Concise Summary简洁概述
When a student named Marcello explained an AI problem by invoking "complexity," Yudkowsky stopped him immediately. The word sounds explanatory but explains nothing — it places a pseudo-causal node behind a mystery without resolving it. This habit of "skipping over" the unknown is pervasive in AI, academia, and everyday discourse, enabled by vacuous words like "complexity," "emergence," and "sufficiently advanced." The real discipline is to linger at what you don't understand, to label it plainly as a gap, and to keep working. Yudkowsky and Marcello adopted the word "magic" as a deliberate placeholder — a reminder that work remains, rather than a false sense of closure.
当学生Marcello用「复杂性」来解释一个AI问题时，Yudkowsky立刻叫停了他。这个词听起来像解释，实则什么都没解释——它只是在谜团背后放了一个伪因果节点，并未真正化解谜团。这种「跳过」未知事物的习惯在AI、学术界和日常话语中无处不在，「复杂性」「涌现」「足够先进」等空洞词汇都是这种习惯的载体。真正的功夫在于：在你不理解的地方驻足，把它坦率地标记为空白，然后继续工作。Yudkowsky与Marcello采用「魔法」一词作为刻意的占位符——提醒自己仍有未竟之功，而非制造虚假的完结感。
Infographic信息图
The Rubik's Cube test
魔方测验
Yudkowsky asked Marcello how an AI might invent the concept of a "macro" to solve a Rubik's Cube — the question was whether the student would skip over the hard part.
Yudkowsky问Marcello：AI如何能自己发明出「宏操作」来解魔方——问题的要害在于这位学生会不会跳过最难的部分。
"Complexity" as a skip-over
「复杂性」是一种跳过
The word sounds technical and explanatory, but it doesn't constrain the space of models or concentrate probability — it just names the gap without filling it.
这个词听起来技术性十足，像是一种解释，但它并不约束模型空间，也不集中概率——它只是给空白取了个名字，却没有填补它。
A microthought below words
低于言语层面的微思维
Skip-overs happen faster than deliberate reasoning; they are default human cognition, not special stupidity. Catching them requires trained instinct, not just logic.
跳过发生的速度快于刻意推理；这是人类默认的认知方式，而非特别的愚蠢。捕捉它们需要训练出的本能，而非单纯的逻辑。
Say "magic" instead
改说「魔法」
Using "magic" as a placeholder keeps the gap visible, prevents false closure, and reminds you that real explanatory work still lies ahead.
用「魔法」作占位符，能让空白保持可见，防止虚假的完结感，并提醒你真正的解释工作还在前方。
Institutional pressure to paper over gaps
掩盖空白的机构压力
Academia rewards seemingly complete models over honestly incomplete ones; startup culture demands the same. The incentive to skip is structural, not personal.
学术界奖励表面上完整的模型，而非诚实地不完整的模型；创业文化要求亦然。跳过的动机是结构性的，而非个人的。
Detailed Summary详细概述
The Rubik's Cube Conversation
Yudkowsky frames the essay around his first encounter with Marcello, a national-level math and computing olympiad competitor he was considering as an apprentice. He had asked Marcello to describe how an AI might invent for itself the concept of a "macro" — the key to solving a Rubik's Cube — not by brute force but by reasoning about the structure of the puzzle. When Marcello responded that the AI would need "complexity to do X, and complexity to do Y," Yudkowsky immediately said: "Don't say 'complexity.'" Marcello's confusion prompted the explanation.
What "Complexity" Actually Does
Yudkowsky's objection is not that complexity is a meaningless concept — he explicitly notes it has legitimate mathematical definitions (Kolmogorov complexity, VC complexity) and intuitive uses (judging whether a hypothesis is too complicated). The objection is that Marcello was using it as a causal node that feels like an explanation but isn't. In the context of explaining how an AI acquires a capability, saying "it needs complexity" is equivalent to saying "then a miracle occurs" — it places a node in the causal chain that doesn't actually constrain what could happen next or concentrate your probability mass over possible mechanisms.
This diagnosis is the key move: Yudkowsky is not correcting a misused word; he is catching a microthought — a skip-over happening below the level of deliberate reasoning, too fast to monitor verbally. The mistake is ancient and default, not peculiar to AI discourse.
