Concise Summary简洁概述
Human academic science is a socially inefficient processor of evidence: each time a scientist shifts belief for unjustifiable reasons — herd behavior, entrenched intuition, collegial comfort — it takes extra justifiable evidence to compensate. Yet at the very frontier, science has always relied on individual scientists who privately know something before the social process confirms it. Generating a hypothesis worth testing requires far more Bayesian "work" than merely confirming it once you have it. So the individual reasoner who does that epistemic work rigorously can — and routinely does — get ahead of Science itself, at least briefly. This is not a bug; it is where the advancing border of knowledge actually lives.
人类学术科学在社会层面是一台低效的证据处理器：每当科学家因不正当理由（从众行为、根深蒂固的直觉、学术圈的舒适感）而改变信念，就需要额外的正当证据来抵消这些噪声。然而在真正的前沿，科学始终依赖这样的个别科学家：他们在社会过程确认之前就已私下知晓了某件事。提出一个值得检验的假说，所需的贝叶斯「功」，远超你拥有假说之后去确认它所需的功。因此，一个严格完成了这番认识论工作的个体推理者，完全可以——而且经常确实——走在科学本身的前面，至少暂时如此。这不是缺陷，而恰恰是知识推进边界的真实所在。
Infographic信息图
Science is socially slow
科学在社会层面是缓慢的
Beliefs shift for epistemic and non-epistemic reasons alike; each unjustified shift requires extra evidence to cancel out, making the collective machinery inefficient.
信念的转变既有认识论上的原因，也有非认识论上的原因；每一次无正当理由的转变都需要额外的证据来抵消，使集体机制低效。
The frontier runs on individual foresight
前沿依赖个体的先见
At the advancing border, individual scientists must privately see a truth — pick the right hypothesis to test — before the experiment provides public confirmation.
在推进边界处，个别科学家必须私下先「看见」真理——选出正确的假说去检验——然后实验才提供公开的确认。
Generating beats confirming
提出假说比确认假说耗功更大
In a large answer space, raising one hypothesis to 10% probability takes far more Bayesian evidence than moving it from 10% to 90% once it is already in view.
在大型答案空间中，将某一假说的概率提升到10%，所需的贝叶斯证据远多于此后从10%推至90%。
Random guessing would take zillions of years
随机猜测需要无数亿年
Science does not specify how to generate hypotheses; but in practice a random Ouija board fails because the answer space is astronomically large.
科学没有规定如何产生假说；但实际上随机方法行不通，因为答案空间是天文数字级的大。
The individual can outpace the institution
个体可以超越机构
A scientist who has done the hard epistemic work already knows something true before it is officially confirmed — they are, briefly, faster than Science.
一位已完成艰难认识论工作的科学家，在官方确认之前就已知晓某件真实的事情——他们暂时比科学跑得更快。
Detailed Summary详细概述
Yudkowsky opens with a sardonic observation: science advances when the mountain of evidence grows so enormous that even resistant scientists cannot ignore it — and distinguishing a scientist from a non-scientist is that the latter ignores it anyway. He then quotes Max Planck's even bleaker version: science advances not by convincing opponents but by waiting for them to die. The practical implication, Yudkowsky notes with amusement, is that truth's institutional fate rests on the aesthetic tastes of graduate students.
Science as a Noisy Processor
The many-worlds interpretation of quantum mechanics illustrates the problem. Its gradual academic acceptance shows that physicists convert not on pure epistemic grounds but when the community pack grows large enough to provide social comfort. Each unjustifiable belief-shift is epistemic noise that requires compensating evidence. From a Bayesian standpoint, human science is a highly inefficient processor of evidence.
The collapse of the wavefunction is Yudkowsky's exhibit: it has no experimental justification, yet it appeals to an undermined intuition of a single world, and it takes extra arguments (like its violation of Special Relativity) to begin unwinding an idea that should never have had non-negligible probability. This is not innocent error — in many fields, debates drag on decades past when they should have been settled, sustained by entrenched parties demanding ever more evidence.
The Frontier Problem
Yet Science still eventually arrives, and this powers civilization. The deeper question is: where do ideas come from? Science doesn't specify. In principle, a robot Ouija board using digits of pi could generate hypotheses for testing; if those hypotheses kept being confirmed, pure Science wouldn't care. But Bayes would combust — because Bayes is falsified by more possible outcomes than Science is.
