1988 年春天，瑞典皇家科学院宣布了当年的克拉福德奖。这个奖是为了补诺贝尔奖没有覆盖的学科而设的，那一年轮到数学。获奖者两位：亚历山大·格罗滕迪克（Alexander Grothendieck），和他当年的学生皮埃尔·德利涅（Pierre Deligne）。

格罗滕迪克拒绝了。他给科学院回了一封信，这年 5 月刊在法国《世界报》上。

信不长，我读了几遍，越读越觉得，今天学术圈里让人难受的那些东西，帽子、内卷、署名之争、算力垄断，在这封信里都能找到雏形。

这篇文章想做三件事。先把他的原意讲清楚。再把这封信放进一个更大的思想谱系里看。最后说说我自己的看法。

格罗滕迪克 1928 年生在柏林。父亲是无政府主义者，后来死在奥斯维辛。战争开头那两年，他和母亲一起被关在法国南部的拘留营里。1942 年起母子分开，他被藏在山区小镇勒尚邦，在那里读完了中学。战后他才开始正经学数学。

1958 年，法国高等科学研究所（IHÉS）成立，他是最早的两位终身教授之一。此后十二年，他带着一群年轻人几乎把代数几何从地基开始重写了一遍：概形、拓扑斯、étale 上同调。讲义和讨论班记录加起来有八千多页。

如果要用一句话说他做了什么：别人在研究方程的解长什么样，他在问，“空间”这个词到底该怎么定义，才能让这些问题变成同一个问题。

1966 年，他获得菲尔兹奖，没有去莫斯科领奖。1970 年，他离开了 IHÉS。那年他四十二岁，正是一个数学家最好的年纪。

信写得很克制，开头甚至有点客气。下面按信的顺序简单复述一遍。

开头是一段礼节性的话。他说，科学院决定把今年的克拉福德奖和一笔可观的奖金授予他和德利涅，他感到荣幸。他还特意在括号里注明，德利涅曾是他的学生。但他很遗憾，他不想接受这个奖，也不接受任何别的奖。理由有三条。

第一条，他不需要钱。教授的工资，加上这年 10 月就要开始领的退休金，养活自己和家人绰绰有余。至于这个奖对他几项奠基性工作的肯定，他相信新想法是否丰饶，只有时间能检验。丰饶要看它生出了什么后代，而不是它得到了什么荣誉。

第二条，奖励流向了错误的方向。他注意到，这类有声望的大奖面向的研究者，都已经到了这样的社会位置：物质富足，声望崇高，还握有与之相伴的权力和特权。然后是全信最尖锐的一句：

第三条，他说比前两条严重得多。科学院表彰的工作是二十五年前做的，那时他还身在科学界，基本认同它的精神和价值观。1970 年他离开了，此后没有放弃对研究的热情，内心却和那个圈子越走越远。而这些年里，至少在数学家中间，职业伦理已经坏到同行之间公然窃取几乎成了通则，受害的尤其是无力自卫的人，而且所有人都容忍，哪怕是最明目张胆的情形。在这种情况下参加“奖项”的游戏，就等于认可一种他认为很不健康的风气。他还断言，这种风气在精神上是自杀性的，在智识上、物质上也是，注定很快消亡。

结尾，他先说明，讲这些不是要批评科学院管理这笔基金的用意。接着留下一个预言：他毫不怀疑，不出本世纪，一些完全意料不到的剧变会彻底改变我们对“科学”、对科学的大目标、对做科学的精神的看法。他还说，毫无疑问，到那时皇家科学院也会是在一场空前的更新里派得上用场的机构和人物之一，而这场更新会发生在一场同样空前的文明终结之后。最后，他为拒奖给科学院添的麻烦道歉。

读完全信，有三处细节值得停一下。

一是括号里那句“德利涅曾是我的学生”。看上去是客套，放在背景里读却很复杂。和他一起得奖的德利涅，正是他在长篇手稿《收获与播种》里指控“埋葬”了自己遗产的那个学生。这层关系下面还会说到。

二是“二十五年前”。奖章表彰的是 1960 年代的格罗滕迪克，那个人还认同这个共同体的价值观。写信的是 1988 年的格罗滕迪克。在拒奖者眼里，被表彰的已经是另一个人了。

三是“容忍”。他控诉的不只是有人偷，而是所有人都容忍。坏的不是几个人，而是整个圈子默认的规则。这一点后面讲马太效应和麦金太尔时，会变得更清楚。

有一点要特别注意。他离开的是那个圈子，不是数学本身。整个八十年代，他还在写数学，《追寻叠》（À la Poursuite des Champs）和《纲领草案》（Esquisse d'un Programme）都是这时候写的，后者到今天还在启发新的研究方向。

《马太福音》里有一句话：

老子把同一件事说成了人间的常态，又拿天道来对照：

1968 年，社会学家罗伯特·默顿（Robert K. Merton）在《科学》杂志上发表了一篇论文，把科学界的这种现象命名为“马太效应”。他发现，同样一项贡献，如果出自名人之手，或者名人和无名之辈合作，功劳往往会算到名人头上。

默顿并没有把这当成纯粹的坏事。他指出，在论文多到读不完的时候，名字是一种过滤器。读者先看大牛的文章，这样做并不愚蠢。马太效应是一套信息系统在注意力稀缺下的自然解。

同样是 1988 年，默顿又写了一篇续作，讨论累积优势。他在里面承认了一件事：二十年前那篇命名马太效应的论文，大量用了合作者哈丽雅特·朱克曼（Harriet Zuckerman）对诺奖得主的访谈研究，本该署两个人的名字。

命名马太效应的那篇论文，本身就是一次马太效应。

我想在这里说出第一个自己的看法。

奖项不是对过去的报酬，而是对未来的配给。 一个奖，表面上是给你过去的工作发一枚奖章，实际上是一张通行证：经费、学生、职位、邀请报告、评审席位，以及最要紧的，定义“什么问题重要”的权力。

放到今天国内的语境，这叫“帽子”。帽子从来不只是帽子。它能换项目、换编制、换招生名额，还能换一个评审别人帽子的位置。戴帽子的人决定谁能戴帽子，一个闭环就这样合上了。

格罗滕迪克的第二个理由，看到的正是这一层。他质疑的不是奖金本身，而是奖项能兑换成什么。在一个经费、名额、版面和注意力都有总量限制的系统里，“一些人的过剩只能以另一些人的必需为代价”不是修辞，是会计恒等式。

