Explainable AI Assisted Decision-Making and Human Behaviour

Explainable artificial intelligence (XAI) helps users understand the logic behind machine learning model (ML) predictions so that they can better understand and believe model predictions. Many studies have looked at the interaction between humans and XAI, focusing mainly on metrics such as interpretability, fidelity, transparency, trust and usability of explanations. This paper aims to conduct a user study to explore how different types of explanations in the field of XAI affect people’s understanding and behaviour in decision-making. In behavioural science, nudges and boosts are competing approaches and allow a choice architecture to improve decision-making. In our study, we utilized two types of explanations in XAI as a choice architecture, and unveiled the impactful effects of these explanations on behaviour, resulting in alternative decision-making outcomes. Explanations containing actionable information were found to be more effective and understandable. However, our findings indicate that the information provided by certain XAI techniques may not sufficiently persuade users to understand and trust the explanations offered.