Noam Chomsky Speaks on What ChatGPT Is Really Good For

Noam Chomsky Interviewed by C.J. Polychroniou

May 3, 2023. Common Dreams.

Artificial intelligence (AI) is sweeping the world. It is transforming every walk of life and raising in the process major ethical concerns for society and the future of humanity. ChatGPT, which is dominating social media, is an AI-powered chatbot developed by OpenAI. It is a subset of machine learning and relies on what is called Large Language Models that can generate human-like responses. The potential application for such technology is indeed enormous, which is why there are already calls to regulate AI like ChatGPT.
人工智能（AI）正席卷全球。它正在改变各行各业，并在这个过程中引发社会和人类未来的重大伦理问题。主导社交媒体的 ChatGPT 是由 OpenAI 开发的 AI 聊天机器人。它是机器学习的子集，依赖于被称为大型语言模型的技术，这些模型能够生成类似人类的回复。这种技术的潜在应用非常巨大，这也是为什么已经有人呼吁监管像 ChatGPT 这样的 AI。

Can AI outsmart humans? Does it pose public threats? Indeed, can AI become an existential threat? The world’s preeminent linguist Noam Chomsky, and one of the most esteemed public intellectuals of all time, whose intellectual stature has been compared to that of Galileo, Newton, and Descartes, tackles these nagging questions in the interview that follows.
人工智能能否超越人类？它是否构成公共威胁？确实，人工智能能否成为生存威胁？世界顶尖语言学家、有史以来最受尊敬的公共知识分子之一诺姆·乔姆斯基，其学术地位堪比伽利略、牛顿和笛卡尔，在接下来的采访中解答了这些棘手的问题。

C. J. Polychroniou: As a scientific discipline, artificial intelligence (AI) dates back to the 1950s, but over the last couple of decades it has been making inroads into all sort of fields, including banking, insurance, auto manufacturing, music, and defense. In fact, the use of AI techniques has been shown in some instance to surpass human capabilities, such as in a game of chess. Are machines likely to become smarter than humans?
C. J. Polychroniou：作为一门科学学科，人工智能（AI）最早可追溯到 20 世纪 50 年代，但在过去几十年里，它已经广泛应用于各个领域，包括银行、保险、汽车制造、音乐和国防。事实上，在某些情况下，人工智能技术的应用已经超越了人类能力，例如在象棋游戏中。机器有可能比人类更聪明吗？

Noam Chomsky: Just to clarify terminology, the term “machine” here means program, basically a theory written in a notation that can be executed by a computer–and an unusual kind of theory in interesting ways that we can put aside here.
Noam Chomsky：为了明确术语，“机器”在这里指的是程序，基本上是写在一个可以被计算机执行的符号系统中的理论——而且是一种非常规的理论，其有趣之处我们在此暂且不论。

We can make a rough distinction between pure engineering and science. There is no sharp boundary, but it’s a useful first approximation. Pure engineering seeks to produce a product that may be of some use. Science seeks understanding. If the topic is human intelligence, or cognitive capacities of other organisms, science seeks understanding of these biological systems.
我们可以在纯粹工程和科学之间进行粗略区分。虽然没有明确的界限，但这是一种有用的初步近似。纯粹工程旨在生产可能有某些用途的产品，而科学则追求理解。如果主题是人类智能或其他生物的认知能力，科学便寻求理解这些生物系统。

• GPT（以及其他大语言模型）本质上是通过大规模数据、强大算力、复杂网络结构来训练出来的；
• 它的行为是可用的，但不可解释的；
• 例如无法清晰地回答：“它是如何学会做加法的？”、“某个神经元学到了什么语言规则？”
• 它不是从科学理论出发构建的，而是从实践出发，训练出了效果好的系统，然后再反过来用科学去解释它的行为。

As I understand them, the founders of AI–Alan Turing, Herbert Simon, Marvin Minsky, and others–regarded it as science, part of the then-emerging cognitive sciences, making use of new technologies and discoveries in the mathematical theory of computation to advance understanding. Over the years those concerns have faded and have largely been displaced by an engineering orientation. The earlier concerns are now commonly dismissed, sometimes condescendingly, as GOFAI–good old-fashioned AI.
在我看来，人工智能的先驱们——艾伦·图灵、赫伯特·西蒙、马文·明斯基等——曾将其视为一门科学，属于当时新兴的认知科学领域，借助计算数学理论的新技术和新发现来深化认知。然而，随着时间的推移，这些关注点逐渐淡化，并被工程化导向所取代。如今，早期的这些关注点常被轻视，甚至有时居高临下地称为“老派人工智能”（GOFAI）。