还有更隐蔽的一层。一个人起步时的小运气，比如导师是谁、第一篇文章投到哪里、赶上了哪个风口，会在累积优势下滚成巨大的差距。二十年后，差距看上去完全是能力造成的。颁奖典礼的作用，就是在这个时刻把运气追认为美德。

要把格罗滕迪克的批判想透，我觉得最好用的工具来自哲学家阿拉斯代尔·麦金太尔（Alasdair MacIntyre）。他在 1981 年的《追寻美德》（After Virtue）里讲过一个例子。

假设你想教一个很聪明的七岁孩子下国际象棋，可他对象棋没兴趣，只想吃糖。你跟他说：每下一盘就给你糖，赢了给双份。孩子开始下棋了，而且下得很努力。可只要有机会作弊而不被发现，他就有充分的理由作弊，因为他要的是糖，不是棋。

麦金太尔由此区分了两种好东西。

一种是“外在之善”：钱、名声、地位、权力。它们可以通过很多途径得到，下棋能得到，作弊也能得到。它们总是归某个人所有，一个人得到的越多，别人能得到的就越少。

另一种是“内在之善”：只有真正投身某项实践才能得到的东西。一步妙棋的美感、对局面的直觉、算清一个复杂变化时的那种通透。它们没法作弊得到，而且一个人得到了，整个下棋的共同体都跟着受益。

麦金太尔还说了一句很要紧的话（这里是转述）：任何实践都离不开制度来维持，象棋需要俱乐部，科学需要大学和实验室；但制度天然以外在之善为中心，实践的理想和创造力，永远暴露在制度的贪婪之下。

用这把尺子去量格罗滕迪克的三个理由，会发现它们其实是一件事：他在抗议外在之善对内在之善的殖民。

这里是我的第二个看法，也是我认为整个问题的根子。

科学是这种冲突最极端的一个案例。它的内在之善，知识和理解，是人类拥有的最不排他的东西。你懂了勾股定理，我懂的一分也不会少。一个证明被一万个人读懂，它的价值只会更大。

可我们组织科学的方式，用的偏偏是最排他的东西。优先权只能有一个第一。奖项每年只有那么几个。教职、帽子、顶刊版面，全是零和的。

我们在用最排他的奖励，激励最不排他的事业。 这个错配不是哪个人道德败坏造成的，它是结构性的。

再回头看格罗滕迪克说的“窃取”。理解是偷不走的。没有人能偷走你对一个定理的理解，人们能偷的永远只有署名。所以学术界的盗窃只可能发生在外在之善的层面。一个领域里抢功、挂名、压学生署名的事越多，说明那里的人越是在为糖工作，而不是为棋。

格罗滕迪克最有名的比喻，出自《收获与播种》。

假设你要打开一颗坚硬的核桃。一种办法是拿锤子和凿子，对准一点，用力敲，直到敲开为止。另一种办法是把核桃泡进一种能软化它的液体里，然后等。几个星期、几个月过去，再用手一按，壳就像熟透的牛油果一样裂开了。

他接着换了一个意象：上涨的海水。海水一点一点漫上来，你几乎觉察不到它在动，也听不到什么东西被敲碎。可是有一天你会发现，原本坚硬的岩石已经被海水包围、浸没，问题自己消失了。

他做数学用的是第二种办法。不去强攻一个具体难题，而是建立新的语言和概念，把整个地形抬高，直到原来的难题变得自然而然。

这个比喻在他自己的生涯里有一个真实的版本。1949 年，安德烈·韦伊（André Weil）提出了一组关于有限域上方程解的个数的猜想，即韦伊猜想。格罗滕迪克的大部分工作，从 étale 上同调到他设想的“标准猜想”和“动机”理论，都可以看作在为这组猜想涨海水。他想要的不只是证明，而是一个能让证明显得理所当然的世界。

1973 年，德利涅证明了韦伊猜想里最难的那一条。他用了老师建造的大量工具，但在关键处用一个巧妙的技巧绕开了标准猜想。数学界为之欢呼，格罗滕迪克据说却很失望。核桃被一把锤子敲开了，海水停在了半路。标准猜想到今天仍然没有被证明。

我的第三个看法是这样的。

我们现行的科研功劳制度，本质上是一本核桃账。功劳记在“谁第一个敲开了哪颗核桃”名下。定理有名字，猜想有名字，可海水没有名字。一个新定义、一套好记号、一个让后来者少走十年弯路的框架，它们的贡献是弥散的，是所有人的，也就不是任何人的。

在这样的账本下面，理性的研究者都会去当敲核桃的人。

把这一点想清楚，就能理解格罗滕迪克为什么在《收获与播种》里写了那么长的《葬礼》篇。他控诉的核心是：他的工具被所有人使用，他的名字却被悄悄抹去。这份私人的委屈，和他对制度的公共批判，其实是同一个结构。他是那个时代最伟大的涨海水的人，而海水恰恰是核桃账记不下来的东西。

这也解释了他的指控为什么既有洞见又有偏颇。他看见了制度的盲区，却把盲区造成的代价，读成了具体某个人的恶意。

还要替锤子说句公道话。没有核桃，也就没有海水。没有韦伊猜想这个具体的目标，就不会有 étale 上同调。海水需要一个方向，锤子也常常是用海水锻造的。德利涅的那把锤子，材料几乎全部来自格罗滕迪克的海。问题不在锤子，在账本。

《收获与播种》里有很多篇幅在写儿童。格罗滕迪克说，发现是孩子的特权。孩子不怕犯错，不怕显得愚蠢，也不知道哪些问题已经被大人宣布为“不重要”或“早就解决了”。他认为这种未经驯服的好奇，是一切创造的源头。

他还回忆了自己刚进巴黎数学圈的时候。身边有很多人比他聪明、比他学得快，能轻松地掌握他要费很大力气才弄懂的东西。可是多年以后回头看，他觉得真正让自己有所创造的，是更早那几年在蒙彼利埃一个人摸索时学会的独处：不靠公认的看法来判断什么是真的，只靠自己去看。

心理学里有一个实验，和这段话惊人地呼应。1973 年，斯坦福的莱珀（Mark Lepper）等人找来一群本来就爱画画的幼儿园孩子。一组孩子事先被告知，画了就能拿到一张漂亮的奖状；另一组没有这个许诺。一两周后，研究者在自由活动时间悄悄观察。结果是，那些曾经为奖状而画的孩子，自发画画的时间明显变少了。

