DOES ‘MOVE 37’ HINT AT THE FUTURE OF AI IN LITIGATION AND ARBITRATION?
In 2016, AlphaGo, an artificial intelligence (AI) system created by Google’s DeepMind, beat the 18-time and reigning world champion in the incredibly complex game of Go. While an AI system beating the world’s best player was a big story, some experts thought the real story was AlphaGo’s ‘Move 37’, initially dismissed by experts as a blunder, but later recognised as a novel and strategically brilliant move.
One expert said it was a move that never would have been made by a human. After the match, AlphaGo was asked the odds of a Go expert making that move. The answer: 1 in 10,000. Since AlphaGo, other specialised AI models have predicted the 3D structures of hundreds of millions of proteins, helped design new treatments for certain types of cancer, and extended our understanding of plasma physics. Move 37 has become a shorthand expression of AI’s capacity for creativity.
It is notable that these AI models were not based on large language models (LLM), which are the dominant AI systems used in the practice of law today. AlphaGo and other scientifically creative models tend to be small, specialised models, trained for a narrow area of expertise, and not based on text. Given the differences, Move 37 would seem to be an inapt analogy for AI’s use in the legal field. But it may be rash to dismiss it too quickly. Coincident with the broad availability of generative AI (genAI), AI has been used for a few years in the discrete legal areas of contract analytics, M&A due diligence and regulatory compliance, but only recently have AI models begun to be applied more generally in litigation and arbitration. LLMs are very good at what they do and provide significant benefits to users, but they also suffer from important limitations.
Jan-Mar 2026 issue
King & Spalding, LLP