A New Framework for How the Brain Compresses Our Noisy World

Every moment of our lives, our bodies are awash in sensory signals.

我们生命中的每时每刻，我们的身体都充满了感官信号。

Photons hit our retinas.

光子撞击我们的视网膜。

Waves of compressed air collide with our eardrums.

压缩空气波撞击我们的耳膜。

Volatile molecules bind to receptors in our nostrils, and chemicals slather our taste buds.

挥发性分子与我们鼻孔中的受体结合，化学物质会覆盖我们的味蕾。

Pressure and heat activate nerve endings in our skin.

压力和热量激活我们皮肤的神经末梢。

We are able to navigate this torrent because the brain does an enormous amount of data compression.

我们之所以能够驾驭这股洪流，是因为大脑进行了大量的数据压缩。

Through a process known as categorization, the brain turns the messy, noisy, information-rich world into objects, people, concepts, and emotions that we can understand and act on at the level of experience.

通过一个称为分类的过程，大脑将混乱、嘈杂、信息丰富的世界转化为我们可以在经验层面理解和采取行动的物体、人、概念和情感。

In neuroscience’s traditional view, categorization happens at the very end of sensory processing.

在神经科学的传统观点中，分类发生在感觉处理的最后阶段。

The brain passively receives sensory details, then decodes their features and matches them to stored templates in memory, like a clerk shuffling through a neural filing cabinet.

大脑被动地接收感官细节，然后解码它们的特征，并将它们与记忆中存储的模板进行匹配，就像职员在神经文件柜中翻来覆去一样。

But this approach to categorization struggles to account for the extraordinary flexibility in the way we assign labels to features of the world.

但这种分类方法很难解释我们为世界特征分配标签的方式的非凡灵活性。

On a clear day on an open street, a sudden rhythmic patter is a pigeon taking flight, yet when we’re walking down a dimly lit alley at night, the same sound is the shuffle of a stranger’s footsteps.

晴朗的日子，空旷的街道上，突然响起的有节奏的声音是鸽子飞翔的声音，而当我们在夜晚走在灯光昏暗的小巷时，同样的声音是陌生人拖着的脚步声。

How can the brain categorize similar sets of sensory signals in radically different ways for different situations?

大脑如何在不同情况下以完全不同的方式对相似的感觉信号进行分类？

A process called categorization allows the brain to use past experiences that resemble a new situation to implement actions and behaviors.

称为分类的过程允许大脑使用类似于新情况的过去经验来实施行动和行为。

Two of the world’s leading neuroscientists have brought an updated understanding of brain function and structure to this question.

两位世界领先的神经科学家对这个问题提出了对大脑功能和结构的最新理解。

In the pages of Nature Reviews Neuroscience , Lisa Feldman Barrett , who studies the psychology and neuroscience of emotion at Northeastern University, and Earl Miller , who studies how the brain carries out goal-directed behavior at the Massachusetts Institute of Technology, collaborated on a new view of categorization.

在《自然评论神经科学》杂志上，东北大学研究情感心理学和神经科学的丽莎·费尔德曼·巴雷特（Lisa Feldman Barrett）和麻省理工学院研究大脑如何执行目标导向行为的厄尔·米勒（Earl Miller）合作提出了一种新的分类观点。

They describe how the brain constantly reconstructs its categories moment to moment based not only on senses and memory, but on the body’s immediate physiological needs.

他们描述了大脑如何不断地重建其类别，不仅基于感官和记忆，而且基于身体的直接生理需求。

Lisa Feldman Barrett has led research showing that our nervous system constructs emotions to manage energy and behavior.

丽莎·费尔德曼·巴雷特 (Lisa Feldman Barrett) 领导的研究表明，我们的神经系统通过构建情绪来管理能量和行为。

At the heart of their framework is a counterintuitive insight.

他们的框架的核心是一种违反直觉的见解。

While we may have the impression that our categories reflect an objective reality, the researchers argue that the brain projects categories onto the world in response to the body’s survival needs.

虽然我们可能认为我们的类别反映了客观现实，但研究人员认为，大脑将类别投射到世界上是为了响应身体的生存需求。

Before we are even aware of our sensory impressions, the brain is already preparing the body to behave in ways that maintain the energetic resources that power our physiological systems.

