Physical 'laws' are not necessarily unique metaphysical truths but are representational choices—compressions of data—that balance prediction error, description length, computational cost, and scope. Different choices sit on a Pareto surface; with modern computation and machine learning we can systematically search for alternative, equally valid formulations.
— If laws are seen as pragmatic compressions, that shifts debates about scientific realism, research funding, and the governance of AI‑assisted theory generation.
Seeds of Science
2026.03.11
100% relevant
The article's explicit optimization proposal (minimize prediction error E, description length L, computational cost C; maximize scope S) and the author's experiment work on small LLM (Microsoft Phi‑2) illustrate both the conceptual framing and a technological path to explore it.
← Back to All Ideas