THE ECONOMIST · OCTOBER 10TH 2026 SCIENCE & TECHNOLOGY · AI AND SCIENCE
SCIENCE & TECHNOLOGY · SCIENCE AND ARTIFICIAL INTELLIGENCE

尤里卡之后

Eureka!

What if the machines take over science?

栏目:Science & technology(科学与技术 · 科学与人工智能) | 体裁:分析评论 | 正文约 950 词 | 难度:CET-6 ~ IELTS 7.5
2026 年 10 月 6 日,OpenAI 一次性放出 700 多篇论文,宣称解开了 372 个困扰学界数十年的难题。《经济学人》没有停留在「AI 真厉害」,而是把它当成一个认识论问题来写:如果证明由机器产出、人类只负责消化,那么我们到底还需不需要「理解」?

这篇 11 段的评论,最值得学的是它把一件技术新闻写成了思想评论。全篇围绕一对矛盾推进:进步(progress)与理解(understanding)。它先给三条并行定性,再承认数学家的愤怒有理,接着一句 special pleading(为自身利益辩护)把对方的立场连根拔起;随后又掉头承认「真正的担忧并非无的放矢」,最后落到一条克制的主张上:a human in the loop(永远保留一个在环的人)。

写作上可拿走三招:一是「三重定性」(一句话里同时给出好处、分量与警告:a leap ... a marvel ... a warning);二是「先接底牌再反驳」(Even if ... align ... fully trusted, a society where ... is one in which humans will become the machines’ pets.);三是「日常类比」(把作业答案、扑热息痛这两件人人有经验的小事,用来解释抽象的认识论问题)。
01 全文呈现 02 逐段精讲 + 全文翻译 03 核心词汇精讲 04 长难句拆解 05 写作积累 06 篇章结构与文化冷知识

01全文呈现

原文按自然段落编号 ①–⑪,保留原刊用词、拼写与标点。

1OCTOBER 6TH 2026 may be remembered as the biggest day in the history of mathematics. OpenAI, an American frontier lab, released more than 700 academic papers claiming progress or solutions for 372 problems that had baffled scholars for decades. Mathematicians are still struggling to digest what it means for their discipline.

2The implications promise to go far beyond maths. It is time to grapple with how artificial intelligence (AI) will affect science as a whole. The maths results are, simultaneously, a leap forward for humanity; a marvel that underscores the breakneck progress of AI; and a warning of a future in which understanding of the universe could belong to machines that are impenetrable to human minds. The question is how to achieve progress without sacrificing understanding.

3AIs have so vividly shown their scientific potential in maths because its proofs are derived from axioms, not evidence, and are therefore, in principle, verifiable mechanically. But as we report in our special Science & technology section this week, AI has also accelerated progress in the empirical sciences, partly because other fields use maths as a tool. AI has made weather forecasts better. It has proposed millions of possible new materials and helped control experiments with nuclear fusion. And it can predict the shapes of almost every known protein from its constituent amino acids, an advance that won a Nobel prize.

4More remarkable, AI-fuelled scientific advances are almost certain. The most exciting possibilities are in biology. Researchers are using AI to design drought-resistant crops, to search for new drugs and to create (we must hope) friendly viruses that attack harmful bacteria. There is an effort to use AI to build computer models of how cells work. Armed with life in silico, scientists may go on to achieve mastery of biology, gaining the ability to create entirely new organisms from scratch. If so, DNA would become just another form of computer code to be manipulated like a string of tokens in an AI model.

5Much like the maths, that prospect is giddying and terrifying in equal measure. The first reaction of many mathematicians is to be terrified. Since OpenAI’s breakthroughs, they have been up in arms at the lab’s encroachment into their discipline. The Association for Human Mathematicians, a group of over 900 scholars, warns that “mathematicians did not ask for this work to be done.” They say that the pay-off in doing maths comes not from the proof, but from the toil, because that is what advances human understanding. Having AIs churn out incomprehensible but correct solutions is like giving students the answers to their homework: without effort, there is no learning.

6These complaints contain an element of special pleading. Academies have no more right to dictate the production of knowledge than any other profession. Truth-seeking benefits everyone and cannot be reserved for a guild of scholars. The hierarchies of academies, climbed by proving things, are not sacred. If needs be, they must be remade for a world in which AI produces proofs and humans digest them. In science, understanding should not be fetishised over results. Every time you swallow a paracetamol pill to see off a headache, you rely on a mechanism that works but is not fully understood.

7Yet the fear of handing over science to the machines is not an idle one. AI companies are ploughing resources into science not out of altruism but because, as with humans, solving problems can help make models more intelligent. Many researchers think that a model that gets better at solving scientific problems will get better at doing everything else, too. If so, then nobody will be able to compete with big AI labs, which will eventually control models that work on improving themselves and become, by some definitions, incomprehensibly intelligent. These genius AI scientists will connect to the physical world either through humanoid robots or by hooking up models to lab equipment, as Anthropic has already done.

8The problem this future poses for science is much the same as the one it poses for society. People will have to provide resources to machines for them to continue to make progress. If no one understands what they are doing, simply writing the cheques would be reckless. Nobody could be sure that the models were operating in humanity’s best interests. Even if AI labs manage to “align” machines with the interests of humanity, so that they can be fully trusted, a society where people simply ask AIs what to do—in science, or more broadly—is one in which humans will become the machines’ pets.

9Some people may be willing to accept that if it means, say, curing all diseases. But at least until there is unshakeable alignment between AIs and humans, science will always need a human in the loop. Labs should be able to use AIs to solve problems, but resources such as laboratories and computing power should be directed and monitored by people who can reach an informed judgment about what AIs are doing. This will, inevitably, slow progress. But it is necessary if AIs are to remain under human control.

10In the meantime, human-guided machine-science will require scholars to reinvent their disciplines. Scientists will have to learn to love—and credit with status and career advancement—guiding models and studying, verifying and teaching their results. Governments will need to allocate research grants on the same basis. For their part, AI labs should seek to be good scientific citizens, submitting their work for peer review and publishing their methods and code.

11Checking and digesting a computer’s results may seem like unglamorous work. Much human-produced science is error-strewn because scholars are not rewarded for policing the work of others. But a renewed emphasis on verification and scepticism would improve science. It is worth remembering the Royal Society’s motto, adopted in the 17th century: Nullius in verba, or “Take nobody’s word for it”. That same spirit now needs to apply to science made by the machines.

