现代人工智能代理结合Lean等形式化证明助手,实际上已经实现了大卫·希尔伯特当年设想的“证明机器”功能 1。这一技术进展与数学基础计划的历史背景形成鲜明对照,因为库尔特·哥德尔的不完备性定理证明了不存在完备且一致的公理系统,从而击碎了希尔伯特的数学基础计划;而艾伦·图灵对停机问题不可判定性的证明,则宣告了自动决定任意数学陈述真伪算法的不可能性 1。然而,这种形式化验证能力的提升也带来了新的认知危机,正如冯·诺依曼所预言,若不可判定性失效,数学将变为纯粹的机械处方,导致现有形式的数学不复存在 1。
当前计算机科学社区面临的核心风险在于逐渐放弃对AI生成内容的深层理解,特别是普遍存在的停止阅读AI生成代码的现象 1。根据柯里-霍华德对应原理(Church-Howard correspondence),程序即证明,因此忽略AI生成的代码等同于忽视AI产生的数学证明,这被批评为对计算机科学追求彻底理解这一核心精神的背叛 1。文章呼吁计算机科学家应通过抽象和重构来保持对代码逻辑的理解,重申“我们必须知道”的精神,以防止人类失去对科学本质的掌控 1。
Modern artificial intelligence, particularly when combined with formal proof assistants like Lean, has effectively realized the "proof machine" concept envisioned by mathematician David Hilbert 1. This development stands in stark contrast to the historical failure of Hilbert’s foundational program, which aimed to establish a complete and consistent axiomatic system for mathematics 1. That program was famously dismantled by Kurt Gödel’s incompleteness theorems, which proved that no such system could exist, and Alan Turing’s demonstration that the halting problem is undecidable, thereby confirming the impossibility of an algorithm capable of automatically determining the truth or falsity of any mathematical statement 1.
Despite these theoretical limits, the emergence of AI agents capable of generating verified proofs suggests a shift toward what John von Neumann predicted would happen if undecidability were somehow bypassed: mathematics would reduce to pure mechanical prescriptions, potentially erasing existing forms of the discipline 1. The author argues that this technical success carries a profound epistemic risk, drawing on the Church-Howard correspondence, which establishes that programs are equivalent to proofs 1. Consequently, the growing trend among computer scientists to stop reading AI-generated code is not merely a convenience but a betrayal of the field’s core pursuit of thorough understanding 1. If the community fails to engage deeply with the logic behind these automated outputs through abstraction and refactoring, humanity risks losing control over the essence of scientific knowledge, violating the spirit of Hilbert’s motto, "We must know – we will know" 1.
评论
还没有评论,欢迎留下第一条。