Accurate, Focused Research on Law, Technology and Knowledge Discovery Since 2002

The Burden of Proof in Legal AI

Adam Ziegler via LinkedIn: “My default posture when learning about new tech, especially legal tech, is “open-minded skepticism.” I’m open to persuasion – even eager to be persuaded – that a machine can do what it’s builder claims. But persuasion requires proof. Legal tech tends to be long on claims and short on proof. Part of what drives this is a bet made by some of those selling products that their target customers won’t demand proof. That claims won’t be challenged or explored. That half-baked promotional content and the blessings of industry gatekeepers will suffice. The negative reaction to the recent pre-print of a study by researchers at Stanford’s RegLab – “Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools” – exemplifies this problem. While product marketing claims were met with little scrutiny and a degree of genuflection, transparent scholarly efforts to test those claims were subject to condemnation. This eagerness to embrace PR while resisting analysis is especially dangerous in the legal world, where errors are consequential and undetected inaccuracy could mean liability. In this world, the opposite should be true: we should expect verifiable proof from those promising reliability. The burden of proof should be on the ones making the claims…”

Sorry, comments are closed for this post.