In practice the Ouija board fails because the answer space is vast. Consider equations specifiable in 32 bits or fewer: roughly 4 billion possibilities. Elevating one hypothesis from the noise to 10% probability requires far more Bayesian evidence — more epistemic work — than confirming it further from 10% to 90%.
The Individual Interval
This means that at the frontier of science, individual researchers must privately do the epistemic work of identifying which hypothesis to test, before the experiment confirms it publicly. They may not consciously know they've succeeded; but if the answer space is large and they pick the right hypothesis, they must have done that work. The interval between a scientist's private knowledge and public confirmation by Science is not trivial — it is where the advancing border lives.
"In between was an interval where the scientist rationally knew something that the public social process of science hadn't yet confirmed. And this is not a trivial interval, though it may be short; for it is where the frontier of science lies, the advancing border."
The essay ends by noting the asymmetry: it is much easier to train people to test ideas than to generate good ones to test — which is exactly where individual epistemic quality matters most.
Yudkowsky 以一句带刺的话开场：科学的进步靠的是证据的大山积累得如此之高，以至于就连抵制的科学家也无法忽视——而区分科学家与非科学家的标准，正是后者无论如何都会忽视。随后他引用了普朗克更为悲观的版本：科学不是靠说服对手获胜，而是靠等待对手死去。Yudkowsky 饶有趣味地指出其实际含义：真理的机构命运取决于研究生们的审美品味。
科学作为噪声处理器
量子力学的多世界诠释提供了一个说明问题的例证。它在学术界渐进的接受过程表明，物理学家的转变并非出于纯粹的认识论理由，而是在学术圈子足够大、能提供社会舒适感时才发生。每一次无正当理由的信念转变都是认识论噪声，需要补偿性证据来抵消。从贝叶斯视角看，人类科学是一台极为低效的证据处理器。
波函数坍缩是Yudkowsky的核心例证：它没有实验上的正当依据，却迎合了关于单一世界的（已被动摇的）直觉，需要额外的论证（例如它违反了狭义相对论）才能开始瓦解一个本不该被赋予不可忽视概率的想法。这不仅仅是无辜的错误——在许多领域，争论拖延的时间远远超过本应解决的时间，由盘踞权力的一方不断要求更多证据来维持。
前沿问题
然而，科学终究会到达，这支撑着文明。更深的问题是：想法从哪里来？科学没有规定。从原则上说，一台用圆周率数字驱动的机器人占卜板可以生成假说去检验；如果这些假说持续得到确认，纯粹的科学不会在乎。但贝叶斯会崩溃——因为比起科学，贝叶斯被更多可能的结果所证伪。
在实践中，占卜板行不通，因为答案空间极为巨大。考虑32位或更少比特可以描述的方程：大约有40亿种可能。将一个假说从噪声中提升到10%的概率，所需的贝叶斯证据——所需的认识论功——远多于此后从10%确认到90%所需的功。
个体的间隔
这意味着，在科学的前沿，个别研究者必须私下完成「识别应该检验哪个假说」的认识论工作，而这发生在实验公开确认它之前。他们可能并不有意识地知道自己已经成功；但如果答案空间很大而他们选对了假说，他们必定已经完成了那番工作。科学家的私人知识与科学的公开确认之间的间隔并不微不足道——那正是推进边界的所在。
「在这中间存在一个间隔，在此期间科学家理性地知晓了某件科学的公开社会过程尚未确认的事情。这个间隔并不微不足道，尽管它可能很短；因为正是在这里，科学的前沿所在，推进的边界。」
文章最后指出一个不对称性：训练人们去检验想法，远比让他们产生值得检验的好想法容易——而个体认识论质量最重要的地方，恰恰正在于此。
FAQ常见问答
Does "faster than Science" mean Science is wrong or bad?「比科学更快」是说科学是错误的或糟糕的吗？
No. Yudkowsky stresses that Science still gets there eventually, and that's sufficient to power technological civilization. The argument is that the individual careful reasoner at the frontier can briefly outpace the slower social machinery — not that Science should be discarded.