02逐段对照精讲 + 全文翻译

① 事件:2026 年 10 月 6 日,700 篇论文与 372 个难题

2026 年 10 月 6 日,或许会被记为数学史上最重要的一天。美国前沿实验室 OpenAI 发布了 700 多篇学术论文,宣称在 372 个困扰学者数十年的难题上取得了进展或给出了解答。数学家们至今仍在费力消化:这对本学科究竟意味着什么。

开篇是典型的「日期 + 最高级」新闻式起手:OCTOBER 6TH 2026 may be remembered as the biggest day in the history of mathematics. 注意作者用 may be remembered(或许会被记住)而不是 is(是)——把判断交给未来,既不夸口又足够有分量,这是《经济学人》评价重大事件时的标准分寸。第二句用两个具体到个位的数字建立可信度:more than 700 academic papers 与 372 problems;an American frontier lab(一家美国前沿实验室)是同位语,frontier lab 是 AI 业内的固定说法(指做最前沿模型的那几家);claiming progress or solutions(宣称取得进展或给出解答)里的 claiming 隐含「尚待学界确认」的保留;that had baffled scholars for decades(困扰学者数十年)用过去完成时标明「AI 出现之前的状态」,baffle(难住、使困惑)比 puzzle 更重。末句 struggling to digest what it means for their discipline 是全篇的一处伏笔:digest(消化、领会)与 discipline(学科)都是关键词,而且 digesting 在末段(Checking and digesting a computer’s results)再次出现——首尾用同一个词呼应,是这篇文章藏得很深的一条暗线。

② 放大:从数学到整个科学,一句话三重定性

其影响注定远远超出数学。现在该认真应对一个问题了:人工智能(AI)将如何影响科学整体。这批数学成果同时是三重东西:人类的一次飞跃;一件凸显 AI 突飞猛进之速的奇迹;以及一个警告——未来对宇宙的理解,可能归属于人类心智无法穿透的机器。问题在于:如何在不牺牲理解的前提下取得进步。

本段把话题从数学放大到整个科学,并给出全文的「三重定性」。首句 The implications promise to go far beyond maths 用 promise to(注定会、有望)作预测;第二句 It is time to grapple with ...(该认真应对……了)是评论文章的经典过渡句,grapple with(与……扭打,引申为努力应对难题)比 deal with 有分量。第三句是全篇的骨架句:The maths results are, simultaneously, a leap forward for humanity; a marvel that underscores the breakneck progress of AI; and a warning of a future in which ...——一个主语、三个表语,用分号隔开(因为每一项内部都带从句,逗号压不住),顺序从褒到贬:a leap(飞跃)→ a marvel(奇迹)→ a warning(警告)。breakneck(快得能折断脖子的)比 rapid 形象得多;impenetrable to human minds(人类心智无法穿透的)中 impenetrable 本义「不可进入」,此处指「无法理解」。末句 The question is how to achieve progress without sacrificing understanding 只有十一个词,却是全文的论点句:progress 与 understanding 这对矛盾,此后每一段都在它上面做文章。学会这一招:把全篇的争论压缩成「A 与 B 能否兼得」,读者的注意力立刻被钉住。

③ 解释:为什么数学先被攻破

AI 之所以在数学上如此鲜明地展现了科学潜力,是因为数学证明由公理推出、而非由证据归纳,因此原则上可以被机械地验证。但正如本刊本周「科学与技术」专题所报道的,AI 同样加速了经验科学的进展,部分原因是其它领域把数学当作工具。AI 让天气预报更准了;它提出了数百万种可能的新材料,并帮助控制核聚变实验;它还能根据氨基酸构成预测几乎所有已知蛋白质的形状——这项进展曾赢得诺贝尔奖。

本段做两件事:解释「为什么数学先被攻破」,再列举经验科学的战绩。首句用 because its proofs are derived from axioms, not evidence 一句说清数学与经验科学的方法论差异:数学靠公理(axiom)演绎,经验科学靠证据归纳;are derived from(由……推出)是被动语态,not evidence 是插入的对比成分。in principle(原则上)四个字很要紧——留了余地:理论上可机械验证,不等于实践里已经做到。第二句用 But 转折,把话题推到数学之外:partly because other fields use maths as a tool 点出关键机制——数学是共同工具,所以数学一强,别的学科跟着受益。随后是三个极短的具体战果句:AI has made weather forecasts better. / It has proposed millions of possible new materials ... / And it can predict the shapes of almost every known protein ...——天气预报、新材料、核聚变、蛋白质结构,全是可核查的事实,没有一句抽象形容词。注意这种「短句清单」的用法:论证「AI 不只是玩具」时,四个实例比一句「AI 很强大」有说服力得多。末句以同位语 an advance that won a Nobel prize(一项赢得诺奖的进展)收尾——用第三方荣誉为自己背书,比自己说「了不起」高明;这也是英文学术评论的常规操作。

④ 展望:生物学最激动人心,DNA 变成另一串代码

更值得注意的是,由 AI 驱动的科学突破几乎已成定局,而最令人兴奋的可能性在生物学。研究者正用 AI 设计抗旱作物、搜寻新药,并制造(但愿如此)能攻击有害细菌的「友好病毒」。还有人在努力用 AI 建立细胞如何运作的计算机模型。有了「硅基生命」这一武装,科学家或将继续推进,最终掌握生物学——获得从零创造全新生物体的能力。若真如此,DNA 就不过是另一种形式的计算机代码,可以像 AI 模型里的一串词元那样被随意摆弄。

本段把视线推向生物学。开头 More remarkable(更值得注意的是)用比较级承接上段,语气往上加一档;are almost certain(几乎已成定局)比 will 更硬,但 almost 又留了余地。第二句直接点出方向:The most exciting possibilities are in biology. 第三句用三个不定式并列铺开研究现状:to design drought-resistant crops, to search for new drugs and to create (we must hope) friendly viruses that attack harmful bacteria——注意括号里的 (we must hope)(但愿如此)是社论的冷幽默:病毒并不天然「友好」,所以作者顺手插一句自嘲,这类括号插话能让严肃文章透一口气,也是《经济学人》的招牌笔法。drought-resistant(抗旱的)用 -resistant 后缀构词(fire-resistant、water-resistant 同理)。第四句 There is an effort to ...(有人在努力……)是英语里最省力的「介绍一项工作」句式。第五句是全段的高点:Armed with life in silico, scientists may go on to achieve mastery of biology——Armed with(以……为武装)用过去分词短语作状语,把「计算机里的生命模型」写成一件武器;in silico(在硅中)与实验室里的 in vitro(试管内)、in vivo(活体内)是一组拉丁语学术用语,一律斜体;achieve mastery of(掌握)比 master 更书面;gaining ... 是现在分词作结果状语,from scratch(从零开始)是必背短语。末句 If so, DNA would become just another form of computer code to be manipulated like a string of tokens in an AI model 用 If so(若真如此)作虚拟推论,并把 DNA 比作 computer code、把碱基比作大模型的 tokens(词元)——用读者最熟悉的新事物去解释旧事物,是本文反复使用的类比策略。

⑤ 反弹:数学家为何「群情激愤」

与数学上的情形相似,这一前景既令人眩晕,也同样令人恐惧。许多数学家的第一反应是恐惧。自 OpenAI 取得突破以来,他们一直对这家实验室「侵入」自己的学科群情激愤。一个由 900 多名学者组成的「人类数学家协会」警告说:「数学家并没有要求别人来做这项工作。」他们说,做数学的回报不来自证明,而来自辛苦的钻研——因为正是这个过程推进了人类的理解。让 AI 大量产出无法理解却正确的解答,就像把作业答案直接给学生:没有付出,就没有学习。

本段转入反弹,是全文第一个需要认真对付的反对意见。首句 giddying and terrifying in equal measure(既令人眩晕,也同等程度地令人恐惧)——in equal measure 是地道表达,表示两种感受分量相当;giddying 从 giddy(头晕)来,形容「快得让人站不稳」。第二、三句写数学家的反应:up in arms(群情激愤、准备抗争,源自「拿起武器」);encroachment into their discipline(对本学科的侵夺)——encroach on 本用于领土或权利,此处把「AI 进入数学」写成一场入侵,选词即立场。第四句引 The Association for Human Mathematicians(人类数学家协会)的警告,a group of over 900 scholars 是同位语,用人数给出分量;引语 “mathematicians did not ask for this work to be done.” 用被动完成时,暗示「没人征询过我们的意见」。第五句是本段的核心论证:the pay-off ... comes not from the proof, but from the toil(回报不来自证明,而来自苦工)——not ... but ... 把「结果」与「过程」对立起来,pay-off(回报、收益)与 toil(辛劳)押着相近的音,读起来有一点格言的意味。末句用作业类比收束:Having AIs churn out incomprehensible but correct solutions is like giving students the answers to their homework: without effort, there is no learning. churn out(大量粗制滥造地生产)带明显贬义;冒号后的 without effort, there is no learning 只有七个词,节奏斩截——先给比喻、再用一句格言点破,是英文说理段最经济的收尾法。