The Structure of the Skip-Over
The essay names the pattern: putting a non-controlling causal node behind something mysterious. The words that enable this are diverse — "complexity," "emergence," "sufficiently advanced," and many others that are legitimate in other contexts. The shared structural error is the skip-over itself, not the particular word. You must:
- Linger at the mystery rather than covering it with a label.
- Feel which parts of your map are blank — develop an instinct, not just a rule.
- Pay close attention to catch yourself doing it, because you cannot monitor every sentence in real time.
Institutional Incentives
Yudkowsky argues that two powerful contexts amplify the default tendency: academia, where presenting a paper with a "seemingly complete model" earns more kudos than an "explicitly incomplete map" that admits unknown steps; and startups, where acknowledging you don't know how to build the thing means confronting the collapse of your plans. The social pressure to skip is structural.
The "Magic" Convention
The essay closes with a practical remedy Yudkowsky and Marcello developed: when they hit something they didn't understand, they would say "X magically does Y" — explicitly flagging the gap as unsolved. This is superior to "complexity" or "emergence" because it creates no illusion of understanding. "Magic" is an honest placeholder, a reminder of work yet to do. The deeper lesson: words that sound like explanations but aren't are more dangerous than silence, because they give you a false sense of closure and remove the felt pressure to keep asking.
The Teachable Moment
Marcello's quick correction — "Oh, like 'emergence.' So now I've got to think about how X might actually happen..." — is what convinces Yudkowsky the student is teachable. The ability to hear the objection, map it onto other instances of the same error, and immediately redirect to the substantive question is exactly the intellectual move the essay is trying to cultivate.
魔方对话
Yudkowsky以他与Marcello的初次会面为框架。Marcello是全国数学与计算机奥林匹克竞赛的选手，Yudkowsky正在考虑是否收他为徒。他让Marcello描述：一个AI如何能自行发明「宏操作」这一概念——这是解决魔方的关键——不是靠暴力穷举，而是通过对谜题结构的推理。当Marcello回答说AI需要「复杂性来完成X，需要复杂性来完成Y」时，Yudkowsky立刻说：「不要说'复杂性'。」Marcello的困惑引出了下面的解释。
「复杂性」实际上做了什么
Yudkowsky的反对并非因为复杂性是一个无意义的概念——他明确指出它有合法的数学定义（Kolmogorov复杂度、VC复杂度）和直觉用法（判断一个假说是否过于复杂）。他反对的是：Marcello用它充当一个听起来像解释却并非解释的因果节点。在解释AI如何获得某种能力的语境中，说「它需要复杂性」等同于说「然后奇迹发生了」——它在因果链中插入了一个节点，却实际上并不约束接下来可能发生的事，也不集中可能机制上的概率质量。
这一诊断是关键所在：Yudkowsky并非在纠正一个被误用的词；他捕捉到的是一种微思维——一种发生于刻意推理层面之下的跳过，速度太快，无法在言语层面加以监控。这个错误是古老的、默认的，并非AI话语所独有。
跳过的结构
文章命名了这种模式：在神秘事物背后安放一个非控制性因果节点。能实现这一点的词汇繁多——「复杂性」「涌现」「足够先进」，以及许多在其他语境中完全合法的词。共同的结构性错误是跳过本身，而非特定的词。你必须：
- 在谜团处驻足，而非用标签覆盖它。
- 感受地图的哪些部分是空白——培养一种本能，而非仅仅一条规则。
- 密切关注自己，以便在当下捕捉到跳过行为，因为你无法实时监控每一个句子。
机构激励
Yudkowsky认为，两种强大的语境放大了这种默认倾向：在学术界，提交一篇包含「看似完整模型」的论文，比提交一张坦承存在未知步骤的「明确不完整地图」能赢得更多赞誉；在创业公司，承认自己不知道如何构建那个东西，意味着要直面计划全盘崩溃的现实。跳过的社会压力是结构性的。
「魔法」惯例
文章以Yudkowsky和Marcello在实践中开发的一个补救方法作结：每当遇到他们不理解的事物，就说「X魔法般地完成了Y」——明确地将这个空白标记为未解问题。这优于「复杂性」或「涌现」，因为它不制造任何理解的幻觉。「魔法」是一个诚实的占位符，提醒你仍有未竟之工。更深层的教训是：听起来像解释却并非解释的词，比沉默更危险，因为它们给你虚假的完结感，并消除了你继续追问的内在压力。
可教的瞬间
Marcello的快速领悟——「哦，就像'涌现'一样。那……现在我必须思考X实际上是如何发生的……」——正是让Yudkowsky认定这个学生可教的原因。能够听进反对意见、将其映射到同一错误的其他实例、并立即转向实质性问题，正是这篇文章试图培育的那种智识举动。
FAQ常见问答
Why is "complexity" singled out — aren't there many such words?为什么单独点名「复杂性」——这样的词不是有很多吗？
Yes, and the essay says so explicitly. "Emergence," "sufficiently advanced," and many legitimate words in other contexts can all do the same skip-over work. "Complexity" is singled out only because that's the word Marcello used in the conversation. The real target is the structural move of placing a non-explanatory node behind a mystery, regardless of vocabulary.
