图灵发表《论可计算数》

1936 年，图灵发表了一篇题为《论可计算数及其在判定问题上的应用》的论文。论文要回答的数学问题是：是否存在一种机械方法，能自动判定任何一个数学命题的真假？图灵的回答是否定的，而为了证明这个否定，他发明了一个思想工具——图灵机。

图灵机不是一台实体机器。它是一个抽象模型：一条无限长的纸带，一个读写头，一组简单规则。图灵证明，只要规则足够清晰，这种机器就能执行任何可计算的运算。反过来，那些无法用图灵机完成的问题，就是「不可判定」的。这个模型第一次把「什么能被机械计算」定义得清清楚楚。

今天看来，这篇论文的意义已经远超数学本身。它证明了「通用机器」的可能性——一台机器可以模拟任何其他机器，只要给它正确的程序。这个思想直接构成了数字计算机的理论地基，也让「机器能否思考」从一个哲学猜想，变成了可以严格讨论的科学问题。

十四年后，图灵发表了那篇更著名的《计算机器与智能》，用「模仿游戏」（后来被称为图灵测试）来讨论机器是否能思考。从图灵机到图灵测试，图灵用一篇文章定义了计算的边界，用另一篇文章开启了智能的追问。这两篇论文，是整个人工智能史共同的起点。

回看《论可计算数》，它的分量不在于解决了某个具体问题，而在于为一个时代提供了坐标。当 1943 年的 McCulloch-Pitts 神经元、1956 年的达特茅斯会议、2017 年的 Transformer 沿着这条时间轴展开时，1936 年那张纸带上的读写头，始终是所有故事的源头。

In 1936 Turing published a paper titled "On Computable Numbers, with an Application to the Entscheidungsproblem." The mathematical question it answered: does a mechanical method exist that can automatically decide the truth of any mathematical proposition? Turing's answer was no, and to prove that negative he invented a thinking tool—the Turing machine.

The Turing machine is not a physical computer. It is an abstract model: an infinitely long paper tape, a read-write head, and a small set of simple rules. Turing proved that as long as the rules are clear enough, this machine can perform any computable operation. Conversely, problems such a machine cannot complete are "undecidable." The model defined, for the first time, exactly what "can be mechanically computed" means.

Today the paper's significance reaches far beyond mathematics. It proved the possibility of a "universal machine"—one machine can simulate any other, given the right program. That idea directly became the theoretical foundation of digital computers, and turned "can machines think" from a philosophical guess into a rigorously discussable scientific question.

Fourteen years later, Turing published the more famous "Computing Machinery and Intelligence," using the "imitation game" (later called the Turing test) to ask whether machines can think. With one paper he defined the boundary of computation, and with another he opened the inquiry into intelligence. Together, these two papers are the shared origin of the entire history of AI.

Looking back at "On Computable Numbers," its weight lies not in solving a particular problem but in providing coordinates for an era. As McCulloch–Pitts neurons in 1943, the Dartmouth workshop in 1956, and the Transformer in 2017 unfold along this timeline, the read-write head on that 1936 paper tape remains the source of all the stories.