Post-hoc explanations for AI in transport planning

Explainability is crucial in high-stakes domains because mere high performance might not meet all stakeholders' needs. In transport, a typical high-stakes domain, stakeholders require post-hoc testing to gain deeper insights into the decision-making process, enhanced error-tracing support, and validation of the AI system's trustworthiness. However, there is a gap between the explainability provided by general post-hoc XAI methods and the actual needs in practice. This talk aims to bridge this gap by introducing an XAI framework tailored to meet the requirements of transport stakeholders for different purposes. We will then showcase two transportation application to demonstrate how XAI methods can explain why a tested model does or does not meet the explainability needs in the transport planning domain.