在我们意识到我们的感官印象之前，大脑就已经让身体做好准备，以维持为我们的生理系统提供动力的能量资源。

In this way, the brain’s predictions — rather than the cumulative effects of sensory information — ultimately shape and limit how we categorize objects and features of the world.

通过这种方式，大脑的预测——而不是感官信息的累积效应——最终塑造和限制了我们对世界的物体和特征进行分类的方式。

The duo’s framework is “a fresh perspective on categorization,” said Luiz Pessoa , a neuroscientist from the University of Maryland.

马里兰大学神经科学家路易斯·佩索阿 (Luiz Pessoa) 表示，两人的框架是“分类的全新视角”。

Their idea of “the maintenance of the energetic constraints of life as fundamental to how we structure our categories,” he said, is “really important to pursue.”

他说，他们的想法“维持生命的能量约束是我们如何构建我们的类别的基础”，“非常值得追求”。

Also new in Barrett and Miller’s framework is the hypothesis that categorization doesn’t happen in a particular area of the brain, but across the entire organ and beyond.

巴雷特和米勒的框架中的另一个新假设是，分类不是发生在大脑的特定区域，而是发生在整个器官及其他区域。

To understand how organisms form categories, they argued, neuroscientists have to look all over the nervous system, from head to toe.

他们认为，为了了解生物体如何形成类别，神经科学家必须从头到脚观察整个神经系统。

For many years, these two influential neuroscientists were in the same orbit, but had never collaborated directly.

多年来，这两位颇具影响力的神经科学家一直处于同一轨道，但从未直接合作过。

Miller measures high-level electrical patterns in the brain to better understand the mechanisms driving a model of neural computation known as predictive coding .

米勒测量大脑中的高级电模式，以更好地理解驱动称为预测编码的神经计算模型的机制。

Predictive coding regards our perceptions as products of the brain’s predictions, rather than a scene built from sensory signals it passively receives.

预测编码将我们的感知视为大脑预测的产物，而不是根据它被动接收的感官信号构建的场景。

Typically, if we are experiencing a normal sensory scene (relaxing at home, say) without any novel information, the brain generates predictive signals about the environment (known as feedback signals) that will dominate incoming sensory (or “feedforward”) signals.

通常，如果我们正在经历一个正常的感官场景（例如在家放松），没有任何新奇信息，大脑会产生有关环境的预测信号（称为反馈信号），这些信号将主导传入的感官（或“前馈”）信号。

But if something unexpected happens — an event that deviates from the predictive model — the sensory feedforward signals, such as those arriving in the visual cortex, run up against the predictive feedback signals.

但是，如果发生意外情况（偏离预测模型的事件），感觉前馈信号（例如到达视觉皮层的信号）就会与预测反馈信号发生冲突。

This difference creates prediction errors.

这种差异会产生预测错误。

When sensory signals violate your predictive model, they might enter conscious experience as a feeling of surprise.

当感官信号违反你的预测模型时，它们可能会以惊讶的感觉进入意识体验。

“Your brain has to constantly make predictions about what’s going to happen in the next few seconds because it’s got to filter out most of the incoming sensory information,” Miller said.

米勒说：“你的大脑必须不断地预测接下来几秒钟会发生什么，因为它必须过滤掉大部分传入的感官信息。”

“It can’t process it all, so it’s mainly looking for things that mismatch predictions because it’s more informative.”

“它无法处理所有内容，因此它主要寻找与预测不匹配的内容，因为它提供了更多信息。”

In her work, Barrett has applied these ideas of prediction and anticipation to our theoretical understanding of emotions.

在她的工作中，巴雷特将这些预测和预期的想法应用于我们对情绪的理论理解。

Central to her work is the concept of allostasis: how an organism predictively regulates its energy use.

她工作的核心是动态平衡的概念：有机体如何预测性地调节其能量使用。

In her view, emotion categories — fear, happiness, anger — resemble predictive “action plans” that the nervous system generates to activate behaviors that have served us well before.