这叫“过度辩护效应”：当一件你本来就喜欢的事被外在奖励解释了，你会开始觉得自己是为了奖励才做它。奖励一停，兴趣也跟着没了。

这个结论后来有过不少争论。比较稳妥的说法是，事先许诺的、和表现挂钩的奖励，最容易挤掉内在兴趣。

再看学术评价，它几乎完全由这种奖励构成。论文数、影响因子、基金、帽子，每一样都是事先公布的，每一样都和表现挂钩。我们设计了一套最容易挤掉兴趣的激励，然后问为什么那么多博士生读到第三年就不想做研究了。

麦金太尔的那个下棋的孩子，故事其实还有后半段。糖可以是一个开始。也许下着下着，孩子就爱上了棋本身，不再需要糖。糖是梯子，不是房子。我们的问题是，学术界把梯子焊死在了墙上，从本科一直架到退休。

现在回到 1970 年。格罗滕迪克离开 IHÉS，直接的原因是他发现研究所的经费里有一小部分来自法国国防部。他要求停掉这笔钱，没有成功，就走了。

这件事不是孤立的。1967 年，他去了正在遭受轰炸的北越，在疏散到乡下的大学里给学生上课。离开 IHÉS 的同一年夏天，他发起了一个叫“生存”（Survivre，后来改名 Survivre et Vivre，意为“生存与生活”）的团体，关心军备竞赛和生态危机，谢瓦莱（Claude Chevalley）等几位数学家很快加入。

1972 年 1 月，他在欧洲核子研究中心（CERN）做了一场演讲，题目叫《我们还要继续做科学研究吗？》。面对一屋子物理学家，他质疑的是当时占主流的一种信仰：科学知识是唯一真正的知识，科学的进步自然会让世界更好。他说这种“进步”是一个神话。在这之前几个月，他在团体的刊物上写过一篇《新的普世教会》，把这种信仰叫作“科学主义”。

这里有一个很深的转向。他不再问某一项具体的研究该不该做，而是问整个科学体制默认的那条公理：能做的，就应该做。

“生存与生活”几年后就散了。可他提出的问题，五十年后换了一身衣服又回来了。今天你去问一个 AI 实验室的研究员，为什么要把模型做得更强，最常听到的回答大概是：“因为别人也在做。”这不是一个理由，这是一场军备竞赛的自我描述。

讲到 AI，我想先讲另一位数学家的故事。

1994 年，菲尔兹奖得主威廉·瑟斯顿（William Thurston）写了一篇文章，叫《论数学中的证明与进步》。他回忆说，七十年代他在叶状结构这个领域里接连证明了一系列很强的定理。结果出乎他的意料：几年之内，这个领域的人几乎走光了。别人觉得问题都被他做完了，剩下的人也很难跟上他那种没有写成基础设施的证明方式。

瑟斯顿由此得出一个结论：数学的进步，衡量标准不是定理被证明了多少，而是人类对数学的理解增加了多少。一个人证明了太多定理，可能反而让一个领域的理解变少了。

现在把瑟斯顿换成机器。这已经不是假设了。光是过去两个月：

• 9 月 8 日，OpenAI 宣布，它的系统解决了千禧年难题里 NS 方程的一个版本：加上适当的光滑外力，流动可以在有限时间内爆破。上万个智能体跑了 88 小时，写出 166 页证明，附带 Lean 形式化。OpenAI 表示不领克雷研究所的奖金。核验还在进行。有预印本指出，Lean 里的证明和文字写的证明对不上；据 Gizmodo 报道，也有读者担心 Lean 验证的其实是一个比公开宣称更弱的命题。
• 10 月 5 日，哥伦比亚大学的 Josh Alman 和 MIT 的 Virginia Vassilevska Williams 挂出预印本，给出约 n^1.9992 时间的 3SUM 算法。“3SUM 不可能有本质上快于 n² 的算法”是细粒度复杂性的地基之一，这下被推翻了，APSP 假设也连带倒下。论文写明，算法是 Anthropic 一个内部研究版的 Claude 发现的：一位员工让它检验一个密码学构造，它在一次无人干预的会话里反而找到了这个算法。Anthropic 9 月在保密协议下把算法交给两位作者，并提出给报酬；作者负责弄懂、简化、加强和推广，写成论文。论文写完以后，Anthropic 又用内部模型在 Lean 里核验了主要结果。
• 10 月 6 日晚，OpenAI 一次性把 722 篇数学手稿放上 GitHub，分成 372 个结果族。其中只有 235 个配了 Lean 说明页，形式化的审核状态一栏写着 unchecked。
• 10 月 26 日，OpenAI 又宣布了三项结果：数学上证明了全维数的局部光滑化猜想，从而推出挂谷猜想；计算机科学上推翻了指数时间假设（ETH）；物理上证明了四维时空里一类带规范场的量子场论存在。这一次它换了说法：模型的产出最好被理解为“原始研究数据”，要请外部专家当“牧羊人”，组织学者把它们变成科学理解；公司不求学术功劳，参与者也不用签保密协议。

AI 是人类造出过的最强的锤子。给它一颗定义清楚的核桃，它会以工业化的规模去敲，而且敲的已经不是奥赛题，是一个领域的地基。

这会把格罗滕迪克的问题推到一个前所未有的位置。

第一个后果关于理解。如果做科学的目的是得到定理，那它可以外包。如果目的是理解，就外包不了。理解是一种心智的状态，不是一件产品。就像你不能雇别人替你跑步来提高自己的心肺功能。一个机器证明了、但没有任何人真正理解的定理，在任何一个人的头脑里都没有涨起一寸海水。

瑟斯顿那个叶状结构的故事，在 AI 时代可能变成整个学科的故事：问题被敲完了，人走光了，留下一堆正确却无人理解的真理。当然，另一种未来同样可能：机器去敲核桃，人终于可以专心涨海水。哪一种会发生，取决于我们那本账怎么记。

第二个后果关于马太效应。研究者的能力和他能调动的 AI 资源之间，很可能是互补的：资源越多，能力越能发挥。如果更好的成果又能换来更多的资源，正反馈就形成了。原本就存在的差距会被放大，而且放大得更快。

这里我想说一个容易被忽略的区别。过去马太效应的硬通货是声望，声望再不公平，至少还得由同行授予。算力这种货币，同行说了不算。 它只认钱。当科研最关键的生产资料从“同行的承认”变成“机构的资本”，格罗滕迪克那句“一些人的过剩只能以另一些人的必需为代价”，就从一句道德质问变成了一张 GPU 排期表。