Continuing with the question, is it likely that programs will be devised that surpass human capabilities? We have to be careful about the word “capabilities,” for reasons to which I’ll return. But if we take the term to refer to human performance, then the answer is: definitely yes. In fact, they have long existed: the calculator in a laptop, for example. It can far exceed what humans can do, if only because of lack of time and memory. For closed systems like chess, it was well understood in the ‘50s that sooner or later, with the advance of massive computing capacities and a long period of preparation, a program could be devised to defeat a grandmaster who is playing with a bound on memory and time. The achievement years later was pretty much PR for IBM. Many biological organisms surpass human cognitive capacities in much deeper ways. The desert ants in my backyard have minuscule brains, but far exceed human navigational capacities, in principle, not just performance. There is no Great Chain of Being with humans at the top.
继续这个问题，程序是否有可能被设计得超越人类能力？我们必须谨慎对待“能力”这个词，原因我稍后会解释。但如果我们将术语理解为人类的表现，那么答案是肯定的。事实上，这种情况早已存在：例如笔记本电脑中的计算器。由于缺乏时间和内存，它就能远远超过人类的能力。对于像象棋这样的封闭系统，在 20 世纪 50 年代人们就清楚地认识到，随着大规模计算能力的提升和长时间的准备，最终会设计出能够击败受内存和时间限制下棋的大师级程序的。多年后的成就基本上是 IBM 的宣传。许多生物有机体在更深的层面上超越了人类的认知能力。我后院里的沙漠蚂蚁脑部微小，但在原则上，而不仅仅是表现上，它们的导航能力远远超过了人类。不存在一个以人类为首的伟大存在链。

The products of AI engineering are being used in many fields, for better or for worse. Even simple and familiar ones can be quite useful: in the language area, programs like autofill, live transcription, google translate, among others. With vastly greater computing power and more sophisticated programming, there should be other useful applications, in the sciences as well. There already have been some: Assisting in the study of protein folding is one recent case where massive and rapid search technology has helped scientists to deal with a critical and recalcitrant problem.
人工智能工程的产品正被广泛应用于各个领域，无论好坏。即使是简单常见的产品也很有用：在语言领域，自动填充、实时字幕、谷歌翻译等程序就很有帮助。随着计算能力的显著增强和编程技术的日益复杂，科学领域也应当会出现更多实用应用。事实上，已经有一些应用了：例如，大规模快速搜索技术在协助科学家研究蛋白质折叠这一关键且难题上就发挥了重要作用。

Engineering projects can be useful, or harmful. Both questions arise in the case of engineering AI. Current work with Large Language Models (LLMs), including chatbots, provides tools for disinformation, defamation, and misleading the uninformed. The threats are enhanced when they are combined with artificial images and replication of voice. With different concerns in mind, tens of thousands of AI researchers have recently called for a moratorium on development because of potential dangers they perceive.
工程项目可能有益也可能有害。这两个问题在人工智能工程中都存在。当前对大型语言模型（LLMs）及聊天机器人的研究，为制造虚假信息、诽谤和误导不明真相的人提供了工具。若将这些工具与人工图像和声音复制结合，威胁将更为严重。出于不同考量，数万名人工智能研究人员近期呼吁暂停开发，因为他们感知到潜在的危险。

As always, possible benefits of technology have to be weighed against potential costs.
始终如此，在衡量技术可能带来的好处时，必须权衡其潜在代价。

Quite different questions arise when we turn to AI and science. Here caution is necessary because of exorbitant and reckless claims, often amplified in the media. To clarify the issues, let’s consider cases, some hypothetical, some real.
当我们谈论人工智能和科学时，会引发完全不同的问题。由于媒体常常夸大那些过度且鲁莽的声明，因此需要保持谨慎。为了澄清这些问题，让我们来看一些案例，其中有些是假设性的，有些是真实的。

I mentioned insect navigation, which is an astonishing achievement. Insect scientists have made much progress in studying how it is achieved, though the neurophysiology, a very difficult matter, remains elusive, along with evolution of the systems. The same is true of the amazing feats of birds and sea turtles that travel thousands of miles and unerringly return to the place of origin.
我提到了昆虫导航，这是一个惊人的成就。尽管神经生理学是一个非常困难的问题，昆虫学家在研究它是如何实现方面取得了很大进展，但系统的进化仍然难以捉摸。鸟类和海龟的惊人壮举也是一样，它们可以旅行数千英里，准确地返回原点。

