SNAPE-PM: Building and Utilizing Dynamic Partner Models for Adaptive Explanation Generation
Amelie S. Robrecht, Christoph R. Kowalski, Stefan Kopp
TL;DR
This work reframes explainable AI as an adaptive, interactive task by modeling the explainee and context with a Dynamic Bayesian Network ($DBN$) and solving a non-stationary Markov Decision Process ($NSMDP$) to select the next explanatory move. The approach, dubbed SNAPE-PM, tightly integrates a groundable knowledge base with PM state and uses Monte Carlo Tree Search ($MCTS$) to enable real-time, adaptive explanations that tailor content to listener expertise, cognitive load, attentiveness, and cooperativeness. Evaluation on five simulated personas across varied feedback patterns demonstrates that the system can develop distinct explanation strategies and adapt quickly as partner states evolve. The framework holds promise for more explainable and user-adaptive dialog systems, with future work including live user studies and end-to-end NLU/NLG integration.
Abstract
Adapting to the addressee is crucial for successful explanations, yet poses significant challenges for dialogsystems. We adopt the approach of treating explanation generation as a non-stationary decision process, where the optimal strategy varies according to changing beliefs about the explainee and the interaction context. In this paper we address the questions of (1) how to track the interaction context and the relevant listener features in a formally defined computational partner model, and (2) how to utilize this model in the dynamically adjusted, rational decision process that determines the currently best explanation strategy. We propose a Bayesian inference-based approach to continuously update the partner model based on user feedback, and a non-stationary Markov Decision Process to adjust decision-making based on the partner model values. We evaluate an implementation of this framework with five simulated interlocutors, demonstrating its effectiveness in adapting to different partners with constant and even changing feedback behavior. The results show high adaptivity with distinct explanation strategies emerging for different partners, highlighting the potential of our approach to improve explainable AI systems and dialogsystems in general.