AI Risks "Hypernormal" Science

Current AI systems excel at making predictions within existing scientific frameworks but struggle to create paradigm shifts that drive major breakthroughs, potentially trapping science in patterns where researchers improve prediction while losing the ability to ask fundamentally new questions. Unlike transformative scientists like Einstein and Darwin who stepped outside prevailing logic to develop simpler, unified theories, today's AI is trained to minimize prediction errors against existing datasets, locking them into current conceptual vocabularies and making them unlikely to discover revolutionary concepts like germ theory or electromagnetic waves that require entirely new ways of thinking.