Suppose Tom Jones, a proponent of engineering AI, comes along and says: “Your work has all been refuted. The problem is solved. Commercial airline pilots achieve the same or even better results all the time.”
假设汤姆·琼斯这位支持工程化人工智能的人出现，并说：“你的工作全都被驳斥了。问题已经解决了。商业航空飞行员一直都能取得相同甚至更好的成绩。”

If even bothering to respond, we’d laugh.
如果连回应都不屑一顾，我们都会笑。

Take the case of the seafaring exploits of Polynesians, still alive among Indigenous tribes, using stars, wind, currents to land their canoes at a designated spot hundreds of miles away. This too has been the topic of much research to find out how they do it. Tom Jones has the answer: “Stop wasting your time; naval vessels do it all the time.”
以波利尼西亚人的航海事迹为例，他们至今仍生活在原住民部落中，利用星辰、风向、洋流将独木舟航行数百英里抵达指定地点。这同样一直是研究的热点，试图弄清他们如何做到的。汤姆·琼斯给出了答案：“别浪费时间了；海军舰船一直在做这件事。”

Same response. 同样的回应。

Let’s now turn to a real case, language acquisition. It’s been the topic of extensive and highly illuminating research in recent years, showing that infants have very rich knowledge of the ambient language (or languages), far beyond what they exhibit in performance. It is achieved with little evidence, and in some crucial cases none at all. At best, as careful statistical studies have shown, available data are sparse, particularly when rank-frequency (“Zipf’s law”) is taken into account.
现在我们来谈谈一个真正的案例——语言习得。近年来，这一领域的研究广泛且富有启发性，揭示了婴儿对周围语言（或多种语言）的了解极为丰富，这远超他们在实际中的表现。这种理解是通过极少的证据形成的，甚至在某些关键情况下完全没有证据。最理想的情况下，正如仔细的统计研究所示，现有数据非常稀少，尤其是在考虑等级频率（即“齐夫定律”）时。

术语	含义
ambient language	指婴儿所接触的“周围语言”（母语/环境语言）
performance	婴儿在语言上的表现（如说话、反应）
rich knowledge	指内在语言理解能力（即“心智语法”）

术语	含义
Zipf’s Law	词频排名规律：出现频率最高的词占总频率极大比例，长尾词极少
sparse data	可获得的词汇/语法结构在语料中非常稀疏

• 即使婴儿不会说话，他们已经掌握了许多语言结构、音韵规律、词序偏好，这是一种 “潜在的语言能力”（competence > performance）。

• 这挑战了“婴儿是靠模仿/学习数据统计规律获得语言”的传统观点，支持“语言能力有先天部分”。

• 高频词如“the”、“is”、“a”占据绝大多数输入；低频、复杂语法结构几乎不出现；但婴儿却学会了这些结构，说明它们不是靠“看到多少就学会多少”的方式学来的。

• 《语言本能》书中也描述，不同语言黑人在庄园间也形成了语言，婴儿之间也可以，并且符合一种边界。

Enter Tom Jones: “You’ve been refuted. Paying no attention to your discoveries, LLMs that scan astronomical amounts of data can find statistical regularities that make it possible to simulate the data on which they are trained, producing something that looks pretty much like normal human behavior. Chatbots.”
Enter Tom Jones: “你已经被反驳了。你不去注意你的发现，扫描天文数字般数据的 LLMs 能够找到统计规律性，从而模拟训练数据，产生看起来非常像正常人类行为的东西——聊天机器人。”

• 行为逼真是否等价于智能存在？像懂语言就等于懂语言吗？

This case differs from the others. First, it is real. Second, people don’t laugh; in fact, many are awed. Third, unlike the hypothetical cases, the actual results are far from what’s claimed.
这个案例与其他不同。首先，它是真实的。其次，人们不会嘲笑；事实上，许多人感到敬畏。第三，与假设案例相比，实际结果远非所声称的那样。

These considerations bring up a minor problem with the current LLM enthusiasm: its total absurdity, as in the hypothetical cases where we recognize it at once. But there are much more serious problems than absurdity.
这些考量引出了一个关于当前 LLM 热潮的小问题：其完全的荒谬性，例如在那些我们一眼就能识别的假设案例中。但存在比荒谬性更为严重的问题。