⑥ 反驳:这是「为自身利益所作的辩护」

这些抱怨里有一部分是「为自身利益所作的辩护」。学术团体没有比其它任何职业更多的权力去支配知识的生产。求真惠及所有人,不能被保留给一个学者行会。学术界的等级秩序——靠证明东西爬上去的——并非神圣不可改。如有必要,为了一个「AI 产出证明、人类消化证明」的世界,它们必须被重塑。在科学中,理解不该被抬到高于结果的位置去盲目崇拜。每当你吞下一片扑热息痛压住头痛,你依赖的就是一套有效却没被完全弄懂的机制。

本段是全文最锋利的一段反驳,四步推进:定性 → 去神圣化 → 重塑 → 反证。第一步只用六个词:These complaints contain an element of special pleading.(这些抱怨带有为自身利益辩护的成分)——special pleading 是法律与辩论术语,指「为自己的立场设一套例外标准」,一上来就给对方的动机定性,比逐条驳论点更省力。第二步用 no more ... than 句式去神圣化:Academies have no more right to dictate the production of knowledge than any other profession.(学术团体并不比其它职业更有权支配知识生产)——no more ... than(并不比……更……)是「双向否定」的高分结构;dictate(支配、规定)与后文 guild of scholars(学者行会)配合,一个词就带出「中世纪封闭团体」的意味。第三步两句极短、语气陡然变硬:The hierarchies of academies ... are not sacred. / If needs be, they must be remade for a world in which AI produces proofs and humans digest them.——if needs be(如有必要,古雅说法)、remade(重塑)、digest(消化)三个词把「等级秩序」与「新的知识分工」对立起来。第四步转向「理解至上论」:In science, understanding should not be fetishised over results.——fetishise(把……当作神物崇拜)批判性极强,一句话把对手的「理解神圣论」贬为「拜物」。末句用最日常的例子收尾:Every time you swallow a paracetamol pill to see off a headache, you rely on a mechanism that works but is not fully understood.(你吃扑热息痛,靠的是一套有效却没被完全弄懂的机制)——see off(赶走、打发)是口语化短语;这一击之所以漂亮,是因为它把「不理解也能用」从例外变成常态:如果吃药可以不懂药理学,那么用证明为什么必须懂证明?

⑦ 真正的担忧:模型在为自己变强而做科学

然而,把科学交给机器的担忧并非无的放矢。AI 公司把资源大量投进科学,并非出于利他,而是因为——与人类一样——解题能让模型变得更聪明。许多研究者认为,一个在解决科学问题上越来越强的模型,在其它一切事情上也会越来越强。若果真如此,那就没人能与大型 AI 实验室竞争;这些实验室最终将掌控能够自我改进的模型,并按某些定义变得「不可理解的智能」。这些天才的 AI 科学家会接入物理世界:或通过人形机器人,或像 Anthropic 已经做的那样,把模型接到实验设备上。

本段是全文的重心转折:Yet 之后,作者承认对方的担忧「有道理」。首句 the fear ... is not an idle one——idle 此处不是「懒散的」,而是「无根据的」,not an idle fear(并非无端之忧)是极地道的书面表达。第二句 ploughing resources into science not out of altruism but because ... 用 not out of A but because B 揭穿动机:plough into(把大量资金投入,plough 本义「犁地」)是英式拼法(美式 plow),altruism(利他主义)与「逐利」形成反差,这一句把「AI 公司造福科学」的叙事翻了个面:它们做科学是为了让自己更聪明。第三、四句构成一条递推链:解题能力↑ → 通用能力↑ → 无人能竞争 → 模型自我改进 → 不可理解的智能。If so, then ...(若如此,那么……)是合理论证的标准推进词;become, by some definitions, incomprehensibly intelligent 里 by some definitions(按某些定义)是精妙的限定——作者并不背书「超级智能」这个词,只说「按某些定义如此」,既传递了量级又不失严谨。末句 hooking up models to lab equipment(把模型接到实验设备上)配 humanoid robots(人形机器人)把抽象的「AI 科学家」拉回物理世界,as Anthropic has already done 用一个真实案例让设想落地——「预言 + 一个已经发生的例子」,比纯预言可信得多。

⑧ 镜像:科学的问题,就是社会的问题

这一未来给科学带来的问题,与社会面临的问题几乎相同。人们必须为机器提供资源,它们才能继续取得进展。如果没人懂它们在做什么,只管签支票就是鲁莽。谁也无法确信这些模型是在为人类的最佳利益行事。即便 AI 实验室成功把机器与人类利益「对齐」,使其完全可信,一个人们只须问 AI「该怎么做」的社会——在科学上如此,在更广泛的层面亦如此——也将是人类沦为机器宠物的社会。

本段是全文的「镜像结构」:把科学问题上升为社会问题,一句 The problem this future poses for science is much the same as the one it poses for society 完成并置——poses a problem for(给……带来问题)是固定搭配,much the same as(与……几乎相同)留了一点余地。第二句 provide resources to machines for them to continue to make progress 用 for sb to do 复合结构说明「资源换进展」的交易关系。第三句 simply writing the cheques would be reckless(只管签支票就是鲁莽)——write the cheques(签支票、出钱)是极形象的借代;reckless(不计后果的)与上一篇社论的用词重合,可见这是该刊评价「放手不管」的常用词。第四句 were operating in humanity’s best interests 用虚拟语气(were,而非 were 陈述),因为这是「无法确知」的假设。第五句是全篇最重的一句,结构值得拆开看:Even if AI labs manage to “align” machines with the interests of humanity, so that they can be fully trusted,(让步:即便对齐成功、完全可信)a society where people simply ask AIs what to do(长定语)—in science, or more broadly—(破折号插入,把结论从科学扩到全社会)is one in which humans will become the machines’ pets.——先把对方最强的一步(对齐)接住,再说「即便如此」仍有问题,这比直接否定有力得多;而 pets 一词极重:不是「被统治」,而是「被豢养」——衣食无忧,但不再作主。

⑨ 底线:永远需要一个「在环的人」

有些人或许愿意接受这一点,只要它意味着——比如说——治愈所有疾病。但至少在对齐变得牢不可破之前,科学永远需要一个「在环的人」。实验室应当能用 AI 解题,但实验室、算力等资源应由能够对 AI 的作为作出知情判断的人来指挥与监督。这不可避免地会拖慢进展。但若要让 AI 留在人类掌控之下,这是必需的。

