你走过小区的垃圾桶，那边趴着几只野猫。有的隔着五米就窜走了，有的会停下来看你一眼，你再走近一点它才不情愿地挪开。
这个「能让你靠多近」的距离，在同一群猫里差别很大，而且能拿尺子量。一万多年前，人类营地旁边的狼群里也有这个差别。
接下来的事听起来像一个好心人的故事，其实一点都不是。
营地边上堆着吃不完扔掉的骨头和内脏。对狼来说这是一份稳定、好拿、不用追的食物。能不能拿到它，取决于一件很具体的事：离人多近就开始恐慌。
恐慌门槛高一点的那些多吃到一口，多养活一点幼崽；幼崽平均又更不怕人一点。一代代下来，这群狼和留在野外的那些分开了。这里没有谁做决定，也没有「为了吃到剩饭而变温顺」这回事：不怕人的变异随机冒出来，营地只是让带着它的那些留下了更多后代。第一步很可能根本不是人做的，是狼自己走过来的。
有一件实物把这条线钉住了。1914 年德国波恩附近挖出一座墓，一男一女和两只狗葬在一起，年代约一万四千两百年前。2018 年有人重查其中那只幼犬的牙齿，发现它十九周龄左右得了犬瘟热——一种病程三周、致死率极高的病——却又活了八周多，牙齿上的病程刻痕说明那八周里有人一直在喂它。这已经不是「跟着捡食」的关系。基因那边给的是另一种刻度：2020 年测出的二十七个古代狗基因组显示，到大约一万一千年前，狗已经分成至少五个差别很大的支系；分化成五支要很长时间，所以驯化本身比这早得多。在哪里开始、发生过一次还是几次，还有争论。
EN The first step of dog domestication was probably not taken by people. Camp refuse is a stable, easy food source, and access to it depended on one measurable trait: how close an animal will come before panicking. Wolves with a higher panic threshold ate more and raised more pups; no wolf decided anything. One hard object anchors the timeline: a grave near Bonn, Germany, dated to roughly 14,200 years ago, holds two humans and two dogs. A 2018 re-examination of one puppy's teeth found it contracted canine distemper at about 19 weeks — a disease with a three-week course and very high mortality — yet survived a further eight weeks, which requires sustained human care. Genetics gives another scale: 27 ancient dog genomes published in 2020 show at least five deeply distinct lineages already in place by about 11,000 years ago, so domestication itself is considerably older. Where it began, and whether once or several times, is still debated.
1959 年起，苏联新西伯利亚的贝利亚耶夫和特鲁特养了一批银狐，每一代只按一个标准挑：伸手过去，它躲不躲、凶不凶。不挑毛色，不挑耳朵，不挑体型。几十代之后，这群狐狸身上出现了一串没人挑过的东西——耳朵软塌、尾巴上卷、毛色出现白斑、繁殖季节变长。这一串在很多家养动物身上都能看到一部分，有个名字叫驯化综合征。
为什么拧一个旋钮会带出一串输出？目前最流行的解释盯住了一批细胞：胚胎很早就在背上分出一群会到处迁移的细胞，它们最后长成几样互不相干的东西，下面这张图画的就是各自的去处。如果你挑的是「应激反应弱一点」，实际挑中的可能是「这群细胞少一点、迁移得慢一点」，那么软骨、色素、颌骨就被一起带着动了。没有人挑过耳朵。
得老实说：这个解释好看，证据还不够硬。2021 年有过一场公开来回，质疑方指出，在家养动物身上找到的那些相关基因只能说明由这批细胞派生的部位变了，不能证明别的性状是被同一个机制连带的。目前没有定论。
那批狐狸自己还出了问题：2019 年底有文章指出，贝利亚耶夫的银狐来自加拿大一处已经养了几十年的毛皮养殖场，那里的狐狸在俄国实验开始之前就被记录过温顺、也出现过白斑。「只挑温顺、凭空长出一整套」这个教科书版本因此被削掉一大块。但核心机制不靠狐狸撑着——你挑一个性状会连带挪动和它连着的其他性状，育种里天天遇到。
EN From 1959, Belyaev and Trut selected silver foxes in Novosibirsk on a single criterion: whether the animal flinched or snapped when a hand reached in. Within a few dozen generations the line showed floppy ears, curled tails, piebald coats and longer breeding seasons — none of which anyone selected for. The most popular explanation points at neural crest cells, a population that migrates out of the embryonic back and gives rise to part of the adrenal glands (the acute stress switch), ear and nose cartilage, skin pigment cells and part of the jaw and teeth; selecting for a weaker stress response may amount to selecting for fewer or slower-migrating cells, dragging the rest along. The evidence is not settled: a 2021 exchange in Genetics argued the gene-level support only shows that neural-crest-derived structures changed, not that other traits are pleiotropically linked, and a 2023 paper proposed disrupted reproductive rhythms instead. The foxes themselves are also compromised — a 2019 analysis showed Belyaev's stock came from Prince Edward Island fur farms where foxes had been bred for decades and where tameness and piebald coats were recorded before the Russian experiment began. The underlying mechanism does not depend on the foxes: selecting one trait moves correlated traits with it, as every breeder knows.
