Abstract
In this era of artificial intelligence (AI), ambiguity presents a significant challenge for information and communication management, often leading to misinterpretations and inefficiencies. The rise of generative AI (Gen AI) has further amplified this issue by producing text that is often unclear or open to multiple interpretations, which could impact decision-making in many critical areas. To address these challenges, we propose AICMA: a framework for AI-driven Identification, Classification, and Mitigation of Ambiguity. The AICMA framework consists of three core stages: Identification, which determines whether a given sentence is ambiguous or not; Classification, which classifies a sentence based on ambiguity level (High - Ambiguous, Low - Ambiguous, or Not - Ambiguous); and Mitigation, which utilizes large language models (LLMs) to adapt and regenerate ambiguous sentences for enhanced clarity while preserving their original intent. By improving textual interpretability, AICMA offers significant value across various domains, including education, healthcare, policy-making, and beyond. This framework is based on our theoretical framing of ambiguity, particularly in AI-generated text. It has the potential to contribute to developing more robust and reliable AI systems that will produce clearer and more interpretable outputs. Its adaptability will allow it to be integrated into existing AI systems, making it a versatile framework for developers and researchers aiming to enhance information and communication effectiveness Ultimately, AICMA represents a significant step forward in addressing the complexities of ambiguity in AI-generated text, paving the way for more transparent and effective AI-driven text and speech generation solutions, from conversational agents to critical decision-support systems.