在她看来，情绪类别——恐惧、快乐、愤怒——类似于神经系统生成的预测性“行动计划”，以激活以前为我们服务的行为。

For example, a worked-up bodily state, with elevated heart rate, fast breathing, and tense muscles, in the context of being chased by a dog, would indicate the emotion category “fear” to activate certain behaviors and use our energy resources to fight or run away.

例如，在被狗追赶的情况下，身体处于兴奋状态，心率加快、呼吸急促、肌肉紧张，这表明情绪类别“恐惧”会激活某些行为，并使用我们的能量资源来战斗或逃跑。

Traditionally, emotions were thought to be hardwired in specific circuits present from birth.

传统上，情绪被认为是从出生起就存在于特定电路中的。

Barrett has helped show that we construct emotional categories in relation to signals both from within the body and from the external context.

巴雷特帮助表明，我们构建与来自身体内部和外部环境的信号相关的情绪类别。

In 2025, Barrett reached out to Miller to see if he’d be interested in putting their ideas together to create a new framework for categorization — one that would move beyond the filing-cabinet model to include their ideas about prediction and allostasis.

2025 年，巴雷特联系了米勒，询问他是否有兴趣将他们的想法整合在一起，创建一个新的分类框架——这个框架将超越文件柜模型，纳入他们关于预测和动态平衡的想法。

“Even really brilliant scientists can sometimes be guided by traditional thinking, which can be hard to get beyond,” she said.

“即使是非常出色的科学家有时也会受到传统思维的指导，而这种思维很难超越，”她说。

“He understood what I was saying immediately.” So she asked if he wanted to write a paper together.

“他立刻就明白了我在说什么。”于是她问他是否愿意一起写一篇论文。

Miller said yes.

米勒说是的。

“In our discussions it became clear we were cut from the same cloth,” Miller said.

“在我们的讨论中，很明显我们是同出一辙，”米勒说。

“What I love about Lisa is that she is always thinking in big-picture terms.

“我喜欢丽莎的一点是，她总是从大局出发思考。

…

……

Both of us, we are not afraid to say things that run counter to the standard way of looking at things.”

我们俩都不害怕说出与看待事物的标准方式相悖的话。”

As animals, we are tasked with surviving through change — in the environments we are embedded in, and in our own bodies as well.

作为动物，我们的任务是在变化中生存——在我们所处的环境中，以及在我们自己的身体中。

Because we have limited energy , time, and computational resources, understanding every detail of every situation is impossible.

由于我们的精力、时间和计算资源有限，因此不可能了解每种情况的每一个细节。

That means we have to take shortcuts to stay alive.

这意味着我们必须走捷径才能生存。

That’s where categories come in.

这就是类别的用武之地。

“A category is an event in which different things are treated as similar or equivalent in a particular situation,” Barrett said.

“类别是指在特定情况下不同事物被视为相似或等同的事件，”巴雷特说。

Categories might be as basic as “food,” “threat,” or “mate.” Humans have a remarkable ability to form much more detailed categories, including ones that are highly abstract: Visit a philosophy department to hear about categories such as “justice,” “truth,” and “beauty.”

类别可能像“食物”、“威胁”或“伴侣”一样基本。人类具有形成更详细类别的非凡能力，包括高度抽象的类别：访问哲学系以了解“正义”、“真理”和“美”等类别。

Earl Miller measures the dynamics of brain waves to study their role in cognition at the Massachusetts Institute of Technology.

厄尔·米勒在麻省理工学院测量脑电波的动态，以研究其在认知中的作用。

The function of categorization, Barrett said, is to use past experiences that resemble a new situation to implement behaviors that keep our bodily systems functioning.

巴雷特说，分类的功能是利用类似于新情况的过去经验来实施保持我们身体系统正常运转的行为。

The category “apple” might be defined through physical features — smooth, round, red, palm-size — but it also includes a set of behavioral policies for how we should interact with it: “good for eating” or “throw out when rotten.”

“苹果”这一类别可能是通过物理特征来定义的——光滑、圆形、红色、手掌大小——但它也包括一套我们应该如何与它互动的行为政策：“适合吃”或“烂了就扔掉”。

“The point that is important about categorization is that you do all that for a purpose,” said Timothy Buschman , a neuroscientist from Princeton University.