是的，文章也明确说了这一点。「涌现」「足够先进」，以及许多在其他语境中完全合法的词，都能完成同样的跳过工作。「复杂性」被单独点名，只是因为那是Marcello在对话中使用的词。真正的靶子是在谜团背后安放非解释性节点这一结构性动作，与词汇无关。
Isn't "complexity" sometimes a genuine explanation?「复杂性」有时难道不是一个真正的解释吗？
Yes — and Yudkowsky is careful to say so. Kolmogorov complexity and VC complexity are well-defined mathematical concepts with genuine explanatory power. Intuitive reasoning about whether a model is "too complex" for its evidence is also legitimate. The problem is contextual: in the step Marcello was trying to take, invoking complexity added no constraining information about which mechanism would produce the desired result.
是的——Yudkowsky也谨慎地承认了这一点。Kolmogorov复杂度和VC复杂度都是定义明确的数学概念，具有真正的解释力。关于一个模型是否「对其证据而言过于复杂」的直觉推理也是合法的。问题是情境性的：在Marcello试图迈出的那一步中，援引复杂性对于哪种机制会产生预期结果，没有增加任何约束性信息。
What does it mean for a word to "concentrate probability mass"?一个词「集中概率质量」是什么意思？
If you say "the AI needs an algorithm that processes the cube's face-rotation group structure," you've ruled out vast swaths of possible mechanisms and focused probability on a specific class of approaches. If you say "the AI needs complexity," you've ruled out nothing — almost any mechanism could be described as involving some amount of complexity. Concentrating probability mass means actually constraining what could be true.
如果你说「AI需要一个处理魔方面旋转群结构的算法」，你就排除了绝大多数可能机制，把概率集中到一类特定方法上。如果你说「AI需要复杂性」，你什么都没排除——几乎任何机制都可以被描述为涉及某种程度的复杂性。集中概率质量，意味着真正约束了什么可能为真。
Why is "magic" better than just staying silent about the gap?为什么「魔法」比对空白保持沉默更好？
Silence could mean "there's nothing to explain here" or "I've already accounted for this." "Magic" is an active, visible marker that says: this specific step is not yet understood, and someone needs to do work here. It keeps the gap in the conversation and on the to-do list, rather than letting it disappear into the background as if resolved.
沉默可能意味着「这里没什么需要解释的」，或「我已经考虑了这一步」。「魔法」是一个主动的、可见的标记，意思是：这个具体步骤尚未被理解，需要有人在此处完成工作。 它把空白保留在对话中和待办清单上，而不是让它悄悄消失于背景之中，仿佛已经被解决了。
The essay says skip-overs happen "below the level of words" — what does that mean practically?文章说跳过发生在「低于言语的层面」——这在实践中意味着什么？
It means the mistake is not a conscious logical error you can catch by checking your reasoning. The feeling that "complexity" is an adequate move in the argument is generated by a cognitive process faster than verbal thought — a trained default that short-circuits deliberation. Correcting it requires a different kind of attention: noticing an emotional or aesthetic signal (a sense of completion, of having accounted for something) that may be false.
