Immanuel Kant and amoral AI judges

Blog post only

Outsiders to think that tax law, like any law, is rule based work and therefore a sector where AI could do very well. Legal specialists know that this misrepresents a lot of the work. Yes, law is rule based, but there is a certain finesse attached to it, that is very non-mechanical. As I described before (https://taxleonard.wordpress.com/2023/06/21/machines-or-back-to-humans/), application of tax law is a judgement area, which requires moving beyond literal wording of provisions, understanding the system and delivering poetic justice. Curiously, research shows that AI is surprisingly competent in making and clarifying normative judgements, but fares less well in overseeing the long term consequences and judicial precedent (https://lnkd.in/egmbECnF https://lnkd.in/d7zQhebu https://lnkd.in/eTpKCdzh).

One way to think about this legal precedent function of judges is Immanuel Kant’s categorical imperative (https://en.wikipedia.org/wiki/Categorical_imperative). Kant used this as the basis of all morality: the requirement any action can only be defended morally if it is based on a principle that you could want to be a general law. In layman’s terms, you should only act in ways that you would welcome other people to act like too.

Kant’s principle has a lot of intuitive appeal and Kantian ethics is one of the main school of ethics, though sometimes criticised as being too rule-based. But for a judge, this is an excellent starting point for writing a verdict. After all, case law means that a verdict quite literally will become a general law and used by other taxpayers as a guide to what the tax consequences are.

Judgements need to be based on the legal provisions as they are drafted and intended. Judgements also need to be just and poetic (i.e. they need to ensure there is a certain fairness / taxpayers getting what they deserve). But judgements also clarify the law and in that sense, they need to set a workable rule for the future. To set that rule, judges need to thoroughly understand how the system holds together to understand how their verdict – as a legal clarification – will impact the system as a whole. If judges only focus on the first task, or even the first two tasks, they will fail in their public role of creating a workable and just legal system.

So, judges are, in essence, tax policy makers, even if they don’t want to be. Some might object to this, see this as judges going beyond their mandate and quote the separation of powers made famous by Montesquieu. Judges favouring an originalist or textualist approach might see this as a moral problem. Here, Kant’s categorical imperative might help too. Would you want a general rule under which judges consider the legal precedent they create for future cases result in a workable overall system? The matter is up for debate, but my gut feeling is that this would be a better general rule than a prohibition to consider those impacts. So from a Kantian perspective, considering how to set juridical precedent appears moral. And again, amoral AI algorithms fail at that task.

Obvious disclaimers: this is not advice. These views are my own and do not necessarily represent my employer.