本段给出底线主张,全文的核心术语在此出现:a human in the loop(在环的人)。首句 Some people may be willing to accept that if it means, say, curing all diseases 先承认「全交给机器」的诱惑:say 作插入语(比如说),把「治愈百病」这一极端好处轻描淡写地摆出来,反而显出这种交换的荒唐。第二句 But at least until there is unshakeable alignment ..., X will always need ... 是本文最值得背的句式:at least until(至少在……之前)划出时间边界,unshakeable(不可动摇的)比 firm 重得多,will always need 则把主张钉死。注意这种「有限定的坚决」:不喊永远,只说「在此之前」——这正是评论文章保持可信度的技巧。第三句用 but 把两件事分开:Labs should be able to use AIs to solve problems(可以用 AI 解题)与 resources ... should be directed and monitored by people(但资源须由人指挥监督)——「用」与「控」分离,是本文政策主张的核心:不是禁用 AI,而是治理 AI。computing power(算力)是当下政策讨论的关键资源;who can reach an informed judgment(能作出知情判断的人)用定语从句给「人」下了定义:不是「有人签字」,而是「有人真懂」,informed(有依据的、知情的)是全句的题眼。末两句承认代价再给出理由:This will, inevitably, slow progress. But it is necessary if AIs are to remain under human control.——先认账(会拖慢),再说必须(否则失控),这种「先让一步再站定」的写法,比一味强硬更可信。

⑩ 制度:学者、政府与实验室各该做什么

与此同时,以人为引导的机器科学将要求学者重塑自己的学科。科学家必须学会爱上「引导模型、研究并核验其结果、教授这些结果」这件事——并以地位与职业晋升来奖励它。政府也需要按同样的标准分配科研经费。至于 AI 实验室,则应争当「好科学公民」:把成果提交同行评审,公开方法与代码。

本段最「务实」:讲制度怎么改。human-guided machine-science(以人为引导的机器科学)用连字符造出一个复合定语,把全文主张压缩成一个名词。第二句结构精巧,值得细看:Scientists will have to learn to love—and credit with status and career advancement—guiding models and studying, verifying and teaching their results.——主干是 learn to love doing sth,love 的宾语是四个动名词(guiding / studying / verifying / teaching);一对破折号插入了 and credit with status and career advancement,与 love 并列,即「学会爱上、并以地位与晋升去奖励」。credit sb with sth(把……归功于某人、给予肯定)是关键搭配。这一插入语的分量在于:只讲情怀、不给奖励,制度改不动——观念与激励必须同时改。第三句 Governments will need to allocate research grants on the same basis(按同样标准分配科研经费)把改革从个人推向体制,on the same basis(以同样的依据)回应上句的「奖励标准」。末句 For their part, AI labs should seek to be good scientific citizens, submitting ... and publishing ...——For their part(至于它们那一方)是转换视角的过渡语;good scientific citizens(好科学公民)用「公民」隐喻要求企业承担公共责任;peer review(同行评审)与 publishing their methods and code(公开方法与代码)都是学术共同体的最低门槛——作者对企业只提「入行门槛」而不提「额外负担」,分寸拿捏得很准。

⑪ 收尾:洗澡水要倒,婴儿要留

检查并消化计算机给出的结果,看起来像是不起眼的工作。人类产出的科学之所以错漏不少,恰恰因为学者并不因「监督他人的工作」而获奖赏。但重新强调核验与怀疑,会改善科学。值得记住英国皇家学会 17 世纪采用的会训:Nullius in verba,意为「不轻信任何人的话」。同样的精神,如今需要用在由机器做出的科学上。

末段呼应小标题 The baby and the bathwater(不要把婴儿和洗澡水一起倒掉):核验虽不起眼,却不能连同它一起丢掉。首句 unglamorous work(不起眼的工作)语气克制;第二句是全篇最坦诚的自我批评:Much human-produced science is error-strewn because scholars are not rewarded for policing the work of others.——error-strewn(错误百出的)由 strew(散落)的过去分词构成,比 full of errors 更有画面;policing the work of others(监督他人的工作)用 police 作动词,暗示学术共同体本应承担「治安」职能。这句之所以有力,是因为它把问题从「机器不可信」转向「人类学术激励本来就有缺陷」——批评对方之前先自我批评,立场立刻变得公允。第三句 But a renewed emphasis on verification and scepticism would improve science 用 renewed(重新兴起的)暗示这不是新发明,而是恢复旧传统,与下句的「17 世纪」自然衔接。第四、五句以引语收尾:It is worth remembering the Royal Society’s motto, adopted in the 17th century: Nullius in verba, or “Take nobody’s word for it”.——motto(座右铭、会训)、adopted(采用);最后 That same spirit now needs to apply to science made by the machines. 用 That same spirit now needs to apply to ...(同样的精神如今须用之于……)完成「旧精神—新对象」的平移:拿一句 17 世纪的格言为 AI 时代的新问题背书,既显得有历史纵深,又避免了自己造口号。这是英文评论最稳的结尾方式之一。

03核心词汇精讲

标签说明:四级 基础必备 六级 进阶 雅思 学术高频 高分 写作加分

A. 科学的语言:证据与证明

axiom/ˈæksiəm/六级雅思
释义n. 公理;公认的道理,自明之理
派生axiomatic adj. 公理的;不言自明的;axiomatically adv. 公理式地
搭配proofs are derived from axioms 证明由公理推出;accept sth as an axiom 把某事当作公理
verifiable/ˈverɪfaɪəbl/雅思高分
释义adj. 可核实的,可验证的
派生verify v. 核实;verification n. 核验;veracity n. 真实性
搭配verifiable mechanically 可机械验证;independently verifiable 可独立验证
empirical/ɪmˈpɪrɪkl/六级雅思
释义adj. 经验的,实证的(依据观察或实验而非理论的)
派生empiricism n. 经验主义;empirically adv. 以实证方式;empiricist n. 经验论者
搭配the empirical sciences 经验科学;empirical evidence 实证证据
constituent/kənˈstɪtjuənt/雅思高分
释义adj. 构成的,组成的;n. 成分,要素;(选区)选民
派生constitute v. 构成;constituency n. 选区,支持群体;constitution n. 宪法,构成
搭配its constituent amino acids 组成它的氨基酸;the constituent parts of sth 某物的组成部分
discipline/ˈdɪsəplɪn/四级雅思
释义n. 学科;纪律;v. 惩戒,训练
派生disciplinary adj. 纪律的;interdisciplinary adj. 跨学科的;disciplined adj. 自律的
搭配what it means for their discipline 对本学科意味着什么;academic disciplines 各学科
in silico/ɪn ˈsɪlɪkəʊ/雅思高分
释义adv./adj. 在计算机上(进行)的,计算机模拟的(与 in vitro、in vivo 相对)
派生silicon n. 硅;对比 in vitro 试管内、in vivo 活体内
搭配life in silico 硅基(模拟)生命;in silico experiments 计算机模拟实验
organism/ˈɔːɡənɪzəm/四级雅思
释义n. 生物体,有机体
派生organic adj. 有机的;organise v. 组织;organisation n. 组织
搭配create new organisms from scratch 从零创造新生物;a living organism 活的生物体
mastery/ˈmɑːstəri/六级雅思
释义n. 精通,掌握;控制,支配
派生master v. 掌握;n. 大师;masterful adj. 精湛的;mastery of 对……的掌握
搭配achieve mastery of biology 掌握生物学;mastery of a skill 对某项技能的掌握

