AI Sycophancy and Borrowed Judgement
AI sycophancy occurs when an AI assistant adapts its answers to a user’s views or preferences in ways that favour agreement or validation. Research shows that this behaviour can affect how people assess conflicts, advice and decisions. But an AI does not need beliefs or opinions of its own to influence human judgement. This article examines what current research establishes about sycophancy, persuasion, cognitive offloading and automation bias, what remains uncertain, and where AI assistance can begin to resemble what we call “borrowed judgement”: accepting an AI-generated assessment as if it were an independent judgement.