I see arbitration’s greatest current challenge as what I will call, for want of a catchier term, the ongoing “scientification” of arbitration. Modern arbitration developed in the context of conventional commercial disputes, and while industry understanding was often seen as desirable in an arbitrator, deep technical understanding usually was not—ultimately, that is why there is expert testimony. But as science continues to penetrate every aspect of our lives, it also increasingly shapes the disputes that come before arbitrators. 

The most obvious way this is happening in 2026, of course, is artificial intelligence (AI). But while every active arbitrator is now aware that large language models (LLMs) can generate fictitious citations, I would argue that the more fundamental challenge for our field is the attractions of that technology for a profession (arbitrators) that has always been focused on mastering procedure and legal interpretation, rather than on the complexities of modern science.

This is because the risks of AI are subtle, and understanding them properly requires an engagement with the underlying technology of LLMs that few people without a passionate interest in science and computing are likely to prioritize. The practical result of that limitation in a field focused on procedural and legal expertise is that most arbitrators are just not incentivised to develop a strong understanding of how LLMs are trained and how this affects their outputs, let alone what a “transformer” is and the fundamental impact transformers have on the way that LLMs function.

Full article below: