Explainable AI for Everyone

The increasing adoption of Machine Learning in sensitive sectors like healthcare and finance raises questions about the transparency of automated decisions. Explainable AI (XAI) aims to clarify how models generate their outputs, but often requires specialized technical skills.

DashAI and the XAI Module

A new study focuses on integrating XAI into no-code ML platforms, designed to democratize access to AI. Researchers have developed an XAI module for DashAI, an open-source platform, which includes techniques such as Partial Dependence Plots (PDP), Permutation Feature Importance (PFI), and KernelSHAP.

Study Results

A user study (N = 20, including novices and experts) evaluated the usability and impact of the explanations provided by the module. The results show a high success rate in explanation tasks (≥80%). Novices rated the explanations as useful, accurate, and trustworthy, while experts expressed greater reservations about their completeness. The explanations improved the perception of predictability and trust in automation, especially among novices.

This study highlights the challenge of making XAI explanations accessible to novices and sufficiently detailed for experts, a crucial aspect for the widespread adoption of AI.