Zhijun Song
What does an AI system’s behavior reveal? Does success demonstrate capability? Does change indicate learning? Does association establish causation? One principle separates defensible judgment from impression: interpretation should never outrun evidence. AI Behavior: A Science of Observing, Measuring, Testing, and Interpreting develops a systematic approach to determining what observed behavior supports and where the evidence leaves conclusions unresolved. For AI researchers, developers, evaluators, and scholars, this book offers a rigorous framework for making more defensible judgments about AI systems and the limits of what can be established about them.