Interactive Context Refinement (ICR): Contextual Blindness in Generative AI Systems

Abstract

Large Language Models (LLMs) have transformed natural language processing across domains such as education, healthcare, and customer support. However, their inability to fully grasp user context-termed contextual blindness-leads to hallucinations, inaccurate personalization, and diminished user trust. This paper introduces Interactive Context Refinement (ICR), a dialoguedriven framework that enables generative AI systems to proactively refine user context before generating responses. We present the architecture, implementation, and evaluation of an ICR prototype, demonstrating its effectiveness in reducing hallucinations and improving response accuracy across multiple domains.

Keywords: Contextual Blindness, Generative AI, Interactive Context Refinement, Dialogue Systems, Hallucination Mitigation, Context-Aware Generation, Reinforcement Learning, Human-AI Interaction, Natural Language Understanding