Using Large Language Models to Compare Explainable Models for Smart Home Human Activity Recognition

Explainability matters for healthcare AI: The paper addresses a critical gap where deep learning models used to monitor daily activities in smart homes (potentially detecting cognitive decline) operate as "black boxes." Clinicians and other non-expert users need to understand why a model makes predictions before they can trust and act on them in clinical settings.

Different explanation methods produce different results: The research found that various explainable AI (XAI) methods generate different explanations for the same activity recognition task. This suggests there's no one-size-fits-all approach, and choosing the right explanation method matters for how well non-experts understand the system.

LLMs can automate evaluation of AI explanations: Rather than relying solely on expensive and time-consuming user surveys to test whether explanations are actually helpful, the authors propose using Large Language Models to automatically evaluate which XAI methods work best. Their preliminary findings suggest this automated approach aligns with what real users prefer.

Practical path toward trustworthy healthcare AI: By combining explainable AI methods with automated evaluation tools, this work offers a more scalable way to develop activity recognition systems that clinicians can understand and trust—potentially accelerating the deployment of AI in healthcare monitoring without sacrificing transparency.