这意味着这个错误不是一个你可以通过检查推理来发现的有意识的逻辑错误。「复杂性」是论证中一个够用的步骤——这种感觉由一个比言语思维更快的认知过程产生，是一种训练出的默认反应，它使审议发生短路。纠正它需要一种不同的注意力：察觉到一个可能是假的情感或审美信号（一种完结感，一种「已经考虑到某事」的感觉）。
Doesn't this demand an impossible standard — fully understanding every step before moving on?这难道不是要求一个不可能的标准——在继续之前完全理解每一步？
No, and the essay explicitly acknowledges this. The remedy is not to stop at every gap until it's solved, but to label the gap honestly rather than covering it with a word that sounds like a solution. You can still make progress in other parts of the model; you just shouldn't deceive yourself that the mystery has been resolved. "Magic" lets you move on while keeping the debt visible.
不，文章也明确承认了这一点。补救方法不是在每个空白前停下来直到它被解决，而是诚实地标记这个空白，而非用一个听起来像解决方案的词来覆盖它。你仍然可以在模型的其他部分取得进展；你只是不应该欺骗自己以为谜团已经被解决。「魔法」让你得以继续前进，同时保持债务的可见性。
In-depth Analysis · Pros & Cons深入解读 · 优缺点
This essay is ostensibly about a word, but actually about a cognitive move: the skip-over. Through a vivid anecdote about an AI apprentice test, Yudkowsky identifies a failure mode so common and automatic that it requires trained instinct, not mere reasoning, to catch.
这篇文章表面上在讨论一个词，实际上在讨论一种认知动作：跳过。通过一个关于AI学徒测验的生动轶事，Yudkowsky识别出了一种如此普遍、如此自动的失败模式，以至于捕捉它需要训练出的本能，而非单纯的推理。
- The anecdote is a perfect teaching device轶事是完美的教学装置By showing the error arising naturally in a real conversation, Yudkowsky avoids the abstract and makes the mistake viscerally recognizable — readers immediately recall their own instances of the same move.通过展示错误在真实对话中自然浮现，Yudkowsky避免了抽象说教，让这个错误变得可以直观辨认——读者会立刻想起自己犯过同样动作的时刻。
- The diagnostic is precise and generalizable诊断精确且可推广Identifying the problem as a structural move (non-controlling causal node) rather than a bad word means the lesson transfers to any domain — philosophy, social science, startup pitches, religious argument.将问题识别为一种结构性动作（非控制性因果节点）而非一个坏词，意味着这个教训可以迁移到任何领域——哲学、社会科学、创业路演、宗教论证。
- "Magic" is a memorable, actionable remedy「魔法」是一个令人难忘且可操作的补救方法Most essays identifying a bias don't offer a usable tool. The "magic" convention is concrete, immediately adoptable, and self-reinforcing — saying "X magically does Y" out loud makes the gap impossible to ignore.大多数识别偏差的文章不提供可用的工具。「魔法」惯例是具体的、可立即采用的、且自我强化的——大声说「X魔法般地完成了Y」，让这个空白变得不可能被忽视。
- Structural honesty about the institutional dimension关于机构层面的结构性诚实Noting that academia and startups reward gap-hiding shifts the critique from a personal failing to a systemic problem, which is both more accurate and more actionable for reformers.指出学术界和创业公司奖励掩盖空白，将批评从个人失败转向系统性问题，这既更准确，对改革者也更具操作意义。
- The standard is stated but not operationalized标准被陈述但未被操作化"Linger at the mystery" and "feel which parts of your map are blank" are evocative but vague. When exactly is a causal node "controlling" enough to count as a real explanation? The essay names the error pattern but doesn't give a test for distinguishing genuine explanatory terms from skip-over terms.「在谜团处驻足」和「感受地图哪些部分是空白」是富有感染力的表述，但很模糊。一个因果节点何时才算「足够控制性」以构成真正的解释？文章命名了错误模式，却没有给出区分真正解释性词汇与跳过型词汇的检验方法。
- The Rubik's Cube example may be misleading魔方案例可能有误导性In the specific AI context, high-level descriptions like "the system needs to represent structure" could be legitimate if they point toward a real algorithmic family, even if they don't fully specify it. The essay doesn't distinguish between premature abstraction (bad) and useful decomposition (good), leaving the reader uncertain about how much specificity is required.在具体的AI语境中，「系统需要表示结构」之类的高层描述，如果它指向一个真实的算法族，即便没有完全指定，也可能是合法的。文章没有区分过早抽象（坏）和有用分解（好），这让读者对需要多高的具体性程度感到不确定。