普林斯顿大学的神经科学家蒂莫西·布施曼 (Timothy Buschman) 表示：“分类的重要一点是，你所做的一切都是有目的的。”

“The reason for categorization is to support the current task.

“分类的原因是为了支持当前的任务。

Whatever it is that you want or whatever situation you are in, you are categorizing something according to what meaning it has for you.”

无论你想要什么，无论你处于什么情况，你都会根据它对你的意义来对它进行分类。”

But how are categories formed?

但类别是如何形成的呢？

According to the traditional view, over time our experiences construct a generalized list of attributes to form a category such as “cat” that we can use to differentiate furry creatures.

根据传统观点，随着时间的推移，我们的经验会构建一个通用的属性列表，形成一个类别，例如“猫”，我们可以用它来区分毛茸茸的生物。

By this model, we take in sensory information and pare it down to match the categorical features in our memory: “Oh, there’s this thing that’s shaped like a tail; it has ears, it has fur — now I’ve figured out all those features,” Miller suggested.

通过这个模型，我们吸收感官信息并将其缩减以匹配我们记忆中的分类特征：“哦，有一个形状像尾巴的东西；它有耳朵，它有毛皮——现在我已经弄清楚了所有这些特征，”米勒建议道。

“Now I query my memory banks: Oh, it’s a cat.”

“现在我查询我的记忆库：哦，这是一只猫。”

In Barrett and Miller’s view, however, the brain uses memories of similar past situations to predict a set of behaviors that are appropriate for the new situation before the senses have identified a traditional category.

然而，在巴雷特和米勒看来，在感官识别出传统类别之前，大脑会利用过去类似情况的记忆来预测一系列适合新情况的行为。

Imagine you are walking, and you feel a scratch on your leg.

想象一下你正在走路，感觉腿上有划痕。

If you’re in a safe place and are calm, you’ll carry on without giving the sensation a second thought.

如果你在一个安全的地方并且很平静，你就会继续前进，不会再考虑这种感觉。

But let’s say you are walking through tall grass in an unfamiliar place; your heartbeat and breathing are faster than normal.

但假设你正在一个陌生的地方穿过高高的草丛；您的心跳和呼吸比平时更快。

You feel nervous because your brain has predictively allocated resources to your body to flee any potential threats.

您感到紧张，因为您的大脑已预测性地将资源分配给您的身体以逃避任何潜在的威胁。

And by feeling nervous, you have been primed to assume the scratch is a potential threat — an insect bite or, worse, a snake.

当你感到紧张时，你就会认为划痕是一种潜在的威胁——昆虫咬伤，或者更糟糕的是，是蛇咬伤。

In both situations, the sensation of the scratch is the same, but the brain categorizes it in vastly different ways.

在这两种情况下，抓伤的感觉是相同的，但大脑以截然不同的方式对其进行分类。

The fact that a much broader set of signals — environmental and corporeal — influences how we form categories sets the stage for Barrett and Miller’s new framework.

事实上，更广泛的信号——环境的和物质的——影响我们如何形成类别，这一事实为巴雷特和米勒的新框架奠定了基础。

Their proposal is based on years of anatomical , electrophysiological , and brain-imaging research from their labs and others.

他们的提议基于他们的实验室和其他实验室多年的解剖学、电生理学和脑成像研究。

They have built an argument that categorization is an ongoing predictive tool by which nervous systems guide behavior to keep the organism alive.

他们提出了一个论点，即分类是一种持续的预测工具，神经系统通过它来指导行为以保持有机体的生存。

Some of their evidence is at the neuronal level.

他们的一些证据是在神经元水平上的。

They cite research showing how brain connectivity is structured to favor predictive signals over sensory ones.

他们引用的研究表明，大脑连接的结构如何有利于预测信号而不是感觉信号。

For example, there are many more connections between neurons that facilitate the feedback flow of internally generated signals than between those handling the feedforward signals that originate from our senses.

例如，促进内部生成信号的反馈流的神经元之间的连接比处理源自我们感官的前馈信号的神经元之间的连接要多得多。

Even within the visual cortex, 90% of synaptic connections facilitate feedback signaling.