第三个后果关于署名，这也是我觉得最有格罗滕迪克味道的地方。

OpenAI 10 月 26 日那套说法，几乎是在用公司的语言复述核桃与海水：机器负责敲开核桃，产出“原始数据”；把数据变成理解，要靠人。连造锤子的人都承认，海水外包不了。

可是谁来当牧羊人？由公司挑。OpenAI 自己也承认，先选定一批人会造成信息不对称。3SUM 那边，是公司在保密协议下把算法交给两位学者，并提出给报酬。在这两种安排里，“涨海水”从一项自由的探索，变成了一份由机器主人分派的差事。

“不求功劳”听上去很慷慨，有点像格罗滕迪克拒奖。但别忘了他的第二条理由：奖只是权力的一种兑换形式。放弃奖章和功劳，并不等于放弃权力。在机器发现的时代，最值钱的外在之善不再是署名，而是先看到的权利：谁先拿到原始数据，谁就先有机会理解，先写出论文，先占住下一片海。这种权利不由同行评议授予，由模型的主人发放。

我觉得以后需要分开谈三件事：

• 知道的权利：能读到别人发现了什么。
• 发现的权利：能亲自参与发现。
• 塑造的权利：能影响科学往哪里走。

AI 可能会让第一种权利空前扩张，人人都能读懂最前沿的结论。但第二种和第三种，完全可能同时变得更加集中。论文免费公开，不等于研究机会平等。人人都有一个 AI 账号，也不等于人人都能用上同样强大的研究系统。

开放模型、便宜的推理、公共的算力和可以独立检验的证明记录，有可能削弱这种集中。这件事今天还没有定论。但它至少说明，“AI 会不会让科学更不平等”是一个可以认真研究的问题，而不只是一种情绪。

现在我要说一点格罗滕迪克没有说、甚至可能没意识到的东西。

再看一遍他的第一个理由：我不缺钱。这句话本身就是特权的证明。

翻一翻历史上那些著名的拒奖者。萨特在 1964 年拒绝了诺贝尔文学奖，理由是作家不该让自己变成一个机构。佩雷尔曼在 2006 年拒绝了菲尔兹奖，2010 年又拒绝了克雷研究所为庞加莱猜想颁发的百万美元奖金，据报道他认为汉密尔顿的贡献并不比自己少。

他们拒绝的时候，都已经站在了顶端，或者早已对这个游戏毫无所求。

能拒绝奖的人，恰恰是最不需要奖的人。真正需要被这套制度解放的人，一个三十五岁、合同还剩两年的青年教师，没有拒绝它的资本。对他来说，拒绝不是一种可选的策略。

所以个人的英雄主义解决不了这个问题。格罗滕迪克的拒奖是一个姿态，一个照亮问题的姿态，但它不是答案。

还有一个他自己的生涯里藏着的反讽。他一生最有创造力的十二年，恰恰是在一个机构里度过的。IHÉS 给了他一份不用申请项目、不用考核、不用证明自己“有用”的稳定薪水。批判制度最狠的人，他最好的工作，是一个替他挡住了竞争的制度养出来的。

这其实指出了方向。如果问题出在外在之善控制了通往内在之善的入口，那么解法就是把入口拆开：让思考的时间不必靠竞争去赢。

这方面已经有一些值得注意的实验。新西兰的健康研究委员会从 2013 年起，探索性资助先由评审筛出够格的申请，再抽签决定资助谁。瑞士国家科学基金会试点三年后，从 2021 年起，在卡在资助线上、难分高下的申请之间抽签。

抽签听上去很荒唐，可它比奖章诚实。在资助线附近，排第 20 名和第 40 名的申请，常常分不出高下。抽签承认了运气的存在。奖章把运气洗成美德，抽签把运气还给运气。

另一个方向是改账本，让海水也能入账。新定义、好讲义、开源的工具、形式化证明库里一行一行写进去的基础引理，这些都是海水。今天它们在评价体系里几乎一文不值。只要账本不变，理性的人就会继续去敲核桃。

这几年国内一直在说“破四唯”，破唯论文、唯职称、唯学历、唯奖项。方向是对的。可是破指标最常见的办法，是再列一张新的指标清单。用一张表去破一张表，只是把糖换了个包装。

写到这里，我必须给格罗滕迪克本人泼一点冷水。

他对昔日同行，尤其是对德利涅的指控，充满了个人情绪，很多数学家并不认同。不少人认为，德利涅恰恰是最尊重、最认真继承他遗产的人之一。

他也不是一个始终站在制度外面的纯粹反抗者。六十年代的他，就是那个数学世界的中心。大量讲义是学生和合作者替他执笔整理的。他后来批判的那种权力，他自己也曾经拥有过。

1991 年以后，他隐居在比利牛斯山区的一个小村子里，几乎切断了和外界的一切联系，写了大量带有神秘主义色彩的手稿。2010 年，他写信要求不得再出版或传播他的任何著作。一个说“想法的价值要看它的后代”的人，最后试图收回自己所有的后代。

他的家庭、他和身边人的关系，也远非可以拿来做榜样的样子。

还有一个更微妙的问题，值得每个读到这里的人想一想。我们之所以认真对待他的批判，很大程度上是因为他拿过菲尔兹奖。一个无名的讲师写出一封一模一样的信，不会登上《世界报》，今天也不会有人写文章重读它。

马太效应连对马太效应的批判也不放过。

所以我不想把他塑造成一个圣人。他留给我们的价值，不在于他的人生可以被模仿，而在于他提出的问题，至今没有人能绕过去。

格罗滕迪克在拒奖信里预言，科学的目标和精神会经历一场深刻的变化。三十多年后，变化确实来了，只是和他想的可能不太一样。它不是来自觉醒的良知，而是来自一台台机器。

AI 会把一个原本只属于哲学课的问题，变成一个真实的预算问题：我们做科学，到底是为了得到更多正确的结论，还是为了让更多的人理解这个世界？

如果是前者，机器会做得比我们好，我们可以放心地把锤子交出去。

如果是后者，那么有一些事情无论如何都不能外包：一个孩子第一次自己想通一个问题时的那种通透，一群人花十年时间把海水慢慢抬高的那种耐心，以及判断哪些问题值得去理解的那份自由。

我并不担心未来的机器比人更接近真理。我担心的是，我们造出了一个能无限接近真理的文明，却让越来越少的人有机会亲自参与这场探索。

天之道，损有余而补不足。科学本来有机会成为人间最接近天之道的那件事，因为它最珍贵的东西，从来不怕被分享。

别让它变成又一件“损不足以奉有余”的事。

In the spring of 1988, the Royal Swedish Academy of Sciences announced that year's Crafoord Prize. The prize was set up for fields the Nobel Prizes leave out, and that year it was mathematics' turn. It went to two people: Alexander Grothendieck, and Pierre Deligne, who had once been his student.