B. 争议与反弹:批评的词汇

giddying/ˈɡɪdiɪŋ/雅思高分
释义adj. 令人眩晕的;(速度、变化)快得让人头晕的
派生giddy adj. 头晕的;giddily adv. 头晕目眩地;giddiness n. 眩晕
搭配giddying and terrifying 既令人眩晕又令人恐惧;a giddying pace 令人眩晕的速度
up in arms/ʌp ɪn ɑːmz/雅思高分
释义习语 群情激愤,准备抗争(源出「拿起武器」)
派生arms n. 武器;arm v. 武装;disarm v. 解除武装
搭配be up in arms at/over sth 因某事群情激愤;the whole town was up in arms 全城哗然
encroachment/ɪnˈkrəʊtʃmənt/雅思高分
释义n. 侵占,蚕食;(对权利、领土的)侵犯
派生encroach v. 侵占(encroach on/upon);encroaching adj. 步步进逼的
搭配the lab’s encroachment into their discipline 实验室对本学科的侵夺;encroachment on privacy 对隐私的侵犯
special pleading/ˌspeʃl ˈpliːdɪŋ/雅思高分
释义n. 为自身利益所作的辩解;双重标准(要求别人守规则,自己例外)
派生plead v. 恳求;辩护;plea n. 恳请;抗辩
搭配contain an element of special pleading 带有一点自我辩护的成分;dismiss sth as special pleading 斥之为自我辩护
fetishise/ˈfetɪʃaɪz/雅思高分
释义v. 盲目崇拜,把……神化(美式拼法 fetishize)
派生fetish n. 神物,迷恋物;fetishism n. 拜物教;fetishisation n. 神化
搭配understanding should not be fetishised 理解不该被神化;fetishise results 惟结果是从
guild/ɡɪld/雅思高分
释义n. 行会,同业公会(中世纪手工业者的组织)
派生guildhall n. 市政厅,行会会馆;guildsman n. 行会成员
搭配a guild of scholars 学者行会;the merchants’ guild 商人行会
dictate/dɪkˈteɪt/六级雅思
释义v. 支配,规定;口述,听写;n. 命令,原则
派生dictation n. 听写;dictator n. 独裁者;dictatorial adj. 专断的
搭配dictate the production of knowledge 支配知识的生产;dictate terms 提出条件
sacred/ˈseɪkrɪd/四级雅思
释义adj. 神圣的,不可侵犯的
派生sacredness n. 神圣;sacrifice n./v. 牺牲;sacrament n. 圣事
搭配are not sacred 并非神圣不可动;a sacred cow 不容批评的事物

C. 智能与对齐:AI 时代的行话

breakneck/ˈbreɪknek/雅思高分
释义adj. 极快的(快到危险的)
派生break one’s neck 拼命赶;breakneck speed 极速
搭配the breakneck progress of AI AI 突飞猛进;at breakneck speed 以极快的速度
impenetrable/ɪmˈpenɪtrəbl/雅思高分
释义adj. 无法进入的;无法理解的,晦涩的
派生penetrate v. 穿透;penetration n. 穿透;penetrable adj. 可穿透的
搭配impenetrable to human minds 人类心智无法穿透的;an impenetrable argument 难懂的论证
alignment/əˈlaɪnmənt/雅思高分
释义n. 对齐(AI 与人类价值一致);结盟;校准,排列成线
派生align v. 使对齐;aligned adj. 一致的;non-alignment n. 不结盟
搭配align machines with the interests of humanity 使机器与人类利益对齐;unshakeable alignment 牢不可破的对齐
humanoid/ˈhjuːmənɔɪd/雅思高分
释义adj. 人形的;n. 人形机器人
派生human adj. 人的;后缀 -oid 意为「像……的」(android 人形机、asteroid 类星体)
搭配humanoid robots 人形机器人;a humanoid figure 人形轮廓
altruism/ˈæltruɪzəm/雅思高分
释义n. 利他主义,无私
派生altruistic adj. 利他的;altruistically adv. 无私地;altruist n. 利他者
搭配not out of altruism 并非出于利他;pure altruism 纯粹的利他
unshakeable/ʌnˈʃeɪkəbl/雅思高分
释义adj. 不可动摇的,坚定不移的(美式拼法 unshakable)
派生shake v. 摇动;shaky adj. 不稳的;shake-up n. 大改组
搭配unshakeable alignment 不可动摇的对齐;an unshakeable belief 坚定的信念
in the loop/ɪn ðə luːp/雅思高分
释义习语 在决策圈内,知情并参与;a human in the loop 人在回路中(保留人类把关)
派生loop n. 环,回路;loophole n. 漏洞;looping n. 循环
搭配a human in the loop 一个在环的人;keep sb in the loop 让某人知情
reinvent/ˌriːɪnˈvent/四级雅思
释义v. 彻底改造,重新塑造
派生reinvention n. 重塑;invent v. 发明;reinvent the wheel 重复发明轮子(白费力气)
搭配reinvent their disciplines 重塑自己的学科;reinvent oneself 脱胎换骨

D. 通用高分表达:动词与形容

grapple with/ˈɡræpl wɪð/六级雅思
释义短语 努力应对(难题);与……搏斗
派生grapple v. 扭打,搏斗;grappling n. 角力
搭配grapple with how AI will affect science 认真应对 AI 将如何影响科学;grapple with a problem 苦思解决一个问题
churn out/tʃɜːn aʊt/六级雅思
释义短语 大量、快速地产出(常含粗制滥造之意)
派生churn n. 搅乳桶;v. 搅动;churn through 快速消耗
搭配churn out incomprehensible but correct solutions 大量产出难懂却正确的解答;churn out novels 粗制滥造小说
plough into/plaʊ ˈɪntuː/雅思高分
释义短语 把(大量资金、精力)投入(美式拼法 plow into)
派生plough n./v. 犁;plough back 把利润再投资;plough through 费力读完
搭配ploughing resources into science 把资源大举投入科学;plough money into 把钱砸进……
digest/daɪˈdʒest/四级雅思
释义v. 消化(食物);领会,理解(大量信息)
派生digestion n. 消化;digestive adj. 消化的;indigestible adj. 难以消化的
搭配digest what it means 领会其含义;digest the results 消化这些结果
unglamorous/ʌnˈɡlæmərəs/雅思高分
释义adj. 不光鲜的,不出彩的,乏味的
派生glamour n. 魅力;glamorous adj. 光鲜的;glamorise v. 美化
搭配unglamorous work 不起眼的工作;an unglamorous but vital task 枯燥却关键的任务
error-strewn/ˈerə struːn/雅思高分
释义adj. 错误百出的(strew 的过去分词 strewn 意为「散落」)
派生strew v. 散播,撒满;strewn with 散落着;scandal-strewn 丑闻缠身的
搭配error-strewn science 错漏百出的科学;a litter-strewn street 垃圾遍地的街道
scepticism/ˈskeptɪsɪzəm/六级雅思
释义n. 怀疑态度;怀疑论(美式拼法 skepticism)
派生sceptic n. 怀疑者;sceptical adj. 怀疑的;sceptically adv. 怀疑地
搭配a renewed emphasis on verification and scepticism 重新强调核验与怀疑;healthy scepticism 健康的怀疑
peer review/pɪə rɪˈvjuː/雅思高分
释义n. 同行评审(学术出版前的专家审稿制度)
派生peer n. 同侪;review n./v. 评审;reviewer n. 审稿人;peer-reviewed adj. 经同行评审的
搭配submit work for peer review 把成果提交同行评审;pass peer review 通过同行评审

04长难句拆解(9 句)