不是。Yudkowsky 强调科学终究能到达，这足以支撑技术文明。论点是：前沿的个体谨慎推理者可以短暂地超越更慢的社会机制——而不是说科学应该被抛弃。
What makes hypothesis generation so much harder than confirmation?是什么让假说的提出比确认难得多？
In a large answer space (e.g., all equations specifiable in 32 bits — roughly 4 billion possibilities), most candidates are wrong. Raising one specific hypothesis to even 10% probability requires enormous Bayesian evidence. Once you already have that hypothesis at 10%, pushing it to 90% is comparatively cheap. This asymmetry is the core of the "Einstein's Arrogance" point Yudkowsky references.
在大型答案空间（例如，32位可描述的所有方程——大约40亿种可能）中，大多数候选项都是错的。将某一特定假说的概率提升到哪怕10%，就需要巨大的贝叶斯证据。一旦你已经有了那个处于10%的假说，把它推到90%相对来说则代价低廉。这种不对称性是Yudkowsky所引用的「爱因斯坦的傲慢」的核心。
What is the Ouija board thought experiment meant to show?「占卜板」思想实验要说明什么？
It shows that Science, as a formal method, does not specify how to generate hypotheses — only how to test them. If a random device somehow produced correct hypotheses, pure Science would endorse them once confirmed. But Bayes would reject the process because a random source tells us nothing about which hypotheses to promote in probability. In practice, the random approach fails because the answer space is too large.
它表明，科学作为一种正式方法，并没有规定如何产生假说——只规定了如何检验假说。如果某个随机装置碰巧产生了正确的假说，纯粹的科学在确认后会接受它们。但贝叶斯会拒绝这一过程，因为随机来源不能告诉我们应该在概率上提升哪些假说。实践中，随机方法行不通，因为答案空间太大了。
How does the inefficiency of academic science relate to epistemic virtue for individuals?学术科学的低效性与个人的认识论美德有何关联？
The social noise in science — herd behavior, entrenched intuitions, comfort-seeking — means individuals who do reason without those distortions can reach correct conclusions sooner. This is not an argument for ignoring consensus, but for understanding why it sometimes lags, and maintaining one's own epistemic discipline rather than outsourcing all reasoning to the institution.
科学中的社会噪声——从众行为、根深蒂固的直觉、寻求舒适感——意味着那些确实在没有这些扭曲的情况下推理的个体，可以更早地得出正确结论。这不是无视共识的论据，而是理解共识为何有时滞后，并保持自身认识论纪律，而非将所有推理外包给机构。
Is the frontier interval only relevant in physics?「前沿间隔」只与物理学相关吗？
Yudkowsky explicitly extends it to any non-routine science with a large answer space. He also mentions the evolutionary psychology debates as a case of socially extended noise. The principle applies wherever the answer is not drawn from a small, obvious set of alternatives — any domain with genuine novelty.
Yudkowsky 明确将其延伸至任何具有大型答案空间的非常规科学。他还将进化心理学的争论作为社会性延长噪声的案例。这一原则适用于任何答案不出自一个小型、显而易见的备选集的领域——任何具有真正新颖性的领域。
Does this mean individual reasoners should trust themselves over scientific consensus?这是否意味着个体推理者应该信任自己超过科学共识？
Only with great care. The essay is not a license for contrarianism. The argument is specifically about the frontier — where consensus hasn't yet formed because the evidence is genuinely new. In well-settled domains, the social machinery has already done its work, and the individual's deviation is more likely noise than signal. Yudkowsky's point is descriptive (this is how knowledge advances) rather than prescriptive (therefore always trust yourself).
只能极为谨慎地这样做。这篇文章不是特立独行的许可证。论点专门针对前沿——在那里，共识尚未形成，因为证据确实是全新的。在已有定论的领域，社会机制已经完成了它的工作，个体的偏离更可能是噪声而非信号。Yudkowsky 的观点是描述性的（知识就是这样推进的），而非规范性的（因此永远信任自己）。
In-depth Analysis · Pros & Cons深入解读 · 优缺点
This essay argues that Science, as a social institution, is epistemically inefficient — and that the frontier of knowledge is always inhabited by individuals who privately "know" things before the institution confirms them. It is less a critique of Science than an explanation of how individual epistemic virtue can outpace collective machinery.