- Somewhat overstates the uniqueness of the problem对这一问题的独特性有所夸大The essay implies that "complexity" and "emergence" are distinctively empty, but much productive science runs on terms like "force," "fitness," or "information," which also don't fully specify mechanisms. The difference may be one of degree or community calibration rather than kind — a nuance the essay passes over.文章暗示「复杂性」和「涌现」是特别空洞的，但许多富有成效的科学运行在「力」「适应度」或「信息」等同样没有完全指定机制的术语上。区别可能是程度上的或社群校准上的，而非种类上的——文章略过了这一细微之处。
- The "magic" solution has its own risks「魔法」解决方案有其自身的风险Labeling every gap as "magic" can normalize permanent incompleteness — in a team or a culture, it may become a comfortable way to defer hard problems indefinitely rather than a spur to resolve them. The essay assumes the label will create productive discomfort, but it could just as easily become ritual acknowledgment of a problem no one feels urgency to fix.把每个空白都标记为「魔法」，可能会使永久性的不完整变得正常化——在一个团队或文化中，它可能成为一种把困难问题无限期推迟的舒适方式，而非解决它们的催化剂。文章假设这个标签会产生富有成效的不适，但它同样可能变成对一个没有人感到紧迫需要修复的问题的仪式性承认。
"Say Not 'Complexity'" is a compact, memorable intervention on one of the most common and invisible failure modes in intellectual work. Its core diagnosis — that skip-overs are microthoughts below the verbal level, requiring instinct rather than rules — is both original and correct. The practical remedy (say "magic") is immediately usable. Its main limitation is that it names the problem more sharply than it operationalizes the solution; the reader finishes knowing exactly what not to do, but only roughly how to do better.
《勿言「复杂性」》是一篇对智识工作中最普遍、最隐蔽的失败模式之一所做的简洁而令人难忘的干预。其核心诊断——跳过是低于言语层面的微思维，需要本能而非规则来捕捉——既原创又正确。实践补救方法（说「魔法」）可以立即使用。其主要局限在于：它命名问题的清晰度高于它对解决方案的操作化程度；读者读完后清楚地知道什么不该做，但只是大致知道如何做得更好。
Original Text原文
Once upon a time . . .
This is a story from when I first met Marcello, with whom I would later work for a year on AI theory; but at this point I had not yet accepted him as my apprentice. I knew that he competed at the national level in mathematical and computing olympiads, which sufficed to attract my attention for a closer look; but I didn’t know yet if he could learn to think about AI.
I had asked Marcello to say how he thought an AI might discover how to solve a Rubik’s Cube. Not in a preprogrammed way, which is trivial, but rather how the AI itself might figure out the laws of the Rubik universe and reason out how to exploit them. How would an AI invent for itself the concept of an “operator,” or “macro,” which is the key to solving the Rubik’s Cube?
At some point in this discussion, Marcello said: “Well, I think the AI needs complexity to do X, and complexity to do Y—”
And I said, “Don’t say ‘complexity.’ ”
Marcello said, “Why not?”
I said, “Complexity should never be a goal in itself. You may need to use a particular algorithm that adds some amount of complexity, but complexity for the sake of complexity just makes things harder.” (I was thinking of all the people whom I had heard advocating that the Internet would “wake up” and become an AI when it became “sufficiently complex.”)
And Marcello said, “But there’s got to be some amount of complexity that does it.”
I closed my eyes briefly, and tried to think of how to explain it all in words. To me, saying “complexity” simply felt like the wrong move in the AI dance. No one can think fast enough to deliberate, in words, about each sentence of their stream of consciousness; for that would require an infinite recursion. We think in words, but our stream of consciousness is steered below the level of words, by the trained-in remnants of past insights and harsh experience . . .
I said, “Did you read ‘A Technical Explanation of Technical Explanation’?”^1^
“Yes,” said Marcello.
“Okay,” I said. “Saying ‘complexity’ doesn’t concentrate your probability mass.”
“Oh,” Marcello said, “like ‘emergence.’ Huh. So . . . now I’ve got to think about how X might actually happen . . .”
That was when I thought to myself, “Maybe this one is teachable.”
