ExplainabilityAccountabilityTrustworthinesscs.AI
Shaping Scientific Explanations to Expert Perspectives with Persona-Conditioned Reinforcement Learning
Susana Nunes, Tiago Guerreiro, Catia Pesquita
March 23, 2026
Key Takeaways
-
Explanation quality is not one-size-fits-all: The research demonstrates that different experts legitimately prioritize different types of evidence and reasoning when evaluating scientific explanations. Rather than seeking a single "best" explanation, the authors show that tailoring explanations to individual expert perspectives significantly improves how relevant and valid experts perceive them.
-
AI can learn expert perspectives without extensive supervision: By creating "personas" that represent how different experts evaluate explanations, the system can generate customized explanations using reinforcement learning without requiring large amounts of labeled expert feedback. This approach reduced the expert feedback time needed by two orders of magnitude (100x faster), making it more practical for real-world deployment.
-
Perspective-conditioned explanations maintain scientific rigor: The personalized explanations didn't sacrifice accuracy—they matched or exceeded the predictive performance of general-purpose explanation systems. This suggests that adapting to expert perspectives is compatible with, rather than opposed to, maintaining scientific validity.
-
This approach addresses a key gap in responsible AI: Most explainable AI systems assume there's an objective "correct" explanation, but this work recognizes that scientific reasoning involves legitimate epistemic variation. Acknowledging and accommodating different expert perspectives represents a more honest and respectful approach to human-AI collaboration in scientific discovery.
Abstract
Explainable AI is increasingly important to scientific discovery. However, existing methods largely ignore that explanation quality is not universal: experts differ in how they assess evidence, prioritize mechanisms, and construct explanatory narratives. We introduce perspective-conditioned explanations, a framework for adapting explanation generation to epistemic variation in expert judgment. Using knowledge graph reasoning paths in drug discovery, we show that preferences organize into coherent epistemic perspectives that can be captured by agentic personas, representations of how experts evaluate explanations. Persona-aligned rewards then guide reinforcement learning-based explanation generation without large-scale expert supervision. Expert user studies show that perspective-conditioned explanations are preferred over general-purpose explanations and improve perceived relevance and validity. Moreover, they match or exceed state-of-the-art predictive performance and reduce expert feedback time by two orders of magnitude. Together, these findings demonstrate that explanation quality is perspective-dependent and that modeling this variation enables scalable and human-aligned explanation generation for scientific discovery.
Paper Info
Responsible AI Topics
ExplainabilityAccountabilityTrustworthiness
Links
Citation data from Semantic Scholar