ניר גרינברג

The advances of recent years in generative artificial intelligence (AI) have provided ample new means to improve conceptual modeling. Yet, relatively little research has examined how AI solutions can benefit from conceptual modeling. Here, we demonstrate how conceptual modeling can support Explainable AI (XAI) rather than black-box solutions in high-stakes decision-making, thus contributing to the model’s interpretability and likelihood of adoption. In particular, we reformulate a complex AI task - finding similar criminal cases - using a conceptual model that facilitates factual and interpretable AI inferences. Currently, attorneys look for similar cases manually, which is time- and resource-consuming, involving many complex comparisons, and resulting in a selection of cases that is potentially biased. Our conceptual model-based solution, in contrast, uses AI to populate values in the conceptual model from the unstructured case text, and learns what makes two cases similar from expert judgment. The findings show that our approach identifies similar cases and outperforms black-box AI solutions by 10.0% in terms of F₁ while delivering interpretable results based on the conceptual model.