EMMM, Explain Me My Model! Explainable Machine Generated Text Detection in Dialogues

AI-generated text poses real security risks in customer service: The paper identifies a concrete threat where malicious actors could use large language models to impersonate users at scale, highlighting why detection systems are necessary for protecting consumers in online conversations.

Explainability matters more than accuracy alone: EMMM's framework prioritizes making detection decisions understandable to non-expert customer service operators (70% of human evaluators preferred its explanations), recognizing that trustworthy AI deployment requires users to understand why a system made a decision, not just whether it was correct.

Speed and interpretability can coexist with accuracy: The system generates explanations and detection results within 1 second while maintaining competitive accuracy, demonstrating that responsible AI doesn't require choosing between being fast, accurate, and interpretable—all three are achievable.

Current detection methods fail in real-world conversational settings: Existing machine-generated text detection approaches struggle specifically in dialogue contexts (like customer service chats), suggesting that lab-tested AI systems may not work reliably when deployed in actual business environments.

Open-sourcing detection tools supports broader responsible AI adoption: By releasing their code and dataset publicly, the researchers enable other organizations to build and improve detection systems, rather than leaving this critical security capability available only to well-resourced companies.