这篇文章论证，科学作为一种社会机制在认识论上是低效的——而知识的前沿始终由那些在机构确认之前就已私下「知晓」某些事情的个体所占据。这与其说是对科学的批判，不如说是对个体认识论美德如何超越集体机制的解释。
- The hypothesis-generation asymmetry is real and underappreciated假说生成的不对称性是真实的且被低估了Yudkowsky's Bayesian analysis of large answer spaces is genuinely insightful: the epistemic work required to identify a hypothesis worth testing is often orders of magnitude larger than to confirm it. This is routinely ignored in simple accounts of the scientific method.Yudkowsky 对大型答案空间的贝叶斯分析确实洞见深刻：识别一个值得检验的假说所需的认识论工作，往往比确认它大几个数量级。这在科学方法的简单描述中经常被忽略。
- Honest about Science's real mechanics对科学的真实机制诚实The many-worlds and wavefunction examples are specific, falsifiable illustrations of social noise — not vague gestures. Planck's quote and the evolutionary psychology reference root the argument in documented history rather than abstract theorizing.多世界和波函数的例子是社会噪声的具体可证伪的说明——而非模糊的手势。普朗克的引文和进化心理学的参照，将论证植根于有据可查的历史，而非抽象理论。
- Clarifies the relationship between Science and Bayes without dismissing either澄清了科学与贝叶斯之间的关系，而不否定任何一方The Ouija board example cleanly separates what Science does specify (testing) from what it doesn't (generation), and shows that Bayesian reasoning is actually more demanding — not a looser standard.占卜板的例子清晰地区分了科学所规定的（检验）与其未规定的（生成），并表明贝叶斯推理实际上要求更高——而非更宽松的标准。
- Actionable conclusion可操作的结论The essay ends with a practical asymmetry: testing is teachable, generating good hypotheses is not. This implies that individual epistemic quality — the hard-to-train part — is where genuine intellectual contribution lives.文章以一个实践性的不对称性收尾：检验可以被教授，产生好的假说则不然。这暗示个体的认识论质量——难以训练的那部分——正是真正智识贡献的所在。
- Conflates two distinct claims混淆了两个不同的主张The claim that Science is socially noisy and the claim that individuals can be faster are both true, but the argument slides between them. A slow, noisy institution does not automatically imply that a specific individual with good priors will outpace it — that individual might also be biased, just differently.「科学在社会层面有噪声」和「个体可以更快」这两个主张都是真实的，但论证在两者之间滑动。一个缓慢、有噪声的机构并不自动意味着某个先验良好的特定个体能超越它——那个个体也可能存在偏差，只是偏差方式不同。
- The many-worlds example cuts both ways多世界的例子是双刃剑Yudkowsky uses many-worlds as evidence of Science's slowness to correct a bias toward collapse. But this presupposes that many-worlds is clearly correct — a position many physicists still contest. Using a contested case as an illustration of how Science fails to reach obvious truths is methodologically circular.Yudkowsky 用多世界作为科学缓慢纠正坍缩偏见的证据。但这预设了多世界是明显正确的——这一立场仍有许多物理学家争议。用一个争议案例来说明科学如何未能抵达显而易见的真理，在方法论上是循环的。
- Understates the cost of premature certainty低估了过早确定性的代价The essay celebrates the interval where an individual "rationally knows" something before Science confirms it. But this interval is also where most contrarian errors live. The asymmetric costs of type-I vs. type-II errors at the frontier are not addressed.文章颂扬个体在科学确认之前「理性地知晓」某件事情的间隔。但这个间隔也是大多数特立独行错误的所在。前沿处第一类错误与第二类错误的不对称代价并未被讨论。
- The Bayesian thermodynamics claim is asserted rather than shown贝叶斯热力学的主张是断言而非论证The reference to Bayesian "work" in a thermodynamic sense is gesturally interesting but technically underdeveloped. The link between information-theoretic surprise and the effort required to generate good hypotheses is not spelled out, leaving the analogy doing more rhetorical work than logical weight.对贝叶斯热力学意义上「功」的提及在手势上很有趣，但在技术上没有充分展开。信息论意义上的意外性与产生好假说所需努力之间的联系没有被明确阐述，使得这个类比承担了比其逻辑分量更多的修辞工作。
A compact and genuinely original analysis of a gap in standard accounts of the scientific method. Its core insight — that hypothesis generation requires more epistemic work than hypothesis confirmation in large answer spaces — is underappreciated and worth taking seriously. The essay is weakened by leaning on contested cases as obvious examples and by not sufficiently hedging the individual-vs.-institution asymmetry it celebrates.
