Episode 24- Zihao li and Jiahong Chen - The Accuracy Paradox: AI Hallucination, Regulation, and the EU-China Divergence

Anush Ganesh and Kena Zheng speak with Dr Zihao Li (University of Glasgow) and Dr Jiahong Chen (University of Sheffield) about their recent paper (with Weiwei Yi) challenging the dominant regulatory approach to AI hallucination. Rather than treating hallucination as a simple failure of factual accuracy, Zihao and Jiahong develop a layered taxonomy that includes sycophancy, consensus illusion, oversimplification, and prompt-sensitivity effects, and argue that hyper-optimising for accuracy may paradoxically deepen the very harms regulators seek to prevent. The conversation examines the EU AI Act, the GDPR, and the DSA before turning to a substantial comparative discussion of China's generative AI regulatory framework, including the Interim Measures for Generative AI Services and earlier rules on deep synthesis and algorithmic recommendation. The episode explores where the European and Chinese approaches converge around accuracy as a regulatory anchor and where they part ways in terms of institutional design and content governance. The discussion closes with Zihao and Jiahong's proposals for moving beyond accuracy toward epistemic trustworthiness, and whether pluralism, confidence calibration, and reflective design can gain traction amid growing pressures around AI competitiveness. Here is a link to their paper published with Computer Law and Security Review- https://www.sciencedirect.com/science/article/pii/S2212473X26000520