Complexity is not a useless concept. It has mathematical definitions attached to it, such as Kolmogorov complexity and Vapnik-Chervonenkis complexity. Even on an intuitive level, complexity is often worth thinking about—you have to judge the complexity of a hypothesis and decide if it’s “too complicated” given the supporting evidence, or look at a design and try to make it simpler.
But concepts are not useful or useless of themselves. Only usages are correct or incorrect. In the step Marcello was trying to take in the dance, he was trying to explain something for free, get something for nothing. It is an extremely common misstep, at least in my field. You can join a discussion on artificial general intelligence and watch people doing the same thing, left and right, over and over again—constantly skipping over things they don’t understand, without realizing that’s what they’re doing.
In an eyeblink it happens: putting a non-controlling causal node behind something mysterious, a causal node that feels like an explanation but isn’t. The mistake takes place below the level of words. It requires no special character flaw; it is how human beings think by default, how they have thought since the ancient times.
What you must avoid is skipping over the mysterious part; you must linger at the mystery to confront it directly. There are many words that can skip over mysteries, and some of them would be legitimate in other contexts—“complexity,” for example. But the essential mistake is that skip-over, regardless of what causal node goes behind it. The skip-over is not a thought, but a microthought. You have to pay close attention to catch yourself at it. And when you train yourself to avoid skipping, it will become a matter of instinct, not verbal reasoning. You have to feel which parts of your map are still blank, and more importantly, pay attention to that feeling.
I suspect that in academia there is a huge pressure to sweep problems under the rug so that you can present a paper with the appearance of completeness. You’ll get more kudos for a seemingly complete model that includes some “emergent phenomena,” versus an explicitly incomplete map where the label says “I got no clue how this part works” or “then a miracle occurs.” A journal may not even accept the latter paper, since who knows but that the unknown steps are really where everything interesting happens?^2^
And if you’re working on a revolutionary AI startup, there is an even huger pressure to sweep problems under the rug; or you will have to admit to yourself that you don’t know how to build the right kind of AI yet, and your current life plans will come crashing down in ruins around your ears. But perhaps I am over-explaining, since skip-over happens by default in humans. If you’re looking for examples, just watch people discussing religion or philosophy or spirituality or any science in which they were not professionally trained.
Marcello and I developed a convention in our AI work: when we ran into something we didn’t understand, which was often, we would say “magic”—as in, X magically does Y”—to remind ourselves that here was an unsolved problem, a gap in our understanding. It is far better to say “magic” than “complexity” or “emergence”; the latter words create an illusion of understanding. Wiser to say “magic,” and leave yourself a placeholder, a reminder of work you will have to do later.
^1^ http://lesswrong.com/rationality/a-technical-explanation-of-technical-explanation
^2^ And yes, it sometimes happens that all the non-magical parts of your map turn out to also be non-important. That’s the price you sometimes pay, for entering into terra incognita and trying to solve problems incrementally. But that makes it even more important to know when you aren’t finished yet. Mostly, people don’t dare to enter terra incognita at all, for the deadly fear of wasting their time.
