T-REX: A Framework to Build Trustworthy Recommenders of Evidence Explanation

The initial enthusiasm for eXplainable Artificial Intelligence (XAI) has been tempered by concerns about the effectiveness and reliability of its explanations. Studies show that some explanations are no more reliable than random ones. Tim Miller suggests a paradigm shift in XAI to address issues of cognitive biases, such as automation bias, which can affect decision-making processes. He advocates for hypothesis-driven support systems to align AI explanations with human cognitive processes. Addressing these issues, we propose the Trustworthy Recommenders of Evidence eXplanations (T-REX) framework. This approach aims to enhance XAI by moving from statistical explanations to those based on trustworthy scientific evidence, enabling AI systems to tackle complex tasks more effectively.