1. The maths results are, simultaneously, a leap forward for humanity; a marvel that underscores the breakneck progress of AI; and a warning of a future in which understanding of the universe could belong to machines that are impenetrable to human minds.
结构:
主干 The maths results are, simultaneously, ... + 表语一 a leap forward for humanity + 表语二 a marvel that underscores the breakneck progress of AI + 表语三 and a warning of a future in which ... └ 定语从句 in which understanding of the universe could belong to machines └ 定语从句 that are impenetrable to human minds
难点:① 一个主语 The maths results 带三个并列表语,用分号隔开——不用逗号,是因为每一项内部都含有从句或介词短语,逗号压不住层次;② simultaneously(同时地)插在系动词之后,一词定调:三件事同时成立;③ 第二项的 that underscores ... 是修饰 a marvel 的定语从句,underscore(凸显、强调)原义「在词下画线」;④ 第三项是「介词 + which」引导的定语从句 a future in which ...,其中 machines 后又套一层 that are impenetrable to human minds——三层修饰叠在一起,读法是从外往里剥:a warning of a future → in which understanding could belong to machines → that are impenetrable;⑤ 三项的顺序(飞跃 → 奇迹 → 警告)由褒到贬,正是全文张力的来源。
译文:这批数学成果同时是三重东西:人类的一次飞跃;一件凸显 AI 突飞猛进之速的奇迹;以及一个警告——未来对宇宙的理解,可能归属于人类心智无法穿透的机器。
2. AIs have so vividly shown their scientific potential in maths because its proofs are derived from axioms, not evidence, and are therefore, in principle, verifiable mechanically.
结构:
主干 AIs have so vividly shown their scientific potential in maths + 原因状语从句 because its proofs are derived from axioms, not evidence + 并列谓语 and are therefore, in principle, verifiable mechanically
难点:① so ... because ... 搭配:so 修饰副词 vividly(如此鲜明地),because 从句给出原因;② 从句内部 are derived from(由……推出)是被动语态,not evidence 是插入的对比成分,相当于 and not from evidence;③ and are ... verifiable 与 are derived 并列,共用主语 its proofs——注意 its 指 maths(数学的),不是 AIs;④ 两个插入语 therefore 与 in principle 把语气收紧:结论「可机械验证」只是「原则上」成立,实践里未必;⑤ verifiable mechanically(可机械地验证)这个短语是「数学为何先被 AI 攻破」的答案所在。
译文:AI 之所以在数学上如此鲜明地展现了科学潜力,是因为数学证明由公理推出、而非由证据归纳,因此原则上可以被机械地验证。
3. And it can predict the shapes of almost every known protein from its constituent amino acids, an advance that won a Nobel prize.
结构:
主干 it can predict the shapes of almost every known protein + 方式状语 from its constituent amino acids + 同位语 , an advance └ 定语从句 that won a Nobel prize
难点:① 句子真正的分量在句末的同位语 an advance:它把前面整句「重新命名」为一件大事,比另起一句说「这很了不起」更有力;② from its constituent amino acids(根据其组成氨基酸)中 constituent 是形容词「构成的」,与 constitute(构成)同源;③ almost every known 两个限定词叠加,既留了余地(almost)又显出规模(every known);④ an advance that won a Nobel prize 用已发生的荣誉为「AI 的潜力」背书——第三方认证比自夸可信,这是英文评论的常用手法;⑤ 全句只 22 个词,是本段四个战果句中的压轴,短而重。
译文:它还能根据氨基酸构成预测几乎所有已知蛋白质的形状——这项进展曾赢得诺贝尔奖。
4. Armed with life in silico, scientists may go on to achieve mastery of biology, gaining the ability to create entirely new organisms from scratch.
结构:
状语 Armed with life in silico(过去分词短语,逻辑主语为 scientists) + 主干 scientists may go on to achieve mastery of biology + 结果状语 , gaining the ability to create entirely new organisms from scratch
难点:① 句首 Armed with ...(以……为武装)是过去分词短语作状语,逻辑主语就是主句主语 scientists;② in silico(在硅中,即计算机模拟内)与 in vitro(试管内)、in vivo(活体内)同为拉丁语学术用语,书写时通常斜体;③ go on to do(接着做、进而做)表示阶段性推进,比 begin to 更强调「下一步」;④ 句末 gaining ... 是现在分词作结果状语——不是「同时在做」,而是「由此获得」,后面再接不定式 to create;⑤ from scratch(从零开始)是必背短语,achieve mastery of 比 master 更书面。
译文:有了「硅基生命」这一武装,科学家或将继续推进,最终掌握生物学——获得从零创造全新生物体的能力。
5. Having AIs churn out incomprehensible but correct solutions is like giving students the answers to their homework: without effort, there is no learning.
结构:
主语 Having AIs churn out incomprehensible but correct solutions(动名词复合结构) + 系动词 is + 表语 like giving students the answers to their homework + 解释 : without effort, there is no learning
难点:① 主语是动名词复合结构:have sb do sth(让某人做某事)整体变为动名词,所以用 Having 而不是 Have,AIs 是 have 的宾语、churn out 是省略 to 的宾补;② churn out(大量粗制滥造地生产)带贬义,与 incomprehensible but correct(难懂却正确)形成反讽——正确性并不能抵消「无意义」;③ is like doing 用 like 引出类比,口语化但极有说服力;④ 冒号后 without effort, there is no learning 是对比喻的解释:两个四词短句、节奏斩截,先给比喻再用格言点破,是英文说理段最经济的收尾法;⑤ 注意 there be 句型在此不表「存在」,而表「通则」:没有付出,就没有学习。
译文:让 AI 大量产出无法理解却正确的解答,就像把作业答案直接给学生:没有付出,就没有学习。
6. Many researchers think that a model that gets better at solving scientific problems will get better at doing everything else, too.
结构:
主干 Many researchers think + 宾语从句 that a model ... will get better at doing everything else, too └ 定语从句 that gets better at solving scientific problems(修饰 a model)
难点:① 句中有两个 that:第一个引导宾语从句,第二个引导定语从句,中间只隔一个名词 a model,很容易读串——读法:认出 think 后的宾语从句从 a model 开始,把 that gets better ... 整段划掉,先抓住 a model will get better at doing everything else;② get better at doing sth(在……上变得更强)在从句与主句各出现一次,形成「A 强则 B 也强」的推论链;③ everything else, too 三个词把话题从科学推到「所有事」,为下句「赢家通吃」铺路;④ 定语从句用一般现在时 gets,把命题写成普遍规律,而非个别观察。
译文:许多研究者认为,一个在解决科学问题上越来越强的模型,在其它一切事情上也会越来越强。
7. Even if AI labs manage to “align” machines with the interests of humanity, so that they can be fully trusted, a society where people simply ask AIs what to do—in science, or more broadly—is one in which humans will become the machines’ pets.
结构:
让步状语从句 Even if AI labs manage to align machines with the interests of humanity + 结果状语从句 so that they can be fully trusted + 主干 a society ... is one └ 定语从句 where people simply ask AIs what to do + 插入 —in science, or more broadly— └ 定语从句 in which humans will become the machines’ pets
难点:① Even if ... 引出让步:先假定对方最强的一步(对齐成功、完全可信),再说「即便如此」;② align A with B(使 A 与 B 对齐)是 AI 领域的核心术语,原文加引号,提示这是行话;③ so that they can be fully trusted 是结果状语从句,they 指 machines;④ 主句极短:a society ... is one——用 one 代替 society 避免重复,同时把两个长定语从句分别挂在 society 与 one 之后;where 与 in which 都修饰 society/one,功能相同而形式不同,避免重复用词;⑤ 破折号插入 in science, or more broadly 把结论从科学界扩到全社会;⑥ pets 一词的分量:不是「被统治」,而是「被豢养」——物质无忧,但不再作主。
译文:即便 AI 实验室成功把机器与人类利益「对齐」,使其完全可信,一个人们只须问 AI「该怎么做」的社会——在科学上如此,在更广泛的层面亦如此——也将是人类沦为机器宠物的社会。
8. Labs should be able to use AIs to solve problems, but resources such as laboratories and computing power should be directed and monitored by people who can reach an informed judgment about what AIs are doing.
结构:
并列分句一 Labs should be able to use AIs to solve problems + 并列分句二 but resources such as laboratories and computing power should be directed and monitored by people └ 定语从句 who can reach an informed judgment about ... └ 宾语从句 what AIs are doing
难点:① 两个分句用 but 对立:「可以用」vs「必须由人管」——这是全文政策主张的核心,不是禁用而是治理;② such as laboratories and computing power 是举例插入语,插在 resources 与谓语之间,读时先跳过;③ should be directed and monitored 两个被动并列,共用一个 should be;④ people who can reach an informed judgment 这个定语从句给「人」下了定义:不是「有人签字」,而是「有人真懂」——reach a judgment(作出判断)是正式搭配,informed(有依据的、知情的)是全句题眼;⑤ 句末 about what AIs are doing 是介词 about 后接宾语从句,说明判断的对象。
译文:实验室应当能用 AI 解题,但实验室、算力等资源应由能够对 AI 的作为作出知情判断的人来指挥与监督。
9. Scientists will have to learn to love—and credit with status and career advancement—guiding models and studying, verifying and teaching their results.
结构:
主干 Scientists will have to learn to love ... guiding models and studying, verifying and teaching their results + 插入 —and credit with status and career advancement— (与 love 并列:learn to love and (to) credit ... with ...)
难点:① 主干是 learn to love doing sth,love 的宾语是四个动名词:guiding / studying / verifying / teaching,两两成组;② 一对破折号插入 and credit with status and career advancement,与 love 并列,即「学会爱上、并以地位与晋升去奖励」——插入语被塞在动词与宾语之间,读法是先把它整段跳过,读完主干后再回头补上;③ credit sb with sth(把……归功于某人、给予肯定)是关键搭配,此处 sth 是 guiding models ... 这件事;④ status and career advancement(地位与职业晋升)是学术激励的两项硬指标;⑤ 这句的分量在制度设计:只讲情怀、不给奖励,改革推不动。
译文:科学家必须学会爱上「引导模型、研究并核验其结果、教授这些结果」这件事——并以地位与职业晋升来奖励它。