• LLM 的危险远不止“说错话”那么简单，它可能影响认知结构、信任系统、甚至社会和政治判断。因为它什么都不懂

One is that the LLM systems are designed in such a way that they cannot tell us anything about language, learning, or other aspects of cognition, a matter of principle, irremediable. Double the terabytes of data scanned, add another trillion parameters, use even more of California’s energy, and the simulation of behavior will improve, while revealing more clearly the failure in principle of the approach to yield any understanding. The reason is elementary: The systems work just as well with impossible languages that infants cannot acquire as with those they acquire quickly and virtually reflexively.
有一点是，LLM 系统被设计成无法告诉我们任何关于语言、学习或其他认知方面的事情，这是一个原则问题，无法弥补。即使扫描双倍 TB 的数据，增加另一万亿个参数，使用更多加利福尼亚的能源，行为模拟也会得到改善，但同时也更清楚地揭示了该方法的根本性失败——无法产生任何理解。原因很简单：这些系统在处理婴儿无法习得的不可能语言和它们能够快速几乎反射性地习得的语言时，效果完全一样。

It’s as if a biologist were to say: “I have a great new theory of organisms. It lists many that exist and many that can’t possibly exist, and I can tell you nothing about the distinction.”
这就像一个生物学家声称：“我提出了一个关于生物的新理论，其中包含了许多实际存在的生物以及许多不可能存在的生物，但我却无法解释它们之间的区别。”

Again, we’d laugh. Or should.
再次，我们会笑，或者应该笑。

• Linus：我终于能退休了（笑

Not Tom Jones–now referring to actual cases. Persisting in his radical departure from science, Tom Jones responds: “How do you know any of this until you’ve investigated all languages?” At this point the abandonment of normal science becomes even clearer. By parity of argument, we can throw out genetics and molecular biology, the theory of evolution, and the rest of the biological sciences, which haven’t sampled more than a tiny fraction of organisms. And for good measure, we can cast out all of physics. Why believe in the laws of motion? How many objects have actually been observed in motion?
并非汤姆·琼斯——现在指的是实际案例。汤姆·琼斯坚持他的科学激进立场，回应道：“在你调查所有语言之前，你怎么可能知道这些？”此时，对常规科学的放弃变得更加明显。按同样的逻辑，我们也可以抛弃遗传学和分子生物学、进化论，以及生物科学的其他领域，这些领域对生物的采样还不到一小部分。而且，我们还可以把物理学全部抛弃。为什么要相信运动定律？实际上又有多少物体被观察到在运动？

• 物理学采样的越多，且科技越发达，理论也随之进步，低速并没有抛弃牛顿。是不是可以类比，现在处于低等级LLM，随着采样越多，也会迎来下一个涌现

There is, furthermore, the small matter of burden of proof. Those who propose a theory have the responsibility of showing that it makes some sense, in this case, showing that it fails for impossible languages. It is not the responsibility of others to refute the proposal, though in this case it seems easy enough to do so.
此外，还有一个证明责任的小问题。提出理论的人有责任证明其合理性，具体来说，需要证明该理论在不可能的语言上会失效。但在这个案例中，其他人并没有义务去反驳该提议，尽管实际上很容易做到这一点。

• 你不能证明我错，所以我可能是对的” —— 这是错误的推理方式

Let’s shift attention to normal science, where matters become interesting. Even a single example of language acquisition can yield rich insight into the distinction between possible and impossible languages.
让我们把注意力转向常规科学，在这里事情变得有趣。即使是语言习得的一个单一例子，也能让我们对可能和不可能的语言之间的区别有丰富的见解。

The reasons are straightforward, and familiar. All growth and development, including what is called “learning,” is a process that begins with a state of the organism and transforms it step-by-step to later stages.
原因很简单，也很熟悉。所有生长和发展，包括所谓的“学习”，都是一个从生物体初始状态开始，逐步转变为后继阶段的过程。

Acquisition of language is such a process. The initial state is the biological endowment of the faculty of language, which obviously exists, even if it is, as some believe, a particular combination of other capacities. That’s highly unlikely for reasons long understood, but it’s not relevant to our concerns here, so we can put it aside. Plainly there is a biological endowment for the human faculty of language. The merest truism.
语言习得是一个过程。初始状态是语言能力的生物禀赋，这毋庸置疑地存在，即便它如某些人认为的那样，是其他能力的特定组合。由于有长期以来的原因，这种情况极不可能，但与我们现在关心的问题无关，因此可以暂时不考虑。人类语言能力显然具有生物禀赋，这毋庸置疑。

Transition proceeds to a relatively stable state, changed only superficially beyond: knowledge of the language. External data trigger and partially shape the process. Studying the state attained (knowledge of the language) and the external data, we can draw far-reaching conclusions about the initial state, the biological endowment that makes language acquisition possible. The conclusions about the initial state impose a distinction between possible and impossible languages. The distinction holds for all those who share the initial state–all humans, as far as is known; there seems to be no difference in capacity to acquire language among existing human groups.
转换过程进展到一个相对稳定的状态，之后只发生表面的变化：语言知识的获得。外部数据触发并部分塑造这一过程。通过研究达到的状态（语言知识）和外部数据，我们可以对初始状态——使语言习得成为可能的生物天赋——得出深远的结论。关于初始状态的结论确立了可能语言与不可能语言之间的区别。这种区别适用于所有共享初始状态的个体——据所知是所有人类；现存人类群体在语言习得能力上似乎没有差异。