从前有这么一个故事……
这个故事发生在我第一次见到Marcello的时候，我后来与他合作了一年研究AI理论；但在那时，我还没有接受他成为我的学徒。我知道他在全国数学和计算机奥林匹克竞赛中名列前茅，这足以引起我的注意，促使我仔细观察他；但我还不知道他能否学会思考AI问题。
我让Marcello说说，他认为AI可能如何发现解决魔方的方法。不是用预先编好程序的方式——那是微不足道的——而是AI本身如何能够摸索出魔方宇宙的规律，并推理出如何加以利用。一个AI如何能够自行发明「操作符」或「宏操作」这一概念？——这正是解决魔方的关键。
在这段讨论的某个时刻，Marcello说道：「嗯，我觉得AI需要复杂性来完成X，需要复杂性来完成Y——」
我说：「不要说'复杂性'。」
Marcello问：「为什么不能说？」
我说：「复杂性本身永远不应该是一个目标。你可能需要使用某种增加一定复杂性的特定算法，但为了复杂性而追求复杂性，只会让事情更难。」（我脑海中想起了那些我听过的人，他们鼓吹互联网在变得「足够复杂」时将会「觉醒」并成为AI。）
Marcello说：「但肯定有某种程度的复杂性能做到这件事。」
我短暂地闭上眼睛，试图想办法用言语把这一切解释清楚。在我看来，说「复杂性」在AI的舞蹈中就是感觉像一个错误的动作。没有人能思考得足够快，快到能用语言审议自己意识流中的每一个句子；因为那将需要无限递归。我们用语言思考，但我们的意识流在低于语言的层面被引导，被过去洞见和惨痛经历训练出的残余所塑造……
我说：「你读过《技术性解释的技术性解释》吗？」^1^
「读过，」Marcello说。
「好，」我说。「说'复杂性'并不集中你的概率质量。」
「哦，」Marcello说，「就像'涌现'一样。嗯。所以……现在我得思考X实际上可能是如何发生的……」
就是在那一刻，我心想：也许这个是可以教的。
复杂性不是一个无用的概念。它有附属的数学定义，比如Kolmogorov复杂度和Vapnik-Chervonenkis复杂度。即便在直觉层面，复杂性也常常值得思考——你需要判断一个假说的复杂度，判断在给定的支持证据下它是否「太复杂了」，或者审视一个设计并尝试让它更简单。
但概念本身无所谓有用或无用。只有用法才有对错之分。在Marcello试图在舞蹈中迈出的那一步里，他试图免费地解释某件事，无偿地得到某样东西。这是一种极为普遍的失误，至少在我的领域如此。你可以加入一场关于通用人工智能的讨论，然后左右看到人们一遍又一遍地做着同样的事——不断地跳过他们不理解的事物，却没有意识到自己正在这么做。
这件事在眨眼之间就发生了：在某个神秘的事物背后安放一个非控制性的因果节点，一个感觉像是解释却并非解释的因果节点。这个错误发生在言语层面之下。它不需要任何特殊的性格缺陷；这是人类默认的思维方式，是他们自古以来的思维方式。
你必须避免的是跳过那个神秘的部分；你必须在谜团处驻足，直接面对它。有许多词可以跳过谜团，其中一些在其他语境下是合法的——比如「复杂性」。但本质上的错误是那个跳过，无论背后安放的是什么因果节点。跳过不是一个思想，而是一个微思想。你必须密切关注，才能在做的时候抓住自己。而当你训练自己避免跳过时，它会成为一种本能，而非言语推理。你必须感受你的地图哪些部分仍然是空白，更重要的是，要注意那种感觉。
我怀疑在学术界有巨大的压力，要把问题扫到地毯下面，这样你才能呈现出一篇看起来完整的论文。一个包含某些「涌现现象」的看似完整的模型，比一张明确不完整的地图——其中的标签写着「我完全不知道这部分是怎么运作的」或「然后奇迹发生了」——会得到更多认可。一本期刊甚至可能不接受后者，因为谁知道那些未知步骤是不是真正所有有趣事情发生的地方？^2^
而如果你在经营一家革命性的AI创业公司，那么把问题扫到地毯下面的压力就更大了；否则你将不得不向自己承认，你还不知道如何构建正确类型的AI，而你当前的人生规划将在你耳边轰然崩塌。但也许我是在过度解释，因为跳过在人类中是默认发生的。如果你在寻找例子，只需观察人们讨论宗教、哲学、灵性，或任何他们未经过专业训练的科学。
Marcello和我在我们的AI研究工作中形成了一个惯例：当我们遇到我们不理解的事物——这是经常发生的——我们会说「魔法」——就像「X魔法般地完成了Y」——来提醒我们自己，这里有一个未解的问题，一个我们理解的空白。说「魔法」远比说「复杂性」或「涌现」好得多；后两个词制造了理解的幻觉。更明智的做法是说「魔法」，给自己留下一个占位符，一个提醒，告诉自己这是以后必须完成的工作。
^1^ http://lesswrong.com/rationality/a-technical-explanation-of-technical-explanation
^2^ 是的，有时确实会发生这种情况：你地图中所有非魔法的部分，结果也都是不重要的。这是你有时必须付出的代价，因为你进入了未知领域，并试图渐进地解决问题。但这反而让知道自己尚未完成变得更加重要。大多数时候，人们根本不敢进入未知领域，因为他们对浪费时间有着致命的恐惧。