神经科学与心理学2022pnas.org
人脑中通信成本是计算成本的 35 倍
Communication costs are 35-fold computational costs in the human brain (2021)
进阶 · 约 60 分钟读完前置知识：热力学基础、计算理论基本概念、神经科学基础
导读
这篇 PNAS 论文从热力学角度估算人脑的能量开销，认为神经元之间的通信成本约为其计算成本的 35 倍，并以此讨论大脑通信效率对神经结构的影响。评论者指出其论证依赖大量假设，且热力学层面的分析与逻辑计算模型之间的可比性存疑。
HN 关注点：论文用热力学分析比较大脑的通信与计算成本，这一大胆的量化结论引发了关于其方法论是否站得住脚的激烈讨论。
HN 评论精选
这几条评论里：质疑 1 补充 3
质疑认为论文是伪科学：热力学分析与逻辑计算模型不可比，论证建立在大量站不住脚的假设上
我有非理科博士的朋友，他们绝对相信发表在科学期刊上的东西一定可靠。读之前我就想“该不会又是伪科学吧”……就我所见，基本上是的。1. 逻辑比特不是热力学比特，对大脑做热力学分析只能和 CPU 的热力学分析相比。2. 每个以“假设、猜想”开头的句子都给整个项目引入了致命假设，几乎没有站得住脚的。认为单个神经元在“计算”而它与他人的“通信”在逻辑意义上不是计算的，这显然是错的。3. 频繁引用“计算”的热力学极限作为与大脑任意部分比较的基线是没有意义的。大脑的热力学效率之所以有趣，只在于任何大脑的逻辑模型似乎都需要多得多的计算资源。
mjburgess查看英文原文
I have friends who are non-science PhDs and absolutely believe anything published in a science journal must be absolutely sound. Before I started reading I thought, "alas, its going to be pseudo-science again right?".... and as far as I can see, yes, basically it is. 1. Logical bits are not thermodynamic bits, so a thermodynamic analysis of the brain can only be compared to a thermodynamic analysis of CPUs (if one wishes to compare at all) 2. Every phrase which begins "suppose, assume, conjecture" introduces fatal assuptions into the whole project, extremely few are defensible. The idea that a single neurone is "computing" whilst its "communication" with others is non-computational in a logical sense is clearly false. One can make a thermodynmic distinction between energy "of the neurone" and "of thier 'communication'" this distinction has no relevance to a computational-logical model of the brain as a computational system. 3. The frequent reference to thermodynamic limits of "computation" (in a logical sense) as a baseline for comparison with (abitary) parts of the brain, is meaningless. The thermodynamic efficiency of the brain is interesting only insofar as any possible logical model of the brain seems to imply vastly more "computational resources". And vastly more thermodynamic resources than CPUs have. Physical limits on theoretical computation pertain, if they are even themsevles coherent (which is disputed), to the absolute minimal possible "piece" of reality, not even, i'd say, to any measurable phenomenon. As soon as a system has to engage in measurement, i'd say it would be millions+ times less "efficient" than 'physical limit's would suggest. One "trick" to see through the pseudoscience of computational neurobiology is simply to apply its methods to actual CPUs and computers. With the above assumptions, this paper would conclude that only one operation in a transistor is "computing" anything, the energy required to do that is "the energy of computation"... and the rest of the energy used across CPU(-RAM-etc.) was "merely communication". As-if all algorithms were logically specified as a purely parallel series of switch-flips. No, algorithms (in a computation-logic) sense have nothing to do with switch flips. And their implementation on digital computers requires many, serial and parallel and "thermodynamic communication" between them to perform the computation in question.
补充类比无线传感器网络：通信比计算耗能时，本地多算以减少通信是常见策略
这让我想起当年的无线传感器网络领域，主要挑战是无线通信通常比本地计算耗能得多，因此有动力在本地执行额外计算以减少通信（例如过滤或聚合数据）。
anonymousDan查看英文原文
It reminds me a bit of the field of wireless sensor networks from back in the day, where the main challenge was that wireless communication was typically much more energy intensive than local computation requirements and so there was an incentive to perform additional computation locally that might reduce communication (e.g. to filter or aggregate data).
补充认为结论与连接成本驱动网络模块化的研究一致，但实证基础薄弱需谨慎
这与[0]的观点一致，即进化过程产生的小世界网络的模块化很大程度上是连接成本的结果。遗憾的是这一领域的科学实证基础如此薄弱，但这并不意味着它是错的，只是需要谨慎看待。
rileyphone查看英文原文
This is in line with [0], which argues that modularity in small-world networks produced by evolutionary processes is largely a result of connection costs. It’s unfortunate that science here has such a limited empirical footing given the importance of the knowledge, but that doesn’t make it wrong, just circumspect. 0. https://royalsocietypublishing.org/doi/10.1098/rspb.2012.2863
补充澄清论文指的是神经元间通信，而非人际交流
是神经元内部的通信，不是“和人说话”（尽管那可能也很累）。
anotherhue查看英文原文
intra-neuronal communication, not 'talking to people' (tiring as that may be).
历次讨论
- 2022Hacker News 讨论帖151 分 · 71 评
同领域还值得读
神经科学家能理解微处理器吗？（2017）
Could a Neuroscientist Understand a Microprocessor? (2017)
提出一个思想实验：把神经科学常用的分析方法（损伤实验、电生理、功能定位等）应用到一个我们完全了解其原理的工程系统——6502 微处理器（运行游戏）上，检验这些方法能否真正“理解”…
神经科学与心理学2017 · 411 分 · 135 评进阶 · 约 60 分钟编辑精选大脑衰老呈现非线性转变，提示中年“关键窗口”
Brain aging shows nonlinear transitions, suggesting a midlife "critical window"
这篇 PNAS 论文研究大脑衰老的非线性变化，发现大脑老化并非匀速进行，而是在中年（约 40-60 岁）存在一个关键的"临界窗口期"，此时干预可能有效。
神经科学与心理学2025 · 295 分 · 159 评进阶 · 约 75 分钟编辑精选社交网络中大规模情绪传染的证据（2014）
Evidence of massive-scale emotional contagion through social networks (2014)
这篇 2014 年发表于 PNAS 的论文研究了 Facebook 上约 70 万用户的信息流中，减少正面或负面内容是否会传染性地影响用户自己发布内容的情绪。
神经科学与心理学2020 · 285 分 · 72 评入门 · 约 45 分钟编辑精选裸盖菇素使人类大脑去同步化
Psilocybin desynchronizes the human brain
这篇 Nature 研究用个体化的 fMRI 精密成像追踪受试者服用裸盖菇素（psilocybin）前后的脑功能连接变化，发现药物会造成大范围的脑网络「去同步化」，尤其是默认模式…
神经科学与心理学2024 · 273 分 · 259 评进阶 · 约 60 分钟编辑精选对智力的赞扬会削弱儿童的动机与表现（1998）
Praise for Intelligence Can Undermine Children's Motivation (1998)
这篇 1998 年的经典论文（Dweck 等人）通过实验研究表扬方式对儿童动机的影响：夸孩子“聪明”（表扬智力）会让他们回避挑战、害怕失败，而表扬努力则更能培养坚持性。
神经科学与心理学2016 · 253 分 · 155 评入门 · 约 45 分钟编辑精选