• 人类学会语言后，其核心语法结构很早就定型了（通常 5 岁前），后期变化是“用语言的方式”变了，而不是“语言能力”本身变了。
• 语言习得不是纯靠环境刺激学来的，而是外部输入激活了大脑中“预设的语言机制”。这就是乔姆斯基的“触发论（triggering model）
• 通过研究语言知识（终态）和输入数据（外部刺激），我们可以推断出初始状态（initial state）——也就是人类大脑在出生时就具有的、支持语言习得的生物结构能力（biological endowment）。

All of this is normal science, and it has achieved many results.
所有这些都属于常规科学，并且已取得许多成果。

Experiment has shown that the stable state is substantially obtained very early, by three to four years of age. It’s also well-established that the faculty of language has basic properties specific to humans, hence that it is a true species property: common to human groups and in fundamental ways a unique human attribute.
实验显示，稳定状态在 3 到 4 岁时就已基本确立。此外，语言能力具有人类独有的基本特性，这使其成为真正的物种属性：普遍存在于人类群体中，且在本质上是人类独有的特质。

A lot is left out in this schematic account, notably the role of natural law in growth and development: in the case of a computational system like language, principles of computational efficiency. But this is the essence of the matter. Again, normal science.
在这个示意图中，许多内容被遗漏了，尤其是在增长和发展中的作用——对于像语言这样的计算系统，计算效率的原则。但这正是问题的核心。再次，常规科学。

It is important to be clear about Aristotle’s distinction between possession of knowledge and use of knowledge (in contemporary terminology, competence and performance). In the language case, the stable state obtained is possession of knowledge, coded in the brain. The internal system determines an unbounded array of structured expressions, each of which we can regard as formulating a thought, each externalizable in some sensorimotor system, usually sound though it could be sign or even (with difficulty) touch.
明确亚里士多德关于知识拥有和使用知识的区分非常重要（用当代术语来说，即能力和表现）。在语言方面，获得的稳定状态是知识拥有，储存在大脑中。内部系统决定了大量结构化表达式的无界数组，每一个都可以被视为阐述一个思想，每一个都可以在某种感觉运动系统中外化，通常是声音，但也可能是手势，甚至（困难地）触觉。

The internally coded system is accessed in use of knowledge (performance). Performance includes the internal use of language in thought: reflection, planning, recollection, and a great deal more. Statistically speaking that is by far the overwhelming use of language. It is inaccessible to introspection, though we can learn a lot about it by the normal methods of science, from “outside,” metaphorically speaking. What is called “inner speech” is, in fact, fragments of externalized language with the articulatory apparatus muted. It is only a remote reflection of the internal use of language, important matters I cannot pursue here.
内部编码系统在使用知识（表现）时被访问。表现包括思维中的内部语言运用：反思、规划、回忆等等。从统计角度看，这是语言最主要的用途。它无法通过内省获取，尽管我们可以通过科学的正常方法，从“外部”，用比喻的说法，来了解很多关于它的信息。“内部语言”实际上是被外部化的语言片段，只是发音器官被抑制了。它只是内部语言使用的遥远反映，是一些我无法在此深入探讨的重要事项。

• 我们日常说话、思考时，调用的是一种内部语言系统，这个系统事先已经在我们大脑中建好、并随时可用。这种内部语言系统分为两种，像美国畜牧学家坦普·葛兰汀属于图像思维，大多数人属于语言思维。
• 你无法“知道”你大脑里到底用了什么语法规则，但科学可以通过语言实验、语误分析、神经成像等方式推断这些规则的存在。所谓“内语”并不等同于大脑真正的语言机制，它只是你已经习得的外语规则的一种安静重演，类似于你在脑中“模拟说话”。

Other forms of use of language are perception (parsing) and production, the latter crucially involving properties that remain as mysterious to us today as when they were regarded with awe and amazement by Galileo and his contemporaries at the dawn of modern science.
语言的其他应用方式有感知（解析）和生成，而生成过程则关键性地包含了那些至今仍让我们感到神秘莫测的属性，这些属性在近代科学初现时，伽利略及其同时代人也曾对此充满敬畏和惊叹。