Grothendieck refused. He wrote back to the Academy, and his letter appeared in the French newspaper Le Monde that May.

The letter is short. I have read it several times, and the more I read it, the more I feel that everything that makes academic life so hard to bear today can already be found there in embryo: what Chinese academia calls “hats” (talent titles), the rat race, the fights over authorship, the monopolies on compute.

This essay sets out to do three things. First, to make clear what he actually meant. Then, to place the letter in a larger family of ideas. Finally, to say what I think myself.

Grothendieck was born in Berlin in 1928. His father was an anarchist who later died at Auschwitz. For the first two years of the war, he and his mother were held together in an internment camp in the south of France. From 1942 mother and son were separated, and he was hidden in Le Chambon-sur-Lignon, a small town in the mountains, where he finished secondary school. Only after the war did he begin to study mathematics properly.

In 1958 the Institut des Hautes Études Scientifiques (IHÉS) was founded in France, and he was one of its first two permanent professors. Over the next twelve years, he and a group of young mathematicians rewrote algebraic geometry almost from the foundations up: schemes, topoi, étale cohomology. The lecture notes and seminar records add up to more than 8,000 pages.

If you had to say in one sentence what he did: other people were studying what the solutions of equations look like, while he was asking how the word “space” ought to be defined so that all those questions would become a single question.

In 1966 he was awarded the Fields Medal, and did not go to Moscow to receive it. In 1970 he left IHÉS. He was forty-two, right in a mathematician's prime.

The letter is restrained, and its opening is even rather polite. Here is a brief paraphrase, in the letter's own order.

It opens with a few courteous words. He says he is honoured that the Academy has decided to award this year's Crafoord Prize, with a substantial sum of money, to him and Deligne. He takes care to note in parentheses that Deligne was once his student. But he regrets that he does not wish to accept this prize, or any other prize. He gives three reasons.

First, he does not need the money. His professor's salary, plus the pension he will start drawing that October, is more than enough to support himself and his family. As for the prize's recognition of some of his foundational work, he is convinced that only time can test whether new ideas are fertile. Fertility is judged by the offspring an idea has, not by the honours it receives.

Second, the rewards flow in the wrong direction. He observes that prestigious prizes of this kind go to researchers who have already reached a social position with material plenty and high prestige, along with the power and privileges that come with them. Then comes the sharpest sentence in the whole letter:

The third reason, he says, is far more serious than the first two. The work the Academy is honouring was done twenty-five years earlier, when he was still part of the world of science and broadly shared its spirit and values. He left that world in 1970. Since then he has kept his passion for research, but inwardly he has drifted further and further from that milieu. And in those years, at least among mathematicians, professional ethics have decayed so far that outright theft among colleagues has become almost the rule, above all at the expense of those who cannot defend themselves, and everyone tolerates it, even in the most flagrant cases. Under these conditions, taking part in the game of “prizes” would mean endorsing a climate he considers deeply unhealthy. He also declares that this climate is suicidal: spiritually, and intellectually and materially too. It is doomed to die out before long.

At the end, he first makes clear that none of this is meant as a criticism of the Academy's aims in administering the fund. Then he leaves a prophecy. He has no doubt that before the century is out, wholly unforeseen upheavals will utterly transform our notion of “science”, of its great aims, and of the spirit in which science is done. He adds that by then the Royal Academy will no doubt be among the institutions and people with a useful part to play in an unprecedented renewal, one that will follow an equally unprecedented end of civilization. Finally, he apologizes for the trouble his refusal has caused the Academy.

Once you have read the whole letter, three details are worth a pause.

First, the remark in parentheses that Deligne was once his student. It looks like a courtesy, but read against the background it is anything but simple. Deligne, who shared the prize with him, was the very student Grothendieck accused, in his long manuscript “Récoltes et Semailles” (Reaping and Sowing), of having “buried” his legacy. We will come back to this relationship below.

Second, the twenty-five years. The medal honours the Grothendieck of the 1960s, a man who still shared this community's values. The letter was written by the Grothendieck of 1988. In the eyes of the man refusing it, the person being honoured was already somebody else.

Third, tolerance. What he condemns is not only that some people steal, but that everyone puts up with it. What has gone bad is not a few individuals but the rules the whole community takes for granted. This will become clearer later, when we come to the Matthew effect and MacIntyre.

One thing deserves special attention. What he left was that milieu, not mathematics itself. All through the 1980s he was still writing mathematics. Pursuing Stacks (À la Poursuite des Champs) and Esquisse d'un Programme (Sketch of a Programme) both date from those years, and the latter still inspires new lines of research today.

There is a verse in the Gospel of Matthew:

Laozi described the same thing as the normal state of human affairs, and set it against the way of heaven:

In 1968 the sociologist Robert K. Merton published a paper in the journal Science that gave this phenomenon in the scientific world a name: “the Matthew effect”. He found that when the same contribution comes from a famous scientist, or from a famous scientist working with an unknown one, the credit tends to go to the famous name.

Merton did not treat this as a purely bad thing. He pointed out that when there are more papers than anyone can read, a name works as a filter. Reading the big names first is not foolish. The Matthew effect is the natural solution for an information system when attention is scarce.

Also in 1988, Merton wrote a sequel on cumulative advantage. In it he admitted something. The paper that had named the Matthew effect twenty years earlier drew heavily on his collaborator Harriet Zuckerman's interview study of Nobel laureates, and it should have carried both their names.

The paper that named the Matthew effect was itself a case of the Matthew effect.

Here I want to offer the first view of my own.

A prize is not payment for the past but an allocation of the future. On the surface, a prize gives you a medal for work you have already done. In reality it is a pass: to funding, students, positions, invited talks, seats on review panels, and, most important of all, the power to define “which problems matter”.

In China today, this is called a “hat”. A hat is never just a hat. It can be traded for grants, for a permanent post, for student places, and even for a seat on the panels that decide other people's hats. Those who wear hats decide who gets to wear hats, and so the loop closes.