这是对科学方法标准描述中一个空白的简洁而真正原创的分析。其核心洞见——在大型答案空间中，假说的生成比假说的确认需要更多认识论工作——被低估了，值得认真对待。文章因依赖争议案例作为显而易见的例证，以及没有充分对冲它所颂扬的个体对机构不对称性，而有所削弱。
Original Text原文
I sometimes say that the method of science is to amass such an enormous mountain of evidence that even scientists cannot ignore it; and that this is the distinguishing characteristic of a scientist, a non-scientist will ignore it anyway.
Max Planck was even less optimistic:
"A new scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it."
I am much tickled by this notion, because it implies that the power of science to distinguish truth from falsehood ultimately rests on the good taste of grad students.
The gradual increase in acceptance of many-worlds in academic physics, suggests that there are physicists who will only accept a new idea given some combination of epistemic justification, and a sufficiently large academic pack in whose company they can be comfortable. As more physicists accept, the pack grows larger, and hence more people go over their individual thresholds for conversion—with the epistemic justification remaining essentially the same.
But Science still gets there eventually, and this is sufficient for the ratchet of Science to move forward, and raise up a technological civilization.
Scientists can be moved by groundless prejudices, by undermined intuitions, by raw herd behavior—the panoply of human flaws. Each time a scientist shifts belief for epistemically unjustifiable reasons, it requires more evidence, or new arguments, to cancel out the noise.
The "collapse of the wavefunction" has no experimental justification, but it appeals to the (undermined) intuition of a single world. Then it may take an extra argument—say, that collapse violates Special Relativity—to begin the slow academic disintegration of an idea that should never have been assigned non-negligible probability in the first place.
From a Bayesian perspective, human academic science as a whole is a highly inefficient processor of evidence. Each time an unjustifiable argument shifts belief, you need an extra justifiable argument to shift it back. The social process of science leans on extra evidence to overcome cognitive noise.
A more charitable way of putting it is that scientists will adopt positions that are theoretically insufficiently extreme, compared to the ideal positions that scientists would adopt, if they were Bayesian AIs and could trust themselves to reason clearly.
But don't be too charitable. The noise we are talking about is not all innocent mistakes. In many fields, debates drag on for decades after they should have been settled. And not because the scientists on both sides refuse to trust themselves and agree they should look for additional evidence. But because one side keeps throwing up more and more ridiculous objections, and demanding more and more evidence, from an entrenched position of academic power, long after it becomes clear from which quarter the winds of evidence are blowing. (I'm thinking here about the debates surrounding the invention of evolutionary psychology, not about many-worlds.)
Is it possible for individual humans or groups to process evidence more efficiently—reach correct conclusions faster—than human academic science as a whole?
"Ideas are tested by experiment. That is the core of science." And this must be true, because if you can't trust Zombie Feynman, who can you trust?
Yet where do the ideas come from?
You may be tempted to reply, "They come from scientists. Got any other questions?" In Science you're not supposed to care where the hypotheses come from—just whether they pass or fail experimentally.
Okay, but if you remove all new ideas, the scientific process as a whole stops working because it has no alternative hypotheses to test. So inventing new ideas is not a dispensable part of the process.
Now put your Bayesian goggles back on. As described in Einstein's Arrogance, there are queries that are not binary—where the answer is not "Yes" or "No", but drawn from a larger space of structures, e.g., the space of equations. In such cases it takes far more Bayesian evidence to promote a hypothesis to your attention than to confirm the hypothesis.
If you're working in the space of all equations that can be specified in 32 bits or less, you're working in a space of 4 billion equations. It takes far more Bayesian evidence to raise one of those hypotheses to the 10% probability level, than it requires further Bayesian evidence to raise the hypothesis from 10% to 90% probability.
