AI systems may identify stable, high‑value patterns in scientific data that are too complex for humans to compress into simple formulas or intuitively grasp. Those discoveries could be usable (for materials design, drug discovery, etc.) even if human researchers cannot fully explain or teach the underlying principles.
— If true, this would change who 'does' science, how results are validated, and how societies govern and trust machine-generated interventions.
Noah Smith
2026.03.22
100% relevant
Noah Smith’s argument that human science favors compressible laws and his LLM example suggesting AI can find complex-but-useful patterns in domains like materials science
← Back to All Ideas