• 比如我们可以从没说过的一句话中，瞬间表达自己的思想，还合乎语法。你能立刻说出嵌套句、多层修饰、结构匹配一致的句子，甚至有时连自己都不知道规则是什么。

The principal goal of science is to discover the internal system, both in its initial state in the human faculty of language and in the particular forms it assumes in acquisition. To the extent that this internal system is understood, we can proceed to investigate how it enters into performance, interacting with many other factors that enter into use of language.
科学的核心目标在于揭示其内在系统，这既包括人类语言能力的初始状态，也包括其在习得过程中呈现的特定形式。只要我们理解了这个内在系统，就能进一步探究它是如何参与语言表现，并与其他多种影响语言使用的因素相互作用的。

Data of performance provide evidence about the nature of the internal system, particularly so when they are refined by experiment, as in standard field work. But even the most massive collection of data is necessarily misleading in crucial ways. It keeps to what is normally produced, not the knowledge of the language coded in the brain, the primary object under investigation for those who want to understand the nature of language and its use. That internal object determines infinitely many possibilities of a kind that will not be used in normal behavior because of factors irrelevant to language, like short-term memory constraints, topics studied 60 years ago. Observed data will also include much that lies outside the system coded in the brain, often conscious use of language in ways that violate the rules for rhetorical purposes. These are truisms known to all field workers, who rely on elicitation techniques with informants, basically experiments, to yield a refined corpus that excludes irrelevant restrictions and deviant expressions. The same is true when linguists use themselves as informants, a perfectly sensible and normal procedure, common in the history of psychology up to the present.
表现数据为内部系统的性质提供了证据，特别是当这些数据通过实验得到精炼时，如标准的田野调查工作中所做的那样。但即使是最大量的数据收集也必然在关键方面产生误导。它局限于通常产生的内容，而不是编码在大脑中的语言知识——这才是那些想要理解语言本质及其使用的人所要研究的主要对象。那个内部对象决定了无限多种可能性，但这些可能性不会在正常行为中使用，因为存在与语言无关的因素，比如短期记忆限制——这些是60年前就研究过的话题。观察到的数据还会包含许多位于大脑编码系统之外的内容，通常是为了修辞目的而有意识地违反规则的语言使用。这些都是所有田野工作者都知道的常识，他们依靠与信息提供者的启发技术——基本上是实验——来产生一个精炼的语料库，排除无关的限制和偏差表达。当语言学家把自己作为信息提供者时也是如此，这是一个完全合理和正常的程序，在心理学史上直到现在都很常见。

Proceeding further with normal science, we find that the internal processes and elements of the language cannot be detected by inspection of observed phenomena. Often these elements do not even appear in speech (or writing), though their effects, often subtle, can be detected. That is yet another reason why restriction to observed phenomena, as in LLM approaches, sharply limits understanding of the internal processes that are the core objects of inquiry into the nature of language, its acquisition and use. But that is not relevant if concern for science and understanding have been abandoned in favor of other goals.
如果继续进行常规科学，我们会发现语言的内部过程和元素无法通过观察现象来检测。这些元素通常甚至不会出现在言语（或写作）中，尽管它们的效果，通常是微妙的，可以被检测到。这就是为什么在 LLM 方法中仅限于观察现象会严重限制对语言本质、其习得和使用的核心研究对象的内部过程的理解。但如果已放弃科学和理解的目标，而转向其他目标，那么这就不相关了。

More generally in the sciences, for millennia, conclusions have been reached by experiments–often thought experiments–each a radical abstraction from phenomena. Experiments are theory-driven, seeking to discard the innumerable irrelevant factors that enter into observed phenomena–like linguistic performance. All of this is so elementary that it’s rarely even discussed. And familiar. As noted, the basic distinction goes back to Aristotle’s distinction between possession of knowledge and use of knowledge. The former is the central object of study. Secondary (and quite serious) studies investigate how the internally stored system of knowledge is used in performance, along with the many non-linguistic factors than enter into what is directly observed.
更普遍地讲，在科学界，数千年间，结论都是通过实验——通常是思想实验——得出的，每一次都是从现象中提炼出的根本性抽象。实验受理论指导，旨在排除进入观察现象中的无数无关因素，比如语言表现。这一切都如此基础，以至于很少被讨论，也很常见。正如所提到的，这种基本区分可以追溯到亚里士多德关于知识拥有和使用知识的区分。前者是研究的核心对象。次要（且相当重要）的研究则探讨内部存储的知识系统如何在表现中发挥作用，以及许多非语言因素如何影响直接观察的内容。

• 就像哲学中对物质和意识的争论一样（唯物主义与唯心主义），语言的本质也可能存在于我们无法直接观察的领域。我们并不能完全依赖感官经验来获得真知，许多重要的知识（例如，内心的感受、思想的运作）无法被直接观察或测量。