When the idea-space is large, coming up with ideas worthy of testing, involves much more work—in the Bayesian-thermodynamic sense of "work"—than merely obtaining an experimental result with p<0.0001 for the new hypothesis over the old hypothesis.
If this doesn't seem obvious-at-a-glance, pause here and read Einstein's Arrogance.
The scientific process has always relied on scientists to come up with hypotheses to test, via some process not further specified by Science. Suppose you came up with some way of generating hypotheses that was completely crazy—say, pumping a robot-controlled Ouija board with the digits of pi—and the resulting suggestions kept on getting verified experimentally. The pure ideal essence of Science wouldn't skip a beat. The pure ideal essence of Bayes would burst into flames and die.
(Compared to Science, Bayes is falsified by more of the possible outcomes.)
This doesn't mean that the process of deciding which ideas to test is unimportant to Science. It means that Science doesn't specify it.
In practice, the robot-controlled Ouija board doesn't work. In practice, there are some scientific queries with a large enough answer space, that picking models at random to test, it would take zillions of years to hit on a model that made good predictions—like getting monkeys to type Shakespeare.
At the frontier of science—the boundary between ignorance and knowledge, where science advances—the process relies on at least some individual scientists (or working groups) seeing things that are not yet confirmed by Science. That's how they know which hypotheses to test, in advance of the test itself.
If you take your Bayesian goggles off, you can say, "Well, they don't have to know, they just have to guess." If you put your Bayesian goggles back on, you realize that "guessing" with 10% probability requires nearly as much epistemic work to have been successfully performed, behind the scenes, as "guessing" with 80% probability—at least for large answer spaces.
The scientist may not know he has done this epistemic work successfully, in advance of the experiment; but he must, in fact, have done it successfully! Otherwise he will not even think of the correct hypothesis. In large answer spaces, anyway.
So the scientist makes the novel prediction, performs the experiment, publishes the result, and now Science knows it too. It is now part of the publicly accessible knowledge of humankind, that anyone can verify for themselves.
In between was an interval where the scientist rationally knew something that the public social process of science hadn't yet confirmed. And this is not a trivial interval, though it may be short; for it is where the frontier of science lies, the advancing border.
