The democratization dilemma: When everyone is an expert, who do we trust? (www.nature.com)

AI’s ability to produce expert-level outputs is reshaping how knowledge and trust are built across professions — creating an “instant expertise” paradox where anyone can generate seemingly authoritative answers without the contextual judgment that underpins real expertise. The Comment argues current governance (notably gaps in the EU AI Act’s transparency and human‑oversight provisions) doesn’t fully address risks such as user over‑trust, fragile LLM reasoning, erosion of professional standards, and homogenization of knowledge networks. For the AI/ML community this matters because it shifts validation from model performance alone to how model outputs are situated within real‑world professional practices and ethical considerations. To close that gap the authors propose a fourth regulatory pillar: Expertise Contextualization. Practically, this would require AI systems to embed dynamic context markers — including dynamic knowledge‑boundary maps that flag where human judgment is needed, contextualized confidence metrics that combine statistical certainty with situational and ethical cues, and domain‑specific expertise frameworks tying outputs to professional standards. Implementation pathways include regulatory pilots, cross‑industry standards, and shared expertise repositories, and the approach draws on polycentric and reflexive governance to enable multiple validation points. For practitioners and policymakers, these mechanisms offer a concrete route to preserve trust, maintain professional norms, and make human‑AI collaboration safer and more intelligible.