We might also recall an observation of evolutionary biologist Theodosius Dobzhansky, famous primarily for his work with Drosophila: Each species is unique, and humans are the uniquest of all. If we are interested in understanding what kind of creatures we are–following the injunction of the Delphic Oracle 2,500 years ago–we will be primarily concerned with what makes humans the uniquest of all, primarily language and thought, closely intertwined, as recognized in a rich tradition going back to classical Greece and India. Most behavior is fairly routine, hence to some extent predictable. What provides real insight into what makes us unique is what is not routine, which we do find, sometimes by experiment, sometimes by observation, from normal children to great artists and scientists.
我们也许还可以想起进化生物学家特奥多尔·道布赞斯基的一个观察，他主要以研究果蝇而闻名：每个物种都是独特的，而人类是所有物种中最为独特的。如果我们想了解自己是什么样的人——遵循 2500 年前德尔斐神谕的告诫——那么我们主要关心的就应该是是什么让人类如此独特，主要是语言和思维，它们紧密相连，正如古典希腊和印度传承下来的丰富传统所认识的那样。大多数行为都是常规的，因此在某种程度上是可以预测的。真正让我们独特的是那些非常规的东西，我们通过实验或观察来发现它们，比如从正常的孩子到伟大的艺术家和科学家。

One final comment in this connection. Society has been plagued for a century by massive corporate campaigns to encourage disdain for science, topics well studied by Naomi Oreskes among others. It began with corporations whose products are murderous: lead, tobacco, asbestos, later fossil fuels. Their motives are understandable. The goal of a business in a capitalist society is profit, not human welfare. That’s an institutional fact: Don’t play the game and you’re out, replaced by someone who will.
在此还有一个最后的评论。一个世纪以来，社会一直受到大型企业运动的影响，这些运动鼓吹蔑视科学，而娜奥米·奥雷斯克等人对此进行了深入研究。这始于那些产品具有杀伤性的公司，比如铅、烟草、石棉，后来还有化石燃料。他们的动机是可以理解的。在资本主义社会中，企业的目标是利润，而非人类福祉。这是一个制度性事实：不参与游戏，就会被淘汰，取而代之的是愿意参与的人。

• 诺贝尔物理学奖颁给了辛顿，规训科学走向AI

The corporate PR departments recognized early on that it would be a mistake to deny the mounting scientific evidence of the lethal effects of their products. That would be easily refuted. Better to sow doubt, encourage uncertainty, contempt for these pointy-headed suits who have never painted a house but come down from Washington to tell me not to use lead paint, destroying my business (a real case, easily multiplied). That has worked all too well. Right now it is leading us on a path to destruction of organized human life on earth.
企业公关部门很早就意识到，否认其产品致命影响的科学证据不断累积将是一个错误，这很容易被驳斥。更好的做法是散布怀疑，鼓励不确定性，轻视那些从未粉刷过房子却从华盛顿下来指示我不要使用含铅油漆的人，从而摧毁我的生意（这是一个真实案例，很容易被复制）。这种策略效果出奇地好。如今，它正把我们引向毁灭地球上有组织的人类生活的道路。

• 转基因问题，谁也不知道鱼和西红柿的基因结合在一起，会不会产生潜在的长期危害。就像之前己烯雌酚（DES）事件

In intellectual circles, similar effects have been produced by the postmodern critique of science, dismantled by Jean Bricmont and Alan Sokal, but still much alive in some circles.
在学术界，后现代对科学的批判产生了类似的影响，尽管这一观点被让·布里蒙和艾伦·索卡尔驳斥过，但至今仍在一些领域内流行。

It may be unkind to suggest the question, but it is, I think, fair to ask whether the Tom Joneses and those who uncritically repeat and even amplify their careless proclamations are contributing to the same baleful tendencies.
提出这个问题或许不太礼貌，但我想说的是，询问汤姆·琼斯们以及那些不加批判地重复甚至放大他们草率言论的人是否在助长同样的不良倾向，我认为这是合理的。

CJP: ChatGPT is a natural-language-driven chatbot that uses artificial intelligence to allow human-like conversations. In a recent article in The New York Times, in conjunction with two other authors, you shut down the new chatbots as a hype because they simply cannot match the linguistic competence of humans. Isn’t it however possible that future innovations in AI can produce engineering projects that will match and perhaps even surpass human capabilities?
CJP: ChatGPT 是一种基于自然语言的聊天机器人，它利用人工智能来实现类人对话。在《纽约时报》的一篇最新文章中，你和另外两位作者一起，将新的聊天机器人称为炒作，因为它们根本无法匹配人类的语言能力。然而，未来人工智能的创新发展是否有可能产生工程项目，从而匹配甚至超越人类的能力？

