Judgments issued ex machina? Are AI-made judicial decisions the future?

In the future, general access to information and electronic communication will be enhanced still further. To establish digital assistance, the integration of artificial intelligence (AI) will increase. AI is considered the key technology of the future and thus opens new doors and questions.

The potential use cases of AI in the Austrian judiciary are currently focused on supporting judges, for example via AI-automated factual legal research or the automatic creation of minutes during hearings using speech recognition. In addition, to optimise internal court workflows, AI could be used to recognise documents and extract metadata as well as to identify internal judicial responsibilities.

Who's in charge? Limitations for AI-made judicial decisions

Any judicial decision must be made by a human and not a robot. In this context, special emphasis must be put on the constitution guaranteeing a judge's independence. AI links data and makes decisions based on algorithms and probabilities. But the algorithms powering AI are programmed by engineers. So who is responsible for the AI's decisions? The judge? The engineer? Particularly if the algorithm is not disclosed, it would be difficult or impossible to understand, question, check or control a decision solely issued by AI. Even if a judge is merely using an AI tool in a supportive capacity, how can they assess the AI's results if they do not understand how the AI-assisted program reached its decision?

What's more, AI is mostly a self-learning and thus constantly evolving system, but one that is only as good as its underlying training data. If AI learns based on misguided views or erroneous past court decisions, these systematic errors will be reflected in future AI decisions. The AI may, for example, recognise the credibility of witnesses based on objective features such as micro facial expressions and therefore be less likely to be guided by subjective impressions, but – apart from triggering potential new data protection issues – value judgements are legitimate and essential parts of our current legal system. Accordingly, free assessment of evidence by the judge in civil proceedings has high value. If judicial decision-making competence were fully transferred to AI, the necessary individualism for a case-by-case decision would be lost, and legal development would come to a standstill.

Robo judges? Not so fast

Replacing a judge with AI is only conceivable in standardised proceedings where the judge bears ultimate responsibility and may review, amend, edit and overrule the AI's decision. In civil proceedings, this could include, for example, cost decisions or automated alimony calculations. AI could also be used in dunning proceedings with low amounts in dispute, where a payment order is issued without close judicial review and the case is heard only upon objection by the defendant. If a judicial decision is based solely on standardised documentary evidence that does not require interpretation or the resolution of a pure question of (simple) law, AI could be used further to assist judges and increase capacities for more complex legal problems. Any such decision would nonetheless have to be challengeable and reviewed by a judge.

"Replacing a judge with AI is only conceivable in standardised proceedings where the judge bears ultimate responsibility and may review, amend, edit and overrule the AI's decision."

In summary, the use of AI in routine and background judicial activities, as well as for supporting the judiciary in standardised smaller proceedings in the future, is welcome in terms of increasing efficiency, shortening the duration of court proceedings and saving costs. Beyond that, judges do not have to fear for their jobs. At least not yet...

authors: Sarah Rosenthaler and Sara Khalil