Grothendieck's second reason sees exactly this. What he questions is not the prize money itself but what a prize can be converted into. In a system where funding, positions, journal space and attention are all limited in total, “the over-abundance of some can only come at the expense of the necessities of others” is not rhetoric. It is an accounting identity.

There is a subtler layer still. Small pieces of luck at the start of a career, such as who your advisor is, where your first paper goes, or which wave you happen to catch, snowball under cumulative advantage into huge gaps. Twenty years later, the gap looks entirely like a matter of ability. What an award ceremony does is step in at exactly this moment and certify the luck, after the fact, as virtue.

To think Grothendieck's critique all the way through, the most useful tool, I think, comes from the philosopher Alasdair MacIntyre. In his 1981 book After Virtue he gives an example.

Suppose you want to teach a very bright seven-year-old to play chess, but he has no interest in chess and only wants candy. You tell him: every game you play earns you candy, and a win earns you double. The child starts playing, and plays hard. But whenever he gets a chance to cheat without being caught, he has every reason to cheat, because what he wants is the candy, not the chess.

From this MacIntyre distinguishes two kinds of goods.

The first kind is “external goods”: money, fame, status, power. They can be had in many ways: by playing chess, and by cheating too. They are always someone's property, and the more one person has, the less there is for everyone else.

The other kind is “goods internal to a practice”: things you can only get by truly giving yourself to a practice. The beauty of a brilliant move, a feel for a position, the clarity of calculating a complicated line all the way through. They cannot be had by cheating, and when one person gains them, the whole community of chess players benefits.

MacIntyre makes one more point that matters a great deal (paraphrased here). No practice can survive without institutions to sustain it: chess needs clubs, science needs universities and laboratories. But institutions are by nature centred on external goods, and the ideals and creativity of a practice are always exposed to the greed of its institutions.

Measure Grothendieck's three reasons with this ruler and they turn out to be one thing: he is protesting against the colonization of internal goods by external ones.

This is my second view, and what I take to be the root of the whole problem.

Science is the most extreme case of this conflict. Its internal goods, knowledge and understanding, are the least exclusive things humanity has. If you understand the Pythagorean theorem, I understand it not one bit less. A proof understood by ten thousand people is only worth more.

Yet the way we organize science runs on the most exclusive things there are. Priority allows only one first. There are only a handful of prizes each year. Faculty posts, hats, pages in top journals: all of them are zero-sum.

We use the most exclusive rewards to drive the least exclusive enterprise. This mismatch is not caused by anyone's moral failings. It is structural.

Now go back to the theft Grothendieck spoke of. Understanding cannot be stolen. No one can steal your understanding of a theorem; all anyone can ever steal is the credit. So theft in academia can only happen at the level of external goods. The more a field is full of grabbing credit, adding the names of people who did no work, and squeezing students off the author list, the more its people are working for the candy, not the chess.

Grothendieck's most famous metaphor comes from “Récoltes et Semailles”.

Suppose you want to open a hard walnut. One way is to take a hammer and chisel, aim at one point, and strike hard until it cracks. The other way is to soak the walnut in a liquid that softens it, and then wait. Weeks or months later, a press of the hand is enough, and the shell opens like a perfectly ripe avocado.

Then he switches to another image: the rising sea. The water creeps in little by little. You can hardly tell it is moving, and you hear nothing being smashed. Yet one day you find that the rock that was so hard has been surrounded and submerged, and the problem has disappeared on its own.

His way of doing mathematics was the second. Instead of storming one particular hard problem, he built new language and new concepts that raised the whole landscape, until the original hard problem fell out naturally.

This metaphor has a real version in his own career. In 1949 André Weil proposed a set of conjectures about the number of solutions of equations over finite fields: the Weil conjectures. Most of Grothendieck's work, from étale cohomology to the “standard conjectures” and the theory of “motives” he envisioned, can be seen as raising the sea for these conjectures. He wanted more than a proof. He wanted a world in which the proof would seem a matter of course.

In 1973 Deligne proved the hardest of the Weil conjectures. He used a great deal of the machinery his teacher had built, but at the crucial point he got around the standard conjectures with an ingenious trick. The mathematical world cheered; Grothendieck, it is said, was disappointed. The nut had been cracked with a hammer, and the sea had stopped halfway. The standard conjectures remain unproved to this day.

My third view is this.

Our current system of research credit is, at bottom, a ledger of nuts. Credit is entered under “who first cracked which nut”. Theorems have names, conjectures have names, but the sea has no name. A new definition, a good notation, a framework that saves those who come later ten years of detours: their contribution is diffuse. It belongs to everyone, and so to no one.

Under a ledger like this, every rational researcher becomes a nut-cracker.

Think this through and you can see why Grothendieck wrote such a long part of “Récoltes et Semailles”, the part called “L'Enterrement” (The Burial). The heart of his complaint is that everyone used his tools while his name was quietly erased. This private grievance and his public critique of the institution have, in fact, the same structure. He was the greatest raiser of the sea in his time, and the sea is exactly what a ledger of nuts cannot record.

This also explains why his accusations are both insightful and lopsided. He saw the institution's blind spot, but he read the cost of that blind spot as the malice of particular people.

I should also put in a fair word for the hammer. Without nuts there would be no sea. Without the concrete target of the Weil conjectures, there would be no étale cohomology. The sea needs a direction, and hammers are often forged from the sea. Nearly all the material in Deligne's hammer came from Grothendieck's sea. The problem is not the hammer. It is the ledger.

A good deal of “Récoltes et Semailles” is about children. Grothendieck says that discovery is the privilege of the child. A child is not afraid of making mistakes or of looking foolish, and does not know which questions the grown-ups have already declared “unimportant” or “settled long ago”. He believed this untamed curiosity is the source of all creation.

He also recalls his first years in the Paris mathematical world. Many people around him were cleverer than he was and learned faster, picking up with ease what cost him great effort to understand. Yet looking back many years later, he felt that what had truly let him create was the solitude he had learned earlier, in the years he spent finding his own way alone in Montpellier: judging what is true not by received opinion, but only by looking for himself.

A psychology experiment echoes this passage strikingly. In 1973 Mark Lepper and his colleagues at Stanford brought together a group of nursery school children who already loved drawing. One group was told in advance that if they drew, they would get a nice certificate; the other group was promised nothing. A week or two later, the researchers quietly watched the children during free play. The children who had once drawn for the certificate now spent noticeably less time drawing of their own accord.

This is called “the overjustification effect”: when something you already enjoy comes to be explained by an outside reward, you start to feel that you do it for the reward. Once the reward stops, the interest goes too.