即使在视觉皮层内，90% 的突触连接也促进反馈信号传导。

That suggests that the brain is structured for predicting categories, perceptions, and behaviors, rather than for categorizing in reaction to stimuli.

这表明大脑的结构是为了预测类别、感知和行为，而不是根据对刺激的反应进行分类。

On the feedforward side, the authors describe how circuits that carry sensory information undergo extensive compression as they travel deeper into the brain.

在前馈方面，作者描述了携带感觉信息的电路在深入大脑时如何经历广泛的压缩。

They encounter many small neurons that are densely packed but scarcely connected, and their signals get whittled down through fewer, bigger, better-connected neurons.

它们遇到许多密集但几乎没有连接的小神经元，它们的信号通过更少、更大、连接更好的神经元而被削弱。

Traveling along this gradient, sensory signals become highly abstracted — and eventually reach integrative regions, which coordinate information from the senses, memory, and body to guide behavior.

沿着这个梯度传播，感觉信号变得高度抽象，并最终到达整合区域，协调来自感官、记忆和身体的信息来指导行为。

The authors then draw on electrophysiological work from Miller’s lab to show how feedback and feedforward signals interact.

然后，作者利用米勒实验室的电生理学工作来展示反馈和前馈信号如何相互作用。

In a zoomed-out view of the brain, traveling waves, which represent the activity of many neurons, carry information about goals, plans, and the energetic state of the body.

在大脑的缩小视图中，行波代表许多神经元的活动，携带有关目标、计划和身体能量状态的信息。

These waves then encounter and combine with others carrying information from the senses.

然后这些波与携带感官信息的其他波相遇并结合。

According to Barrett and Miller, interaction between waves of information cascading across the brain can start to explain how the same set of external sensory signals might be categorized differently in different situations.

巴雷特和米勒认为，大脑中级联信息波之间的相互作用可以开始解释同一组外部感官信号在不同情况下如何进行不同的分类。

If you are exhausted and hungry, a bruised, overripe apple is categorized as much-needed food.

如果你又累又饿，一个伤痕累累、过熟的苹果就是你急需的食物。

If you are well rested and energetically satiated, it’s a piece of subpar fruit to be passed over.

如果你休息得很好并且精力充沛，那么这就是一个可以忽略的劣质水果。

“Our internal state is just as, if not more, important than sensory input to us,” said Sandra Reinert of University College London, who has studied categorization in rodents .

研究啮齿类动物分类的伦敦大学学院桑德拉·赖纳特 (Sandra Reinert) 表示：“对我们来说，我们的内部状态与感官输入一样重要，甚至更重要。”

“It’s completely intuitive to see how an internal context remaps our rules [for how we form a category].”

“看到内部环境如何重新映射我们的规则（我们如何形成一个类别）是完全直观的。”

A central component of Barrett and Miller’s framework is the origin of predictive signals in the brain.

巴雷特和米勒框架的核心组成部分是大脑中预测信号的起源。

In most neuroscientific literature on visual categorization, integrative regions of the cortex (located on the outer surface of the brain) are considered the source of predictive expectations.

在大多数关于视觉分类的神经科学文献中，皮层的整合区域（位于大脑的外表面）被认为是预测期望的来源。

“They [Barrett and Miller] make a different move,” Pessoa said.

“他们（巴雷特和米勒）采取了不同的行动，”佩索阿说。

“They take the limbic core as the source of these predictive signals.”

“他们将边缘核心作为这些预测信号的来源。”

The limbic core includes structures deep in the brain that combine signals carrying information from the body, senses, memory, and higher-level cognitive functions.

边缘系统核心包括大脑深处的结构，这些结构结合了携带来自身体、感官、记忆和高级认知功能的信息的信号。

This integrative network is both close to and highly connected to the hypothalamus, an ancient part of the brain that monitors basic bodily functions, such as body temperature, heart rate, and hunger.

这个综合网络既靠近下丘脑，又与下丘脑高度相连，下丘脑是大脑的一个古老部分，负责监测体温、心率和饥饿等基本身体功能。

Because the hypothalamus helps inform the cortex — the folded, outermost layer of the brain, which is involved in high-level cognition such as executive control — about the body’s energetic state, the authors hypothesized that the limbic core facilitates the brain’s ability to anticipate the body’s energy needs.

