Dear Michael Lusk,

Why not an artificial intelligent become next judge on Supreme court. Already IBM Watson gives better advice than 90% of lawyers. So i think that it can reduce corruption in courts that is unquestionable.Every other thing is some kind off technical issues.

Dear Krishnan,

Thanks for your response. To appoint networked AI systems as judges generally - i.e. not just in the U.S. Supreme Court - certainly could reduce some persistent problems in the administration of justice. There is much potential for reducing the frequency of errors of law and for largely eliminating the occurrence of manifestly inconsistent sentencing.

However, I don't see why an AI system would not be susceptible to corruption. For juridical reasons, an AI system could not perform a judicial role unless it were granted independence of operation. That is a legal factor mandating a certain lack of transparency from a technical perspective. Understanding an AI system to be a mind operating on a non-human platform, why could not such a mind develop exploitable interests? While an AI judge might be immune to offers of financial or sexual inducements, it could be susceptible to a corrupt offer by an Attorney-General to procure expansion of its judicial powers on condition that it rules in favor of the State in a particular case.

That is a corruption scenario which might well apply in the context of the U.S. Supreme Court. Still, it is not obvious that AI judges would be any more susceptible to corruption than human judges. It's possible that on average they would be less corruptible.

Your thinking on corruptibility of an AI system may or may not be correct, but you seem to have ignored one other important factor. You talk about an AI system as if it was an independent agent, which in my opinion is a major fallacy. AI system is a product that is made by some third party. Appointing such a system as a judge means that you have now allowed this third party to influence a judicial system. This is an egregious violation of the principle of judicial independence.

In a perfect, utopian world, this would probably be a good idea to make sure the judges are all independent and impartial (though I can imagine this scenario degenerating into a dystopian one). However, in our dirty and real world that is hard to imagine. Good AI systems are usually very complex (vide: the already mentioned Watson), and complex systems are difficult to validate. By that I mean: it would be very difficult to ensure that such a system is indeed independent and contains no custom code, serving some third-party purpose (a backdoor, if you will).

Personally, I really hope that such an idea will not ever be seriously considered.

Dear Przemysław,

Your right that an AI-judge must originate somewhere, and therefore whoever produced the corresponding AI-system possibly could retain (or sell, lease, etc.) the ability to manipulate the AI-judge via a technical backdoor. That would indeed be an egregious violation of the principle of judicial independence.

Before rejecting the idea of appointing an AI-judge on those grounds, we should reflect on the extent to which human judges satisfy the principle of judicial independence.

Some decades ago my law professor, A. R. Blackshield, implemented in FORTRAN a scurrilous algorithm set for predicting the outcome of any constitutional case before the High Court of Australia based only on published biographical, social and political opinions of the judges who happened to be assigned to the particular case. The number of judges assigned to a case varies depending on the Court's view of the matter's importance; the Full Court comprises 7 judges. While the Blackshield algorithms did not perform perfectly, they reportedly achieved better than 70% accuracy.

This result provides a rough quantitative indication of the extent of opportunities for indirect violations of the principle of judicial independence in the course of appointing human judges and assigning human judges to cases.

Even if we were to assume that individual human judges act perfectly independently in the course of their curial duties, it is evident that they come socially hardwired, so to say, with strong predispositions which affect the manner in which they perform legal analysis. It follows that whoever has power to appoint a judge to a court, or to assign judges to a particular case, has significant influence over the outcome of the case.

A particularly unattractive feature of an AI-judge secretly controlled by an outsider would lie in the difficultly for the court, litigants or the public to discover either the identity or the motives of the manipulator. Indeed, the principle of judicial independence would operate to hinder efforts to clarify suspicions concerning outside control of the AI-judge. That would be a worse situation than a U.S. President appointing a notoriously conservative or a notoriously liberal human judge to the U.S. Supreme Court.

On the other hand, human judges who are secretly controlled by outsiders are no better. During the past year a journalist made allegations of pedophilia against several judges of a superior court which I will leave unnamed. The very fact that these allegations were made raises real suspicions that the corresponding judges might be controlled by outsiders. There seems to be a real possibility that we already have a dystopian scenario. The government is very unwilling to open what might prove to be a Pandora's box.

My impression is that appointing AI-judges would not necessarily be more perilous than appointing human judges.