This finding has been much debated since. The safer statement is that rewards promised in advance and tied to performance are the most likely to crowd out intrinsic interest.

Now look at academic evaluation. It is made almost entirely of rewards like these. Paper counts, impact factors, grants, hats: every one is announced in advance, and every one is tied to performance. We have designed the incentives most likely to crowd out interest, and then we ask why so many PhD students no longer want to do research by their third year.

MacIntyre's chess-playing child actually has a second half to his story. Candy can be a start. Perhaps, as he keeps playing, the child will come to love chess itself and stop needing the candy. Candy is a ladder, not a house. Our problem is that academia has welded the ladder to the wall, all the way from undergraduate years to retirement.

Now back to 1970. The immediate reason Grothendieck left IHÉS was that he found out a small part of the institute's funding came from the French Ministry of Defence. He demanded that this money be stopped. When that failed, he left.

This was not an isolated episode. In 1967 he went to North Vietnam while it was being bombed, and taught students at a university that had been evacuated to the countryside. In the summer of the year he left IHÉS, he founded a group called Survivre (“Survive”; later renamed Survivre et Vivre, “Survive and Live”), concerned with the arms race and the ecological crisis. Claude Chevalley and several other mathematicians soon joined.

In January 1972 he gave a talk at CERN, the European Organization for Nuclear Research, titled “Allons-nous continuer la recherche scientifique ?” (Shall we go on with scientific research?). Facing a room full of physicists, he challenged a belief that was dominant at the time: that scientific knowledge is the only true knowledge, and that scientific progress will naturally make the world better. This “progress”, he said, was a myth. A few months earlier he had written an article in the group's journal, “La nouvelle Église universelle” (The New Universal Church), which called this belief “scientism”.

This marks a deep turn. He no longer asks whether some particular piece of research should be done. He questions the axiom the whole scientific establishment takes for granted: what can be done should be done.

Survivre et Vivre broke up a few years later. But the question he raised has come back fifty years on, in different clothes. Ask a researcher at an AI lab today why they are making models more powerful, and the answer you will hear most often is probably “Because others are doing it too.” That is not a reason. It is an arms race describing itself.

To talk about AI, I want to start with a story about another mathematician.

In 1994 the Fields medalist William Thurston wrote an essay called “On Proof and Progress in Mathematics.” He recalled that in the 1970s he proved a string of strong theorems, one after another, in the field of foliations. What followed surprised him: within a few years, people had almost all left the field. Others felt he had already done all the problems, and those who stayed found it hard to follow his style of proof, which was never written up as infrastructure.

From this Thurston drew a conclusion. Progress in mathematics is measured not by how many theorems get proved, but by how much human understanding of mathematics grows. One person proving too many theorems can actually leave a field with less understanding.

Now put a machine in Thurston's place. This is no longer hypothetical. In the past two months alone:

• On September 8, OpenAI announced that its system had solved a version of the Navier–Stokes problem, one of the Millennium Prize Problems: with a suitable smooth external force, the flow can blow up in finite time. Over ten thousand agents ran for 88 hours and produced a 166-page proof, with a Lean formalization. OpenAI said it would not take the Clay Mathematics Institute's prize money. Verification is still under way. A preprint has pointed out that the proof in Lean and the proof written in prose do not match; according to Gizmodo, some readers also worry that what Lean verified is actually a weaker statement than the one publicly claimed.
• On October 5, Josh Alman of Columbia University and Virginia Vassilevska Williams of MIT posted a preprint giving a 3SUM algorithm that runs in about n^1.9992 time. “No algorithm for 3SUM is substantially faster than n²” is one of the foundations of fine-grained complexity. Now it has been overturned, and the APSP hypothesis falls with it. The paper states that the algorithm was found by an internal research version of Anthropic's Claude: an employee asked it to check a cryptographic construction, and in a single session with no human intervention it found this algorithm instead. In September, Anthropic gave the algorithm to the two authors under a non-disclosure agreement and offered them compensation; the authors' part was to understand, simplify, strengthen and generalize it, and write the paper. After the paper was written, Anthropic used an internal model to check the main results in Lean.
• On the evening of October 6, OpenAI put 722 mathematical manuscripts on GitHub in one go, grouped into 372 result families. Only 235 of these came with a Lean description page, and there the review status of the formalization reads “unchecked.”
• On October 26, OpenAI announced three more results: in mathematics, a proof of the local smoothing conjecture in all dimensions, which implies the Kakeya conjecture; in computer science, a refutation of the Exponential Time Hypothesis (ETH); in physics, a proof that a class of quantum field theories with gauge fields exists in four-dimensional spacetime. This time it changed its framing. The model's output is best understood as “raw research data,” and outside experts are to be brought in as “shepherds,” organizing scholars to turn it into scientific understanding. The company does not seek academic credit, and participants do not have to sign non-disclosure agreements.

AI is the most powerful hammer humans have ever built. Give it a well-defined nut and it will crack away at industrial scale. And it is no longer cracking olympiad problems. It is cracking the foundations of a field.

This pushes Grothendieck's questions further than they have ever gone.

The first consequence concerns understanding. If the point of doing science is to get theorems, it can be outsourced. If the point is understanding, it cannot. Understanding is a state of mind, not a product. You cannot hire someone to go running for you and expect your own heart and lungs to get fitter. A theorem that a machine has proved, but that no one truly understands, has not raised the sea a single inch in anyone's head.

In the age of AI, Thurston's story about foliations could become the story of whole disciplines: the problems all cracked, the people all gone, and left behind a heap of truths that are correct but understood by no one. Of course, another future is just as possible: machines crack the nuts, and people can at last devote themselves to raising the sea. Which one happens depends on how we keep our ledger.

The second consequence concerns the Matthew effect. A researcher's ability and the AI resources they can command are likely to be complements: the more resources, the more the ability can do. If better results in turn bring in more resources, a positive feedback loop forms. Gaps that already existed get amplified, and amplified faster.

Here I want to point out a distinction that is easy to miss. In the past, the hard currency of the Matthew effect was prestige. However unfair prestige may be, it at least has to be granted by peers. Compute is a currency your peers have no say over. It answers only to money. When the most important means of production in research shifts from “recognition by peers” to “institutional capital,” Grothendieck's line “the over-abundance of some can only come at the expense of the necessities of others” turns from a moral challenge into a GPU schedule.

The third consequence concerns authorship, and this is the part that feels most like Grothendieck to me.

