Tag: artificial-intelligence

AI in the courtroom: promise, safeguards, and the digital reality

Where the AI Act draws the line

When the IBA panel asked me to say a few words about the EU AI Act, one question stood out: what does it mean for courts? The Act treats certain AI uses in the administration of justice and democratic processes as high risk because they can pose risks to democracy and the rule of law. An example is a system that helps a judicial authority reach a ruling with legal effect. Such systems must meet requirements designed to prevent or reduce those risks, including risk management, data quality, technical documentation, registration, transparency, and human oversight. As public authorities and entities providing public services, courts and judiciaries must also conduct a fundamental rights impact assessment for high-risk systems.

Support the judge, not replace them

AI can support judicial decision-making and judicial independence, but it must not replace the person making the final decision. The Act also draws a distinction between that kind of assistance and purely ancillary administrative work that does not affect the administration of justice in individual cases. Examples of the latter include anonymizing or pseudonymizing judicial decisions, documents, or data; facilitating communication between personnel; and carrying out administrative tasks. The high-risk classification should not extend to systems used solely for those ancillary activities.

What are courts using today?

To see how courts in Council of Europe member states are using technology for administration and decision support, I turned to the most recent CEPEJ evaluation report. If CEPEJ’s “case management” category is treated as ancillary administration, it has the highest score on the ICT Deployment Index: 5.66. If “decision support” includes assistance in reaching rulings with legal effect, its score is much lower: 2.64. Those comparisons depend on how the categories are interpreted. CEPEJ says the figures show that countries are still focused on basic digital infrastructure, such as e-filing and case registration.

The next wave is taking shape

The EU Justice Scoreboard reports that nine out of 27 EU member countries report using AI for “core activities”. The current CEPEJ evaluation cycle also shows that AI is being used in areas such as class actions, automatic anonymization of judgments and specialized translation. These developments may become more visible in the next evaluation cycle. For now, the contrast is striking: while the possibilities are expanding, the foundations of court digitalization are still being built. Those statistics are dated 2022. CEPEJ says it will publish its next evaluation in December 2026, with more recent information on the use of AI.

AI in the courtroom: a world-wide experiment

We are all part of a worldwide experiment. An experiment is a procedure carried out to determine the efficacy or likelihood of something previously untried, to provide insight into cause-and-effect by demonstrating what outcome occurs when a particular factor is manipulated. For the world-wide AI and justice experiment, we all provide materials, input, for the experiment to test the effects of the output of artificial intelligence. Most of the input is probably case law published online. The output is AI results, the factor to be manipulated is the workings of the AI.

Lawyers’ and case parties’ case information needs to be correct, and it also needs to be explainable, that is, based on solid reasoning and recognized sources. When the discourse found AI did not explain  what its sources were, the output factor was manipulated, and the references were included. So now, we can at least see where the information comes from. But that has not solved the problem. AI’s sources are not always real; AI fabulates sources, apparently because it is trained to please the users. The sources may be non-existent, they may exist but be irrelevant to the case, and they may argue something completely different. Courts and lawyers found this out when they checked the sources in claim documents. Court staff now need to check each reference, which creates a lot of extra work. Judges using AI for their judgments also need to conform to some rules. One is fair procedure: parties must have equal opportunities to put their own case forward and test the other party’s case. This is standard case law now. Judges must also be transparent in their use of AI: in general search results, in testing evidence parties must have a chance to test the results. For judges finding general search results on the internet, judges searching for information. factual data not in the case file, but found on the internet of its own accord the case law is clear. The information needs to be disclosed and parties need to be given the opportunity to present arguments on the matter. Judges also need to be transparent about AI in composing their judgments. By now, there is an abundance of guidance on the use of AI. I will discuss that in another blog.