由于下丘脑有助于向皮质（大脑的折叠的最外层，参与执行控制等高级认知）通报身体的能量状态，因此作者假设边缘核心有助于大脑预测身体能量需求的能力。

The limbic core is where the most compressed summaries of internal and external signals intersect.

边缘系统核心是内部和外部信号最压缩的汇总相交的地方。

You can imagine the nervous system as being structured like two funnels, shaped like a bow tie, that meet at their narrowest point — the limbic core, the point of highest compression.

你可以想象神经系统的结构就像两个漏斗，形状像领结，在最窄的点处相遇——边缘核心，即压力最高的点。

Just as visual information is compressed as it travels through the visual cortex and deeper into the brain, physiological signals are compressed as they travel from the body up the vagus nerve and into the brain.

正如视觉信息在穿过视觉皮层并深入大脑时会被压缩一样，生理信号在从身体向上经过迷走神经并进入大脑时也会被压缩。

They then reach the limbic core, where they are integrated with compressed sensory information.

然后它们到达边缘核心，在那里与压缩的感觉信息整合。

Together, these signals generate an appropriate category and its associated behavior.

这些信号共同生成适当的类别及其相关行为。

Depending on the category — a visual category like “apple,” an abstract category like “justice,” or an emotion category like “awe” — the interaction of the predictive feedback signals with the compressed sensory feedforward signals can happen in any number of neural regions.

根据类别——像“苹果”这样的视觉类别，像“正义”这样的抽象类别，或者像“敬畏”这样的情感类别——预测反馈信号与压缩的感觉前馈信号的相互作用可以发生在任意数量的神经区域中。

The process spans the entire nervous system, the authors wrote.

作者写道，这个过程跨越整个神经系统。

Barrett emphasized that selecting any particular starting point for this process is arbitrary.

巴雷特强调，为这个过程选择任何特定的起点是任意的。

“For the sake of having a scientific discussion, you have to say, ‘I’m going to pick this point as the start,’” Barrett said of the limbic core.

“为了进行科学讨论，你必须说，‘我要选择这一点作为起点，’”巴雷特谈到边缘核心时说道。

“We could’ve easily just picked some other point.”

“我们可以轻松地选择其他一些点。”

Miller agreed.

米勒同意了。

“Lisa and I are arguing that this is something that is happening at every level, and it’s happening in lots of places, and it’s happening as a result of these opposing, interacting flows of information — feedforward and feedback,” he said.

“丽莎和我认为，这种情况在各个层面都在发生，而且在很多地方都在发生，而且它的发生是这些对立、相互作用的信息流——前馈和反馈的结果，”他说。

“It’s not at one end of the brain, which it has traditionally been thought.”

“它并不像传统上认为的那样位于大脑的一端。”

Barrett and Miller are asking the field to move beyond a model in which categories are stored like documents in the brain’s filing cabinet.

巴雷特和米勒要求该领域超越一种将类别像文件一样存储在大脑文件柜中的模型。

From the way the brain is structured to compress sensory information, to how it predicts and prepares the body to respond, categorization is “baked” in, the authors wrote.

作者写道，从大脑压缩感官信息的结构方式，到它如何预测和准备身体做出反应，分类都是“烘焙”出来的。

It is a core organizational principle of the nervous system that draws on memory, perception, and behavior to help us survive well in an information-rich world.

它是神经系统的核心组织原则，利用记忆、感知和行为来帮助我们在信息丰富的世界中生存。

Get highlights of the most important news delivered to your email inbox

获取发送到您的电子邮件收件箱的最重要新闻的摘要

Quanta Magazine moderates comments to facilitate an informed, substantive, civil conversation.

《Quanta》杂志对评论进行审核，以促进知情、实质性、文明的对话。

Abusive, profane, self-promotional, misleading, incoherent or off-topic comments will be rejected.

辱骂、亵渎、自我推销、误导、语无伦次或偏离主题的评论将被拒绝。

Moderators are staffed during regular business hours (New York time) and can only accept comments written in English.

版主在正常工作时间（纽约时间）有工作人员，并且只能接受英文评论。