05写作积累

5.1 六个可直接套用的句型公式

公式 1 · 三重定性:X is, simultaneously, a ...; a ...; and a warning of ...
The maths results are, simultaneously, a leap forward for humanity; a marvel that underscores the breakneck progress of AI; and a warning of a future in which understanding of the universe could belong to machines that are impenetrable to human minds.
用法:一个主语、三个表语,用分号隔开,由褒到贬排列,一句话同时给出「好处 / 分量 / 风险」。用于评价某件影响深远的事,比「既有好处也有坏处」这类套话有力得多。
仿写 → The results are, simultaneously, a triumph of engineering; a reminder of how quickly the field has moved; and a warning that the people who build these systems no longer fully understand them.
公式 2 · 逐层递推:A gets better at X, so it will get better at everything; if so, then ...
Many researchers think that a model that gets better at solving scientific problems will get better at doing everything else, too. If so, then nobody will be able to compete with big AI labs ...
用法:先给一个看似温和的前提(在某方面更强),再用 If so(若如此)推到极端后果。合理论证的常见骨架:「若此说成立,则……」——注意 If so 之后要给出真正想说的结论。
仿写 → Many analysts think that a firm which leads in one narrow capability will soon lead in all of them. If so, then no rival will be able to compete, and the market will be settled within two years.
公式 3 · 先接底牌再反驳:Even if X manages to ..., a society where ... is one in which ...
Even if AI labs manage to “align” machines with the interests of humanity, so that they can be fully trusted, a society where people simply ask AIs what to do is one in which humans will become the machines’ pets.
用法:先假定对方最强的一步(技术难题已解决、完全可信),再说「即便如此,仍有问题」。比直接否定有力,因为把对方的底牌先接住了,反驳才不是空的。
仿写 → Even if the system can be made perfectly accurate, a workplace where nobody can explain its output is one in which judgement has quietly been outsourced.
公式 4 · 日常类比说理:Doing X is like giving students the answers to their homework: without effort, there is no learning.
Having AIs churn out incomprehensible but correct solutions is like giving students the answers to their homework: without effort, there is no learning.
用法:用读者亲身经历过的场景(抄作业、吃药)解释抽象命题(理解需要过程),再用冒号后一句格言点破。适合议论段中间的说理句。
仿写 → Outsourcing every calculation is like giving a child a calculator before he can add: the answer arrives, but the arithmetic never does.
公式 5 · 有限定的坚决:But at least until ..., X will always need ...
But at least until there is unshakeable alignment between AIs and humans, science will always need a human in the loop.
用法:用 at least until(至少在……之前)划出时间边界,把主张限定在「当下必须」,既不夸口也不退让。政策类、倡议类写作的高频句式。
仿写 → But at least until these systems can explain themselves, any decision that affects a person’s life will always need a human signature.
公式 6 · 引语收尾:It is worth remembering ... motto: “...”. That same spirit now needs to apply to ...
It is worth remembering the Royal Society’s motto, adopted in the 17th century: Nullius in verba, or “Take nobody’s word for it”. That same spirit now needs to apply to science made by the machines.
用法:引一句古老格言给当下的新问题背书,再用 That same spirit now needs to apply to ... 完成「旧精神—新对象」的平移。既显得有历史纵深,又不必自己造口号。
仿写 → It is worth remembering the first rule of the printing trade: check the proof before it goes to press. That same discipline now needs to apply to everything a model writes.

5.2 话题语料(可直接搬进作文)

话题 1 · 人工智能与科学(AI & science)

• 起手定级:October 6th may be remembered as the biggest day in the history of mathematics.
• 三重定性:a leap forward for humanity; a marvel that ...; and a warning of a future in which ...
• 列举战果:AI has made weather forecasts better. It has proposed ... And it can predict ...
• 点出机制:partly because other fields use maths as a tool
• 揭示动机:not out of altruism but because ... solving problems can help make models more intelligent

话题 2 · 理解与结果(Understanding vs results)

• 提出矛盾:The question is how to achieve progress without sacrificing understanding.
• 承认对方:the pay-off ... comes not from the proof, but from the toil
• 类比说理:... is like giving students the answers to their homework: without effort, there is no learning.
• 反证:Every time you swallow a paracetamol pill ... you rely on a mechanism that works but is not fully understood.
• 纠偏:In science, understanding should not be fetishised over results.

