AI科研的验证困境：为何RLVR可能难以突破大科学发现

导语

在编程、数学等具有快速、明确验证闭环的领域，强化学习已展现出强大威力。然而，将同样的方法应用于科学发现，可能面临根本性挑战。科学进步的历史表明，理论验证的周期往往极其漫长，且一个“更好”的理论在提出之初，可能还不如现有理论的预测准确。本文以哥白尼日心说、海王星与祝融星的发现、以及化学元素原子量的争论为例，剖析了科学验证的复杂性，并探讨了这对AI驱动的科学研究意味着什么：大概念性突破难以被即时训练，而科学进步仍需依赖个体研究者带有偏见的长期坚持。

金句

• "The verification loop for theories can be on the order of decades and centuries, and even then we know today as the better theory can often actually make worse predictions." —— 理论的验证周期可能长达数十年甚至数世纪，而且即便如此，我们如今知道，一个更好的理论在初期常常做出更差的预测。
• "In some sense, [Copernicus's theory] wasn't [better]! You couldn't have known ex ante that heliocentrism married with Kepler's 3 laws (1619) is a much cleaner and more accurate theory..." —— 在某种意义上，哥白尼的理论在1543年并不算更好！你无法先验地知道日心说与开普勒三大定律结合后是一个更简洁、更精确的理论……
• "What I'm trying to say is that ex ante, one couldn't have known which research program would be more productive. We need to invest in all of them concurrently." —— 我想说的是，事前无法预知哪个研究方向会更富成效。我们需要同时对它们进行投资。
• "Big conceptual breakthroughs cannot be easily verified. They are recognized decades or centuries later, when it turns out they were much more productive than the alternatives available." —— 重大的概念性突破无法被轻易验证。它们在数十年甚至数世纪后才被认可，因为事实证明它们比当时的其他替代方案更具生产力。
• "What this means for AI for science is that 1. You can’t easily train an RL loop for big conceptual breakthroughs." —— 这对AI科学应用意味着：第一，你无法轻易为重大的概念性突破训练一个强化学习循环。

原文链接

查看完整访谈