Project R-14478

Personalized, intelligible user interfaces through human-in-the-loop generative AI (Research)

Research in AI is a hot topic that continues to grow rapidly. AI, however, tends to present itself as a black box: it is unclear what it does "under the hood" and users typically get results without explanation or justification. This is where explainable AI (XAI) comes into play. XAI, however, mainly succeeds in making this black box(somewhat) accessible to developers, using cookie cutter methods, and not in assisting the actual end users in a practical manner. We envision an approach to enhance the accessibility of the intelligibility of AI systems, taking into account the contextual differences between end users. This leads us to the central research challenge of this PhD: take advantage of the power of generative AI to create personalized, intelligible user interfaces that offer end users the necessary tools to understand, interrogate and control AI-based systems. These user interfaces aim at instilling an appropriate level of trust in AI, by explicating the uncertainties that are inherent to such systems. The focus is on human-centered AI, with human-inthe-loop approaches to ensure the controllability and adaptability of AI systems. We will apply this research to another contemporary topic: the energy sector. While there is a lot of research on AI in this domain, in particular on predictive models, our focus is on end users with different information needs at different times and how (X)AI can provide them with practical insights and actionable recommendations.

16 September 2023 - 15 September 2027