• counterfactuals
• human in the loop
• interactive learning
• machine learning
• skeptical learning
• xai

The growing integration of Artificial Intelligence into decision-making processes has raised increasing concerns about model transparency and user trust. While many machine learning systems achieve high predictive performance, their lack of interpretability limits their use in scenarios where human oversight is required. This thesis explores the dynamics of human–AI interaction within evolving systems, focusing on situations where the user is empowered to question and correct model outputs. To address this, an extension of an existing continual learning framework is proposed, enriched with mechanisms aimed at generating interpretable explanations and incorporating user feedback. Through simulated experiments involving diverse user profiles and data distributions, the study assesses the quality of explanations, the evolution of user skepticism, and the impact of corrective interventions. The results contribute to the development of more adaptive and transparent AI systems, capable of fostering human trust and enabling collaborative decision-making.