Reposted by kind permission of the School of Advanced Study Careers Service.
Why should you be interested in this question? I would argue because it goes to the very heart of where power lies in your country. The rule of law means that nobody is above the law (not even former US presidents) but who decides what the law is? Who develops the law? Is there transparency in the decisions that are made? Do you trust those decisions? How free from bias are the decisions?
Post the covid pandemic the use of technology has grown within the court system, for example with Online Dispute Resolution. As technology advances we are seeing increased use of AI in many aspects of our lives and this includes the judicial function. The level of use varies depending on the jurisdiction, with China probably the most advanced so far. Academics broadly agree that there are likely to be three potential levels of AI use in the judicial function from the most basic judge assist, to judge support, to full AI judge with no human intervention. Currently in the UK ChatGPT is being used by at least one Court of Appeal judge – Lord Justice Birss found it a ‘jolly useful tool’ when he used it to draft part of a judicial decision. I should stress that the content of the paragraph remained fully his responsibility. In the US lawyers were chastised by a judge for using ChatGPT to produce documents submitted to the court which included citations to fictitious cases. The issue was not the use of the chatbot per se but the absence of the necessary responsibility by the lawyers. So we can say that the first level of AI use is already here. Buocz warns that there is a danger that we may have AI imbedded in the judicial function to a significant extent before we are even aware of it.
The second level, judge support, could be for example the production of template decisions by AI. In the Netherlands AI was used to produce settlements for separating couples. Interestingly these were resisted as the couples required a human review of the settlements. The third level is where AI decides without human intervention. Despite reports to the contrary (see Estonia’s denial that they have fully robot judges) there are not currently AI judges that operate in the way we would expect in a full court scenario. But academics such as Sourdin believe this is probably only a matter of time. In Mexico AI is being used to produce decisions relating to pension entitlement. The use that is acceptable is likely determined by i) the type of use being proposed (assist, support, replacement) and ii) the nature of the case being heard (Sir Geoffrey Vos in his speech to the Bar Council argues that what we might accept for minor offences is likely to be different to child welfare where he concludes AI decisions are probably never going to be acceptable).
So, given that AI is already playing a part in our judicial function why should we be concerned and are there any benefits?. Humans are bias therefore human judges are bias. Every human judge will have a different bias. If AI is being used to make judicial decisions and all AI is derived from the same data set then it would logically follow that all judges would now have the same bias. Even when algorithms are expressly programmed to avoid bias (and who decides what that means in the first place?) it has been shown that they can be bias. The COMPAS AI system used in the USA to decide on bail was designed not to take into account factors such as race. However it was found to be racially bias in its decisions. Partly due to using zip codes (in the UK post codes) to define bail risks, which correlated to race.
If the data sets are to be unbiased who is going to create these data sets and from what data? In France the judiciary have actively worked to prevent their judicial decisions being made available for programming future decision making algorithms. If the judiciary are not involved in creating the data sets is it acceptable for the state or large private sector organizations to create these data sets and what does that mean for the separation of powers? It goes to the heart of power. If the data sets can be influenced then the outcome of court cases can be. This is even more significant in a common law jurisdiction where senior courts create precedent and therefore develop the law.
Human judges act with discretion and humanity. McInerney argues that AI’s lack of humanity means that it will never be capable of making judicial decisions. Buckland cites the example of King Solomon and the baby. Two women bring a baby to the King both claiming to be the child’s mother. The King declares that the baby should be cut in half. One women immediately says the other woman should take the baby to spare it coming to harm. King Solomon understood the first woman, who was willing to give the child away, was the mother but questions whether AI could ever reach this level of understanding of human nature. Would an AI judge just have seen an equitable division of assets? Others argue that AI is developing fast and may one day be able to reach levels of understanding and discretion not yet imaginable.
Is the conclusion therefore that AI should not replace human judges? In order to make a balanced decision consideration needs to be given to the potential benefits. The OECD estimated that in 2016 4 billion people did not have access to justice. In the English and Welsh courts there were 60,000 outstanding cases at the end of 2022. Delays or absence of justice have profound effects on peoples lives, not only victims but also witnesses and the accused. Is it better to have an AI judge decision promptly than to wait years for a human judge? Countries such as Brazil and China are actively pursing this in order to help with the large burden of cases their courts are facing. The benefit can be seen here but it should be noted that according to the World Justice Project both of these countries rank relatively low on the Rule of Law Index (Brazil and China rating about 0.5 compared with Australia, Estonia and UK around 0.8). There are concerns that the rule of law needs to be maintained as these AI judicial tools are being developed.
Coupled with access to justice is reduced costs and efficiencies. Bell et al describe the ‘saving of cost and time as the most compelling reason for the use of AI in the justice system and courts’. Countries such as Australia and Estonia are further down this journey than the UK but there is no doubt that there are government drivers to cut the costs of the judicial system and substantial investment in AI as well as other technologies within the UK. AI judges do not need lunch breaks or holidays. However there is likely to be a substantial cost associated with the creation of the data sets that the large language models need.
The question of bias was raised as a disadvantage, but Chen’s research on ‘having your day in robot court’ found that in some circumstances the public believed that computers could be less bias than human judges. In China, AI is reportedly used to review human judge decisions to decide if the outcome was what was expected. A principle of the rule of law is that like cases are decided alike but this does not mean that it would be considered generally acceptable to have an AI judge reviewing human decisions in many jurisdictions.
Sourdin summarized the ethical concerns relating to AI use in the judicial function as fairness, transparency , accountability, reliability, privacy and trust. It is not for me to tell you whether judges should be replaced by AI. It is dependent on your jurisdiction (your geographical lens so to speak), the type of AI use (assist, support, replacement) and your trust in all things AI. It is a global dilemma. But it is one we should be talking about before we have gone further than is acceptable under the rule of law without even realizing it.