All of this is more true for non-routine science than for routine science, because it is a notion of large answer spaces where the answer is not "Yes" or "No" or drawn from a small set of obvious alternatives. It is much easier to train people to test ideas, than to have good ideas to test.
我有时说，科学的方法就是积累如此巨大的证据山，以至于就连科学家都无法忽视它；而这正是区分科学家与非科学家的特征——非科学家无论如何都会忽视它。
马克斯·普朗克甚至更不乐观：
「一个新的科学真理并不是通过说服其反对者、让他们看到光明而获胜的，而是因为它的反对者最终都死去了，而熟悉这一真理的新一代人成长起来了。」
我对这一想法颇感好笑，因为它意味着科学区分真伪的能力，最终建立在研究生们的品味之上。
多世界在学术物理学界被逐渐接受，这表明有些物理学家只会在获得认识论上的正当依据与足够大的学术圈子（他们能在其中感到舒适）的组合时，才会接受一个新想法。随着越来越多的物理学家接受，圈子变得越来越大，因此越来越多的人越过了各自的转变阈值——而认识论上的正当依据基本上保持不变。
但科学终究会到达那里，这对于科学的棘轮向前转动、提升起一个技术文明已经足够。
科学家可能受到毫无根据的偏见、被动摇的直觉、纯粹的从众行为的驱使——所有这些人类缺陷的大全。每当科学家因认识论上无法正当的理由而转变信念，就需要更多证据或新论证来抵消这种噪声。
「波函数坍缩」没有实验上的正当依据，但它迎合了关于单一世界的（已被动摇的）直觉。然后，可能需要一个额外的论证——比如坍缩违反了狭义相对论——才能开始缓慢地从学术上瓦解一个本不该被赋予不可忽视概率的想法。
从贝叶斯视角看，作为整体的人类学术科学，是一台高度低效的证据处理器。每当一个无法正当的论证转变了信念，就需要一个额外的可正当的论证来将其转变回来。科学的社会过程依赖额外的证据来克服认知噪声。
换一种更宽容的说法是：科学家采用的立场在理论上会不够极端，与科学家在成为贝叶斯人工智能、能够信任自己清晰推理时将会采用的理想立场相比。
但也不要太宽容。我们谈论的噪声并不全是无辜的错误。在许多领域，争论在本应解决之后拖延了数十年。不是因为双方的科学家都拒绝信任自己并同意应该寻找更多证据。而是因为一方不断提出越来越荒谬的反对意见，并要求越来越多的证据，从一个根深蒂固的学术权力地位出发，远在证据风向明朗之后。（我在这里想到的是围绕进化心理学发明的争论，而不是多世界。）
个体人类或群体是否有可能比整体的人类学术科学更高效地处理证据——更快地得出正确结论？
「想法通过实验来检验。这是科学的核心。」这一定是真的，因为如果你不能信任僵尸费曼，你还能信任谁呢？
然而想法从哪里来？
你可能会回答，「它们来自科学家。还有别的问题吗？」在科学中，你不应该在乎假说从哪里来——只关心它们是否通过或未通过实验。
好的，但如果你去除所有新想法，整个科学过程就会停止运转，因为它没有可供检验的备选假说了。所以发明新想法不是该过程中可以省去的部分。
现在再戴上你的贝叶斯眼镜。如爱因斯坦的傲慢中所描述的，有些问题不是二元的——答案不是「是」或「否」，而是从一个更大的结构空间中抽取的，例如方程的空间。在这种情况下，将一个假说提升到你的注意力范围内，比确认该假说，需要多得多的贝叶斯证据。
如果你在所有可以用32位或更少比特来描述的方程的空间中工作，你就是在40亿个方程的空间里工作。将其中一个假说提升到10%的概率水平，所需的贝叶斯证据，远多于进一步将该假说从10%提升到90%概率所需的贝叶斯证据。
当想法空间很大时，提出值得检验的想法，涉及的功——在贝叶斯热力学意义上的「功」——远多于仅仅获得一个对新假说相较旧假说p<0.0001的实验结果。
如果这乍看起来不显而易见，请在此暂停，阅读爱因斯坦的傲慢。
科学过程始终依赖科学家通过某种科学本身没有进一步规定的过程来提出待检验的假说。假设你想出了某种完全疯狂的生成假说的方式——比如，用圆周率的数字来驱动一个机器人控制的占卜板——而由此得到的建议不断地被实验验证。纯粹理想本质的科学不会有丝毫停顿。纯粹理想本质的贝叶斯会猛然燃烧并死去。
（与科学相比，贝叶斯被更多可能的结果所证伪。）
这并不意味着决定检验哪些想法的过程对科学来说不重要。它意味着科学没有规定这一过程。
在实践中，机器人控制的占卜板行不通。在实践中，有一些科学问题的答案空间足够大，以至于随机选取模型来检验，将需要无数亿年的时间才能碰到一个能做出良好预测的模型——就像让猴子打出莎士比亚一样。
在科学的前沿——无知与知识的边界，科学推进的地方——这个过程依赖至少一些个别科学家（或工作小组）看到那些尚未被科学确认的事情。这就是他们在检验本身之前，知道应该检验哪些假说的方式。
如果你摘下贝叶斯眼镜，你可以说，「好吧，他们不必知道，他们只需猜测就好。」如果你重新戴上贝叶斯眼镜，你会意识到以10%概率「猜测」，与以80%概率「猜测」相比，需要在幕后成功地完成几乎同等的认识论工作——至少对于大型答案空间而言。
科学家可能并不知道他已经成功地完成了这番认识论工作，在实验之前；但他必须，实际上，已经成功地完成了它！否则他甚至不会想到正确的假说。反正是在大型答案空间中。
所以科学家做出了新颖的预测，进行了实验，发表了结果，而现在科学也知道了。它现在是人类公开可获取知识的一部分，任何人都可以自行验证。
在这中间存在一个间隔，在此期间科学家理性地知晓了某件科学的公开社会过程尚未确认的事情。这个间隔并不微不足道，尽管它可能很短；因为正是在这里，科学的前沿所在，推进的边界。
所有这些对于非常规科学比对于常规科学更为真实，因为它是一个关于大型答案空间的概念，答案不是「是」或「否」或来自一组显而易见的少量备选项。训练人们去检验想法，远比让他们拥有值得检验的好想法容易得多。