Luxembourg, 30 September 2024
This document outlines the Artificial Intelligence Strategy of the Court of Justice of the European Union. It starts with an introduction covering a definition of A.I., typologies and a brief status quo, continues by articulating a vision supported by goals and objectives, states the principles, makes a short readiness assessment, gives an overview of the risks, and concludes with a proposal for the governance and architectural approach.
Executive summary
Artificial Intelligence (A.I.) has been around since the 1950s, but its recent exponential growth can be attributed to two factors: first, the availability of larger datasets for training and second, increased computational power. We can expect further rapid advancements in the future as A.I. continues to evolve. It holds significant potential for the Court of Justice of the European Union (ECJ), on the one hand by enabling automation of simple tasks in both the judicial and administrative areas and, on the other hand, by offering new possibilities in legal research, translation, interpretation, enhancing accessibility and providing a modernised approach to accessing information.
This technology can be leveraged to achieve multiple goals, with a primary focus on three key areas: improving the efficiency and effectiveness of our administrative and judicial processes, enhancing the quality and consistency of judicial decisions, and increasing access and transparency for EU citizens.
Automating various tasks can significantly enhance efficiency and effectiveness. In the judicial field, several examples include the detection of references in documents, suggesting metadata, proposing subject matters during case logging, and generating transcripts of hearings through speech-to-text algorithms. Besides automation, A.I. could soon be available for more advanced support: search engines that understand context, summaries of large documents proposed by a machine, or anonymisation algorithms. In the administrative realm, we could witness improved support for building operations, automation in financial processes (such as automatic invoice handling), and the integration of virtual assistants into our daily tools. These virtual assistants could assist in scheduling meetings, preparing presentations and documents, and composing response emails, among other tasks.
The quality and consistency of judicial decisions could also be enhanced through this technology by enabling automatic correlation and classification of cases, automating the processing of originating documents, and potentially employing advanced legal research tools as technology evolves. The current capabilities already allow for automatic detection and visual display of related cases. It is worth considering conducting a study to better understand the implications, and particularly the risks, of incorporating predictive A.I., as lawyers outside the ECJ appear to be open to its adoption.
A.I. could bring benefits in increasing access to justice and transparency vis-à-vis EU citizens. In the case of people with disabilities, they could benefit, for example, from having automatic subtitles for hearings that are streamed and virtual assistants could support them in satisfying their inquiries with video or audio responses in their own language. Chatbots, virtual assistants, A.I. avatars integrated within our website could also support the general public or legal professionals in their quest for information. It is expected that translation and interpretation tools will evolve in the coming years and therefore these tools will need to be integrated everywhere in order to decrease language barriers. It is expected that soon, effective and affordable tools will be able to perform a sentiment analysis on the reactions of the general public or legal professionals. This type of feedback will help us fine-tune our communication.
The adoption of such technology does not come without risks. It is essential to have a comprehensive understanding of these risks to make informed assessments and develop effective mitigation strategies. Chapter 4 of this paper addresses this topic.
One of the main risks associated with the adoption of A.I. technology is the possibility that involuntary biases may be introduced during the training of A.I. models, resulting in unintentional discrimination. Another risk pertains to the potential disclosure of sensitive data when using algorithms in the cloud.
Additionally, there is concern regarding the generation of false, inaccurate, or irrelevant information by generative A.I. algorithms (such as ChatGPT), which can lead to phenomena known as “hallucinations” where A.I. algorithms invent information that sounds plausible but is not factual. Over-reliance on technology without the application of critical human thinking can also result in improper usage.
The rapid advancement of A.I. algorithms may lead to uncontrolled or excessive usage (hyper abuse) without adherence to all the in-house good-practice rules. Unfortunately, A.I. can provide new tools for sophisticated cyberattacks, enabling impersonation and data pollution that can be exploited by malicious actors. Understanding and addressing these risks are crucial for responsible and secure use of A.I. technology within the ECJ.
One of the important mitigation strategies is to adopt an appropriate governance model, fully integrated into the existing model. The proposal in this paper is to create an A.I. Management Board, which will support the Court in this process, via a “risk-based” approach, based on the principles highlighted in Chapter 3. This Board will define the “red lines” (the areas in which this technology should not be used due to the high risk), the areas of special attention (in which the adoption of A.I. algorithms needs to be done with prudence and appropriate controls), and areas that, due to their specificity, present a low risk. It is important to highlight that one of the proposed objectives is the adoption of a governable A.I., which is seen as a continuous process.
At the technical level, it is crucial to adopt or create A.I. algorithms in a controlled manner and with a clear architectural approach to ensure high reusability. This approach ensures that the same type of algorithm is implemented only once and can be reused across different cases where it can bring benefits. The paper presents a clear and comprehensive high-level capabilities map for A.I.
At the resources level, the paper emphasises three key areas. Firstly, it highlights the importance of creating the necessary instruments to upskill and reskill the internal workforce, including IT professionals, managers, and staff. The paper suggests leveraging training provided through an Emerging Technologies Academy, establishing a mobility board, re-evaluating the overall sourcing strategy, and implementing robust change-management capabilities. Secondly, the paper emphasises the ongoing use of the Innovation Lab and existing innovation network as catalysts and central hubs for eliciting, testing, and evaluating innovative initiatives throughout the ECJ. Lastly, the paper proposes analysing the required level of investments based on the level of ambition set by the ECJ. Chapter 5 provides an initial overview of investment considerations in this regard.
With three years of experience in exploring and experimenting with artificial intelligence technology, the ECJ is now ready to step into an “industrialisation phase.” Besides a thoughtful reflection, this requires the development of a strategy for moving forward, and the implementation of several key steps to establish a strong foundation. The paper outlines these steps in detail throughout its various chapters and provides a summary at the end.
Based on the strategy document provided, here is a table of pros and cons of using artificial intelligence (AI) in the Court of Justice of the European Union (CJEU):