OpenAI's framing of October 26 all but retells the walnut and the rising sea in corporate language: machines crack the nuts and produce “raw data”; turning the data into understanding takes people. Even the people who build the hammer admit the sea cannot be outsourced.

But who gets to be a shepherd? The company picks. OpenAI itself admits that choosing a group of people first creates an information asymmetry. On the 3SUM side, it was the company that gave the algorithm to two scholars under a non-disclosure agreement and offered them compensation. In both arrangements, “raising the sea” turns from a free exploration into a job handed out by the owners of the machine.

“Not seeking credit” sounds generous, a bit like Grothendieck turning down the prize. But remember his second reason: a prize is only one form power can be cashed out in. Giving up medals and credit is not the same as giving up power. In the age of machine discovery, the most valuable external good is no longer authorship but the right to see first: whoever gets the raw data first gets the first chance to understand it, to write the paper, to claim the next stretch of sea. This right is not granted by peer review. It is handed out by the owners of the models.

I think from now on we will need to talk about three things separately:

• The right to know: being able to read what others have discovered.
• The right to discover: being able to take part in discovery yourself.
• The right to shape: being able to influence where science goes.

AI may expand the first right as never before, so that everyone can understand results at the very frontier. But the second and the third may well become more concentrated at the same time. Papers being free to read does not make research opportunities equal. Everyone having an AI account does not mean everyone can use equally powerful research systems.

Open models, cheap inference, public compute and proof records that can be checked independently could weaken this concentration. The question is not settled today. But it at least shows that “will AI make science more unequal?” is a question that can be studied seriously, not just a mood.

Now I want to say something Grothendieck did not say, and may not even have realized.

Look again at his first reason: I don't need the money. That sentence is itself proof of privilege.

Run through history's famous refusers. Jean-Paul Sartre turned down the Nobel Prize in Literature in 1964, on the grounds that a writer should not let himself be turned into an institution. Grigori Perelman turned down the Fields Medal in 2006, and in 2010 he turned down the Clay Mathematics Institute's million-dollar prize for the Poincaré conjecture; he reportedly felt that Richard Hamilton's contribution was no smaller than his own.

When they refused, they were all already at the top, or had long since stopped wanting anything from the game.

The people who can refuse a prize are exactly the ones who need it least. The person who really needs to be freed from this system, a thirty-five-year-old junior lecturer with two years left on their contract, cannot afford to refuse it. For them, refusing is not an available strategy.

So individual heroics will not solve this problem. Grothendieck's refusal was a gesture, one that lit up the problem, but it was not an answer.

There is also an irony hidden in his own career. The twelve most creative years of his life were spent inside an institution. IHÉS gave him a steady salary with no grant applications, no evaluations, no need to prove himself “useful.” The man who attacked the system most fiercely did his best work in the care of a system that shielded him from competition.

This actually points the way. If the problem is that external goods control the gateway to the goods internal to a practice, then the solution is to take the gateway apart: time to think should not have to be won in a competition.

There are already some experiments worth noticing. Since 2013, the Health Research Council of New Zealand has handled its exploratory grants this way: reviewers first pick out the applications that are good enough, then a lottery decides which of them get funded. The Swiss National Science Foundation (SNSF), after a three-year pilot, has since 2021 used a lottery to choose among applications stuck right at the funding line that are too close to call.

A lottery sounds absurd, but it is more honest than a medal. Near the funding line, the application ranked 20th and the one ranked 40th often cannot be told apart. A lottery admits that luck exists. A medal launders luck into virtue; a lottery gives luck back to luck.

The other direction is to change the ledger, so that the sea can go on the books too. New definitions, good lecture notes, open-source tools, the basic lemmas written line by line into libraries of formal proofs: all of this is sea. Today it is worth almost nothing in the evaluation system. As long as the ledger stays the same, rational people will keep cracking nuts.

For several years now, China has been talking about the campaign to “break the four onlys”: no more judging only by papers, only by titles, only by degrees, only by awards. The direction is right. But the most common way to break one set of metrics is to draw up a new list of metrics. Breaking one checklist with another just puts the candy in a new wrapper.

Having come this far, I have to pour a little cold water on Grothendieck himself.

His accusations against former colleagues, and against Deligne above all, are full of personal feeling, and many mathematicians do not accept them. Quite a few believe that Deligne was in fact one of the people who most respected his legacy and carried it on most faithfully.

Nor was he a pure rebel who always stood outside the system. In the 1960s he was the center of that mathematical world. Much of his lecture material was written up by students and collaborators on his behalf. The kind of power he later attacked, he once held himself.

After 1991 he lived in seclusion in a small village in the Pyrenees, cut off almost all contact with the outside world, and wrote a great many manuscripts tinged with mysticism. In 2010 he wrote a letter demanding that none of his works be published or distributed any further. A man who said an idea is judged by its offspring ended up trying to take back all of his own offspring.

His family life, and his relationships with the people around him, were also far from anything you would hold up as a model.

There is a subtler problem too, one worth thinking over for everyone who has read this far. We take his critique seriously largely because he won a Fields Medal. If an unknown lecturer wrote exactly the same letter, it would not appear in Le Monde, and no one today would be writing an essay rereading it.

The Matthew effect does not spare even the critique of the Matthew effect.

So I don't want to make him into a saint. What he left us is valuable not because his life can be imitated, but because no one has yet found a way around the questions he raised.

In his letter of refusal, Grothendieck predicted that the goals and the spirit of science would go through a profound change. More than thirty years later, the change has indeed come, though perhaps not quite as he imagined. It has come not from an awakened conscience but from machine after machine.

AI will turn a question that once belonged only in philosophy class into a real budget question: do we do science to get more correct conclusions, or so that more people can understand the world?

If it is the former, machines will do it better than we can, and we can hand over the hammer without worry.

If it is the latter, then some things cannot be outsourced, whatever happens: the clarity a child feels the first time they work something out on their own; the patience of a group of people who spend ten years slowly raising the sea; and the freedom to judge which questions are worth understanding.

I am not worried that future machines will get closer to the truth than people do. What worries me is that we will have built a civilization that can come ever closer to the truth, while fewer and fewer people get the chance to take part in the search themselves.

The way of heaven takes from what has too much and gives to what has too little. Science had the chance to be the one human pursuit closest to the way of heaven, because what it treasures most has never had anything to fear from being shared.

Let's not let it become one more case of “taking from those who have too little to serve those who have too much.”