
An LCIA-led submission by 16 arbitration and dispute-resolution organisations addressed the European Commission’s draft guidance on the classification of high-risk AI systems. The BITS post on that submission focused on a distinction between ancillary organisational work and assistance with assessing evidence or reaching a judgment.
The submission is a contribution to a regulatory discussion. It is not, by itself, a court decision or a binding determination that every tool used in an arbitration falls into a particular category.
For a reader, that distinction suggests a better first question than “Does the tribunal use AI?” Ask what the tool does and how its output enters the decision-making process.
A summary can become a filter
Consider a hypothetical tribunal reviewing a large document collection. A tool produces a concise account of the chronology. That may save time, but the account also selects what appears important and what can be left out.
If the tribunal treats the summary as the record, rather than a guide back to the record, an organisational aid can begin to influence its findings. The risk is not limited to an obviously wrong answer. An accurate but incomplete account can still change the apparent strength of a case.
This is an editorial example, not a reported case. It helps isolate the practical issue: the degree of reliance on the output and the opportunity to check it.
Questions to put into a procedural discussion
Parties and tribunals can make the discussion more useful by naming the proposed task, the information involved and the intended level of reliance. Will confidential documents leave an approved environment? Will the output be used to find material, or to evaluate it? Who checks the result against the original source?
Where assistance could affect substantive reasoning, fairness also comes into view. Parties may need an appropriate opportunity to address material or reasoning that influences the decision. The precise obligations depend on the applicable law, rules and circumstances.
An agreed approach should be capable of being followed in practice. A broad promise to use tools “responsibly” does not tell a team how to handle confidential records or an arbitrator how to verify a generated proposition.
Keep responsibility with the people appointed to decide
Our view is that the quality of the procedure matters as much as the sophistication of the software. Verification, confidentiality and control over reasoning need to be built into the actual workflow.
That does not require treating every digital convenience as a threat to an award. It requires enough precision to distinguish an appointment reminder from an assessment of disputed evidence.
The original BITS post and the linked LCIA submission are below. Read the submission for the organisations’ specific requests; the practical questions above are BITS commentary on the wider issue.
