Presented at Gikii 2019, Queen Mary University London, September 2019.

Abstract

“What made it possible for me to exercise freedom of choice also made it impossible for me to know the future.”

[spoiler warning]
Ted Chiang’s 1998 novella Story of Your Life beautifully weaves together themes around the philosophy of language, determinism, and the nature of loss and free will. As the story’s protagonist learns the language of an alien race visiting Earth, she begins to develop their ‘simultaneous’ consciousness, breaking out of the sequentially-bound limits of her ordinary human consciousness. The result is a concurrent awareness of past, present and future; of the a priori completeness of our stories.
[/spoiler warning]

How does this connect to law? Oliver Wendell Holmes famously argued in The Path of the Law that the practice of law is, in essence, the art of predicting what courts will do – it’s telling the legal future on the basis of an (informed) understanding of the legal past, and it’s something lawyers do on a daily basis. Law is inescapably bound up in language and text – the stability of legal institutions and the possibility of contest are intimately connected with printed expressions of past decisions, agreements, and rules.

But what happens when we rely on legal analytics systems to make predictions of judicial decisions on our behalf? Natural language processing (NLP) is already being used for this purpose in the US and, more recently, to predict cases in the European Court of Human Rights. The predictions of such systems are necessarily based on the corpus of existing law, inductively deriving the ‘future’ from the past.

What might it mean for legal practice if future legal arguments are recommended by such systems, the outputs of which in turn become the inputs of judicial reasoning? In adversarial legal systems, the judge cannot look beyond the evidence and arguments that are presented to her, implying that the resulting judgments will become mediated by such predictions. As subsequent written judgments are added to the training data for the prediction model, we could see a narrowing of argumentation styles as the breadth of predicted – and thus recommended – approaches increasingly filters out the kinds of leftfield argument that a talented (or simply conscientious) lawyer might have successfully argued in favour of her client. With the drive to reduce litigation costs while simultaneously increasing success, the danger is that practitioners cease to do the kind of research that leads to such outcomes, instead opting for the safe predictive efficiency of the machine.

Thus, the is of past legal text becomes the ought of future legal text, potentially ignoring relevant authorities or future social changes in a seductively efficient race to the bottom. The sequential flexibility of the law, adapting and responding to the society it serves, might therefore reach an inflection point, beyond which it starts to become ‘fixed’: legal outcomes become increasingly predetermined, and the law’s ‘consciousness’, embodied in legal text, becomes ‘simultaneous’, its development increasingly pre-ordained in ways incompatible with a democratic society.

You can view a recording here.