This post was submitted by myself and Pauline McBride as a letter to the Editor in response to Richard Susskind’s article, ‘I asked ChatGPT to write some laws – this is what happened’ (The Times, 4 April 2024) (non-paywall version of the article available here).
Since it wasn’t published, we’re reproducing it here. This represents our own views only.
Sir,
Richard Susskind makes various claims (Apr 4) about the capacity and desirability of using AI to draft and enforce the law. His article fails to consider the very real practical limitations of such technologies (so-called ‘hallucinations’ are an intrinsic part of how generative AI systems, such as ChatGPT, work), or the significant issues around how they are designed and built (for example the relevance of the sources of data on which they are trained, and how and by whom those data are gathered and prepared).
He also fails to mention the significant scholarly and practical concerns about the use of AI technologies in the creation and operation of law. Laws ought not to be directly articulated in computer code for enforcement purposes; law works on the basis of the interpretability of natural language rules. Removing this crucial element in favour of immediate automated application means undermining an essential element of what makes our legal system democratic. Susskind aspires to law-as-code which precludes non-compliance except in case of emergencies. This too is undemocratic; it removes agency from citizens and rules out competing interpretations of the law (it’s also far from clear how what counts as an ‘emergency’ can be represented in the machine). Laws that cannot be broken are not laws, they are brute force commands. We have seen, too, with the Post Office Horizon and Volkswagen ‘defeat device’ scandals, just what can happen when we assume that computer code is operating as expected.
Susskind’s claims about the use of AI to assist legislative drafters and to explain legislation to different audiences have some merit, but this is a world away from the picture that he paints of automated enforced compliance. The Rule of Law is as much about having the capacity to interpret the law before its application as it is about every one of us being subject to the same set of rules. To embed automated enforcement within the very concept of the Rule of Law, as Susskind suggests, is to gut the Rule of Law of its democratic character. Such a dystopian vision should be vigorously resisted.
Dr. Laurence Diver, former senior researcher, Counting as a Human Being in the Era of Computational Law, Vrije Universiteit Brussel
Dr. Pauline McBride, Research Fellow, Northumbria University