NC: Credit for the article should be given to the actual author, Jeffrey Watumull, a fine mathematician-linguist-philosopher. The two listed co-authors were consultants, who agree with the article but did not write it.
NC：文章的署名应归于实际作者杰弗里·沃图穆尔，一位杰出的数学家、语言学家和哲学家。列出的两位合著者仅作为顾问，他们认同文章的观点但并未参与撰写。

It’s true that chatbots cannot in principle match the linguistic competence of humans, for the reasons repeated above. Their basic design prevents them from reaching the minimal condition of adequacy for a theory of human language: distinguishing possible from impossible languages. Since that is a property of the design, it cannot be overcome by future innovations in this kind of AI. However, it is quite possible that future engineering projects will match and even surpass human capabilities, if we mean human capacity to act, performance. As mentioned above, some have long done so: automatic calculators for example. More interestingly, as mentioned, insects with minuscule brains surpass human capacities understood as competence.
确实如此，聊天机器人原则上无法与人类的语言能力相媲美，原因如前所述。它们的基本设计限制了它们达到人类语言理论最低充分条件的能力，即区分可能和不可能的语言。由于这是设计本身的特性，因此无法通过这种 AI 的未来创新来克服。然而，如果指的是人类行动能力或表现，未来的工程项目完全有可能匹配甚至超越人类的能力。如前所述，有些人早已做到了，例如自动计算器。更有趣的是，如上所述，拥有微小大脑的昆虫在人类所理解的能力方面已经超越了人类。

CJP: In the aforementioned article, it was also observed that today’s AI projects do not possess a human moral faculty. Does this obvious fact make AI robots less of a threat to the human race? I reckon the argument can be that it makes them perhaps even more so.
CJP：在上述文章中，也观察到当今的人工智能项目缺乏人类的道德能力。这个明显的事实是否使人工智能机器人对人类种族的威胁减轻？我认为可以认为这甚至可能使它们更具威胁。

NC: It is indeed an obvious fact, understanding “moral faculty” broadly. Unless carefully controlled, AI engineering can pose severe threats. Suppose, for example, that care of patients was automated. The inevitable errors that would be overcome by human judgment could produce a horror story. Or suppose that humans were removed from evaluation of the threats determined by automated missile-defense systems. As a shocking historical record informs us, that would be the end of human civilization.
NC：事实上，广义上理解“道德能力”是显而易见的。如果不加谨慎控制，人工智能工程可能会带来严重威胁。例如，假设患者护理实现了自动化。人类判断可以克服的必然错误可能会酿成一场恐怖故事。或者假设人类被排除在自动导弹防御系统对威胁的评估之外。正如令人震惊的历史记录所告知我们的，那将是人类文明的终结。

• 指斯坦尼斯拉夫·彼得罗夫 (Stanislav Petrov) 的事迹。1983年，苏联的核预警系统错误地报告美国发射了多枚导弹。彼得罗夫作为值班军官，凭借其经验和直觉判断这可能是系统故障，而没有立即上报并触发核反击程序，从而可能避免了一场全球核战争。

CJP: Regulators and law enforcement agencies in Europe are raising concerns about the spread of ChatGPT while a recently submitted piece of European Union legislation is trying to deal with AI by classifying such tools according to their perceived level of risk. Do you agree with those who are concerned that ChatGPT poses a serious public threat? Moreover, do you really think that the further development of AI tools can be halted until safeguards can be introduced?
CJP：欧洲的监管机构和执法部门正对 ChatGPT 的传播表示担忧，而最近提交的欧盟立法草案正试图通过根据其感知的风险等级来对这类工具进行分类，以应对人工智能。你是否同意那些担心 ChatGPT 构成严重公共威胁的人？此外，你是否真的认为人工智能工具的进一步发展可以被阻止，直到可以引入安全措施？

NC: I can easily sympathize with efforts to try to control the threats posed by advanced technology, including this case. I am, however, skeptical about the possibility of doing so. I suspect that the genie is out of the bottle. Malicious actors–institutional or individual–can probably find ways to evade safeguards. Such suspicions are of course no reason not to try, and to exercise vigilance.
NC：我很容易理解人们试图控制先进技术所带来的威胁的努力，包括这个案例。但是，我对这样做的前景表示怀疑。我怀疑这瓶里的精灵已经跑出来了。恶意行为者——无论是机构还是个人——很可能会找到规避安全措施的方法。当然，这些怀疑并不是不尝试和不保持警惕的理由。