Generative large language models as tools in the legal domain have the potential to improve the justice system. However, the reasoning behavior of current generative models is brittle and poorly understood, hence cannot be responsibly applied in the domains of law and evidence. This paper presents reasoning benchmarks that are dynamically varied, scalable in their complexity, and have formally unambiguous interpretations. In this study, we illustrate the approach on the basis of witness testimony, focusing on the underlying argument attack structure. We dynamically generate both linear and non-linear argument attack graphs of varying complexity and translate these into reasoning puzzles about witness testimony expressed in natural language. We show that state-of-the-art large language models often fail in these reasoning puzzles, already at low complexity. Obvious mistakes are made by the models, and their inconsistent performance indicates that their reasoning capabilities are brittle. Furthermore, at higher complexity, even state-of-the-art models specifically designed for reasoning make mistakes. We show the viability of using a parametrized benchmark with varying complexity to evaluate the reasoning capabilities of generative language models, which contribute to a better understanding of the limitations of the reasoning capabilities of generative models.

Machine learning is increasingly used in the legal domain, typically operating retrospectively by treating past case outcomes as ground truth. However, legal outcomes are often shaped by human interventions that are not captured in most machine learning approaches. A final decision may result from a settlement, an appeal, or other procedural actions. This creates label indeterminacy: the outcome could have been different if the intervention had or had not taken place. We argue that legal machine learning applications need to account for label indeterminacy. Methods to impute indeterminate labels exist, but are all grounded in unverifiable assumptions. In the context of classifying cases from the European Court of Human Rights, we show that the way that labels are constructed during training can significantly affect model behaviour. We therefore position label indeterminacy as a relevant concern in AI & Law and demonstrate how it can shape model behaviour.

Data-driven AI systems can make the right decisions for the wrong reasons, which can lead to irresponsible behavior. The rationale of such machine learning models can be evaluated and improved using a previously introduced hybrid method. This method, however, was tested using synthetic data under ideal circumstances, whereas labelled datasets in the legal domain are usually relatively small and often contain missing facts or inconsistencies. In this paper, we therefore investigate rationales under such imperfect conditions. We apply the hybrid method to machine learning models that are trained on court cases, generated from a structured representation of Article 6 of the ECHR, as designed by legal experts. We first evaluate the rationale of our models, and then improve it by creating tailored training datasets. We show that applying the rationale evaluation and improvement method can yield relevant improvements in terms of both performance and soundness of rationale, even under imperfect conditions.

In AI and law, systems that are designed for decision support should be explainable when pursuing justice. In order for these systems to be fair and responsible, they should make correct decisions and make them using a sound and transparent rationale. In this paper, we introduce a knowledge-driven method for model-agnostic rationale evaluation using dedicated test cases, similar to unit-testing in professional software development. We apply this new quantitative human-in-the-loop method in a machine learning experiment aimed at extracting known knowledge structures from artificial datasets from a real-life legal setting. We show that our method allows us to analyze the rationale of black box machine learning systems by assessing which rationale elements are learned or not. Furthermore, we show that the rationale can be adjusted using tailor-made training data based on the results of the rationale evaluation.

The justification of an algorithm’s outcomes is important in many domains, and in particular in the law. However, previous research has shown that machine learning systems can make the right decisions for the wrong reasons: despite high accuracies, not all of the conditions that define the domain of the training data are learned. In this study, we investigate what the system does learn, using state-of-the-art explainable AI techniques. With the use of SHAP and LIME, we are able to show which features impact the decision making process and how the impact changes with different distributions of the training data. However, our results also show that even high accuracy and good relevant feature detection are no guarantee for a sound rationale. Hence these state-of-the-art explainable AI techniques cannot be used to fully expose unsound rationales, further advocating the need for a separate method for rationale evaluation.