Submissions to Justice Mitchell, Supreme Court of Western Australia dated 4 October 2025

I. Introduction

1. This submission responds to the Supreme Court of Western Australia’s Consultation Note on a proposed AI Practice Direction (2025). We commend the Court for its leadership in initiating consultation on this important issue. This submission sets out the national baseline on generative AI use in courts, answers the consultation questions posed, and explains what can be done differently. It endorses the Law Council of Australia’s balanced approach while adding the perspective that architecture and methodology matter: AI can be designed to reflect law’s discipline, not just probability.

II. National Baseline Understanding

2. Across jurisdictions, courts have converged on a baseline understanding of generative AI:
   
   1. LLMs are probabilistic text generators that predict the next word.
   2. They do not reason in a human or legal sense.
   3. They are prone to hallucinations (non-existent cases).
   4. Their processes are opaque (no audit trail).
   5. They conflate fact, inference, and opinion.
   6. Human verification of all citations is essential.