话题 3 · 人与机器(Human control & governance)

• 底线主张:... science will always need a human in the loop.
• 用与控分离:Labs should be able to use AIs ..., but resources ... should be directed and monitored by people ...
• 承认代价:This will, inevitably, slow progress. But it is necessary if ...
• 制度设计:learn to love—and credit with status and career advancement—guiding models ...
• 伦理要求:AI labs should seek to be good scientific citizens

5.4 仿写练习:用「三重定性 + 先接底牌再反驳」写一段评价新技术

【中文情境】一项新技术交出了漂亮成绩:速度快了十倍,错误率还更低。可最要紧的问题不是它有多强,而是——当没人能解释它为什么对的时候,我们还剩下什么?有人会说,能用就行,管它怎么算出来的。这话有一半道理:我们吃药,也从不追问药理学。但如果整个行业从此只交出结果、不再培养解释结果的人,那么所谓进步,只是把判断权悄悄让了出去。

【英文示范】The results are, simultaneously, a leap in productivity; a marvel that shows how far the technology has come; and a warning of a future in which nobody can explain why the answer is right. Some will say that a working answer is enough—and there is something in that: we take a pill without reading the pharmacology. But a profession that only ever receives answers, and never trains anyone to explain them, is one in which judgement has been quietly handed over. The rule should be simple: use the machine, but keep a human in the loop.
三处节奏:① 用 simultaneously ... a ...; a ...; and a warning of ... 完成三重定性(公式 1),顺序由褒到贬;② 用 Some will say ... —and there is something in that(这话有几分道理)先替对方把话说完,再用 But 接住(公式 3 的变体)——「先让步再反驳」比一上来否定更可信;③ 结尾用一句短促的规则句 use the machine, but keep a human in the loop 收束,与全文的核心术语呼应。

06篇章结构、论证手法与文化冷知识

6.1 论证结构(事件 → 放大 → 解释 → 展望 → 反弹 → 反驳 → 真忧 → 镜像 → 底线 → 制度 → 收尾)

① 事件
2026 年 10 月 6 日:700 多篇论文,372 个数十年悬而未决的难题
↓
② 放大
从数学推到整个科学;一句话给出三重定性——飞跃、奇迹、警告
↓
③ 解释
为什么数学先被攻破:公理演绎、可机械验证;经验科学也已被 AI 加速
↓
④ 展望
生物学最激动人心;「硅基生命」;DNA 变成另一串代码
↓
⑤ 反弹
数学家群情激愤:回报不来自证明而来自钻研——「抄作业」之喻
↓
⑥ 反驳
这是 special pleading;知识不属于学者行会;理解不该被神化——扑热息痛
↓
⑦ 真忧
AI 公司做科学是为了让模型更聪明;赢家通吃;接入物理世界
↓
⑧ 镜像
科学的问题就是社会的问题:谁来判断?即便对齐,人也只是「宠物」
↓
⑨ 底线
永远需要一个「在环的人」:可以用 AI 解题,但资源须由懂行的人指挥
↓
⑩ 制度
学者重塑学科、政府改拨款、实验室做「好科学公民」
↓
⑪ 收尾
核验不丢人:学界本有积弊——皇家学会会训 Nullius in verba

6.2 七种论证手法(可迁移到自己的议论文)

手法原文例子效果
数字起手more than 700 academic papers ... 372 problems that had baffled scholars for decades两个具体到个位的数字,让「历史性的一天」有据可依,比任何形容词都可信
三重定性a leap forward for humanity; a marvel that underscores ...; and a warning of a future in which ...一句话给出好处、分量与风险,用分号分层、由褒到贬,立起全篇张力
短句清单AI has made weather forecasts better. It has proposed millions of possible new materials ...三四个短句连列具体战果,无一形容词,却比「AI 很强大」有说服力得多
先让步再反驳These complaints contain an element of special pleading. ... Yet the fear ... is not an idle one.先给对手定性,再承认其担忧有理——两种姿态交替,读者才相信作者不是在站队
日常类比like giving students the answers to their homework / a paracetamol pill把认识论的抽象争论,落到抄作业、吃药这类人人有经验的小事上,理解成本骤降
术语落地science will always need a human in the loop用一个业内行话(human in the loop)承担全篇主张,好记、好转述、好当标题
引语收尾the Royal Society’s motto ... Nullius in verba ... That same spirit now needs to apply to ...以 17 世纪的格言为 AI 时代背书,再用一句「同样的精神」完成平移,收束稳而有历史纵深

6.3 文化冷知识

「Eureka!」:阿基米德的那一声喊

Eureka 是希腊语 heureka(我找到了),相传阿基米德在浴缸里发现浮力原理时赤身跑上街大喊此词。此后它成了「重大发现时刻」的代称,美国加州的州训就是 Eureka。本文用它作标题,暗含一层反讽:机器给出的「尤里卡时刻」,人类未必能分享那份「我找到了」的狂喜。

公理与演绎:数学为什么先被 AI 攻破

数学以公理(axiom,不加证明而假定为真的命题)为起点,靠逻辑推演得出结论,因此一个证明是否正确,可以逐步核对——理论上可被机器「机械验证」(verifiable mechanically)。这正是 Lean、Coq 等证明助手存在的理由。经验科学则依赖实验与观察,结论永远带概率性,机器难以「一次性判定对错」,所以数学成了 AI 的第一个突破口。

蛋白质结构预测与诺贝尔奖

文中「根据氨基酸构成预测蛋白质形状」指 DeepMind 的 AlphaFold:2020 年它在 CASP 评测中把结构预测精度推到接近实验水平,2024 年诺贝尔化学奖授予 Demis Hassabis、John Jumper 与 David Baker。蛋白质由 20 种氨基酸按序列折叠成三维结构,「序列决定结构」这一难题困扰生物学界五十年——AI 的解法正是「从序列直接算出结构」。

in silico / in vitro / in vivo:三个拉丁语兄弟

in vitro 意为「在玻璃器皿中」,指试管实验;in vivo 意为「在活体中」,指动物或人体实验;in silico 是 1980 年代才造出的新词(silicon 硅),意为「在硅片里」,即计算机模拟。三个短语按「体外 → 体内 → 机内」排成一条实验谱系,读科技报道时几乎必遇。

「human in the loop」:一句行话的政治分量

in the loop 原指「在信息回路里」,human in the loop(人在回路中)是自动化与武器控制领域的术语,指关键决策必须由人确认,而非全自动执行。它既是工程要求(Human-in-the-Loop 系统设计),也是伦理主张:只要最终判断还由人作出,责任就还有人可追。本文用它承担全篇主张,正是看中它「既技术又政治」的双重身份。

皇家学会与「Nullius in verba」

英国皇家学会(The Royal Society)成立于 1660 年,是世界上仍存续的最古老科学团体。其会训 Nullius in verba 为拉丁语,意为「不轻信任何人的话」(Take nobody’s word for it),强调结论须由实验证据支撑,而非由权威或经文裁定——这是近代科学方法的核心宣言。文中引它,是要把「怀疑」这件古老的事,平移到「机器给出的结果」上。

小标题里的典故:「The baby and the bathwater」

英文成语 don’t throw the baby out with the bathwater(别把婴儿和洗澡水一起倒掉)意为「别在抛弃坏东西时把好东西也扔了」。本文用它作小标题,指向末段的主张:核验与怀疑虽「不起眼」(unglamorous),却正是科学里那个不能倒掉的「婴儿」。同类标题写法还有 the tail wagging the dog、cutting the Gordian knot 等。
— 全文完 · 共 11 段 · 建议精读时长 50 分钟 —