Artificial Intelligence in ADR: A Neutral’s Evolving Perspective

Back in April, I spoke on a panel about how JAMS arbitrators are navigating the burgeoning world of artificial intelligence (AI). It was a lively exchange, showcasing the diverse ways my colleagues are integrating AI into their practices. One panelist, a self-professed coder, enthusiastically detailed his deep dive into AI, crafting intricate scripts to integrate various models for highly specific tasks. Another shared his positive experience with Westlaw's AI features, highlighting how they had streamlined his award writing, making him a more efficient and effective neutral.

A Healthy Dose of Skepticism

As the final speaker, I took a more cautious stance. My perspective then was that current AI offerings, while brimming with potential, weren't quite ready for the nuanced demands of our industry. I pointed to the steep learning curve and the observation that many applications weren't tailored to the specific workflows of arbitrators. While acknowledging the existence of sophisticated legal research tools and AI-powered contract drafters, I noted that I had not come across a comprehensive solution capable of synthesizing briefs, exhibits, transcripts and an arbitrator's personal notes, and then drafting a cohesive award. Moreover, experienced neutrals have well-established workflows, and the prevailing AI tools didn't seamlessly integrate into these existing practices. In some, I suggested that the AI-cautious audience wait until the market matures.

The Reluctant Adopter

Fast forward six months, and I find myself cautiously yet optimistically reengaging with the world of AI. For a considerable period, I had been what one might affectionately call an "AI resister." My reluctance stemmed from three core concerns. First, initial results often felt unreliable and lacked my distinct voice, which created more work. Second, I harbored a deep distrust of AI-generated summaries; I firmly believe in the importance of engaging directly with primary sources, processing information independently and then formulating my own conclusions. I didn't and still don't believe that one can master the facts and issues just by reading AI summaries. Third, I faulted AI for not getting the nuance of my terse questions, and the premise that everyone has to become a "prompt engineer" in order to extract precisely what we need from the technology.

A Turning Point

It took a candid, and perhaps overdue, conversation with my wife to illuminate the root of my resistance. As a program manager at Google tasked with increasing the internal adoption of Google's AI tools, she possesses a unique perspective on the technology's potential and pitfalls. She gently but firmly pointed out that my pithy requests, or “zero-shot prompting” in AI lingo, were, to put it mildly, destined to fail. I was providing the AI with no context, no references, no training data reflecting my unique writing style or thought processes, so how could I reasonably expect it to "speak in my voice"? Furthermore, I showed no patience to provide consistent feedback, which is crucial for AI models to iterate and improve. Communication, or rather the lack thereof, was the core issue. Yet, as she reassured me, I am someone who is open to learning and adapting. And I am using the paid version of an AI model whose Terms of Use expressly state that my prompts and uploads remain confidential and are not used as training data.

Mentoring the Machine

So, I am now embarking on a new journey: actively teaching AI my specific workflows and the nuances of my writing style. In turn, I hope that this iterative process will show me how to leverage AI more effectively, ultimately helping me to refine my practice and become an even better neutral. It's a familiar feeling, reminiscent of my Biglaw days when I was responsible for training junior associates. It appears I'm back in the role of a mentor, this time with a silicon-based student. And in case you’re wondering, I wrote the first draft of this myself, but AI edited it, and then I made further revisions.