野生小麦和大麦成熟之后，穗子会自己一节节断开，籽粒撒到地上——撒到地上才能明年发芽。
人拿镰刀去割，割回来的是那些还没断、籽粒还挂在穗上的。它们被带回营地，第二年被种下去。于是「穗子不自己断」这个在野外等于绝后的性状，在人手里成了唯一能传下去的性状，旁边几样东西也跟着被挑走：种子不再分几年零星发芽而是一起发芽，种皮变薄。
这里没有哪一代人是在「驯化小麦」。近东的考古记录里，大麦在公元前 9500 年之前几乎全是会自己散落的野生型；到公元前 9000 到 8500 年，不散落的穗轴碎片在少数遗址里占到三成；要到公元前 7400 年之后，才有遗址是清一色的不散落。前后两千年，中间每一代人只是在割穗子。
玉米更极端：祖先墨西哥类蜀黍的每一粒籽外面都裹着一层硬壳，一个基因的变化让硬壳张开——从那以后籽粒既散不出去，也失去了保护。
还有一件顺带发生的事值得单独记住：拿去种的只是野生群体里很小一撮，一代代下来，同一种作物内部的差别越来越小。整片田是同一个答案，病原体只要解开这一个答案就能通吃——红皇后那期讲的军备竞赛在这里变成一边倒，病原体一年迭代好几轮，作物的基因组被我们冻在原地。十九世纪爱尔兰的马铃薯就这么塌了。
EN Wild wheat and barley shatter at maturity, dropping their grain — that is how they survive. A sickle brings home the ears that have not shattered, and those are the grains sown next year, so a trait that is lethal in the wild becomes the only route forward under cultivation, dragging along synchronous germination, thinner seed coats and self-pollination. No generation was "domesticating wheat". In Near Eastern assemblages, barley is almost entirely wild-type before 9500 BC; non-shattering rachis fragments reach about 30% at a few sites between 9000 and 8500 BC; only after 7400 BC do some sites show fully non-shattering populations. Two thousand years for one trait. Maize is starker: its ancestor teosinte has few grains, each sealed in a hard casing. One gene change opens the casing, another turns a branching bush into a single stalk — after which the plant can neither disperse nor protect its seed. One side effect deserves its own note: only a small slice of the wild population was ever planted, so variation within a crop keeps shrinking. A whole field is one answer, and a pathogen that solves it takes everything — the arms race of the Red Queen becomes one-sided when the pathogen iterates yearly and the crop genome is frozen by us. Ireland's potatoes collapsed exactly that way.
把智人的头骨和三十万年前的早期化石、和尼安德特人摆在一起，方向相当一致：眉弓变平、面部前突减小、颅骨变圆、牙齿变小。这一串和家养动物那套有不小的重叠。行为上还有一件更扎眼的事：你每天和几百个不认识的人挤在一节车厢里，一路不打架；把同样密度的陌生黑猩猩关在一起会出人命。
自我驯化假说说的就是这件事：筛选压来自群体内部。一种具体版本是，语言和结盟让一群人可以商量着一起对付本群里最霸道的那个，于是最容易当场发作的那些人留下的后代更少。这里没有「为了群体好」这回事——起作用的只是各自的后代数量：被联手对付掉的那个，后代就少。
最硬的一条旁证是倭黑猩猩。它和黑猩猩是近亲，大约一两百万年前分开，全程没有人参与；它对同类的攻击性明显更低，颅骨和牙齿也朝偏幼态的方向变了一些。2012 年有人正式把它写成一个自然发生的自我驯化案例——有这个自然实验就意味着，这套变化不需要有人来筛。
为什么它到现在还只是假说？狗那边你说得出谁在筛、筛的是哪个性状；人这边「谁在筛」全靠推断，性状清单也有事后拼凑的嫌疑——颅骨变圆可以有别的解释，比如脑发育节奏改变。它有意思、有部分旁证，远没有定论。
还有一条必须说明白：这个假说拿的是物种之间的对比，问的是人和黑猩猩为什么不一样，尺度是几十万年。任何把它挪到「哪群人比哪群人更驯化」上的读法都没有证据——一个是物种间的差异，一个是群体内部的差异，两个尺度根本对不上。
EN Set modern human skulls beside 300,000-year-old fossils and beside Neanderthals and the direction is consistent: flatter brow ridges, less facial projection, rounder braincase, smaller teeth — a list that overlaps noticeably with domesticated animals. Behaviourally, the striking fact is that you ride a packed train with hundreds of strangers and nobody fights; chimpanzees at that density would kill. The self-domestication hypothesis holds that the selection came from inside the group: language and coalitions let people jointly deal with the most domineering member, so the most reactively aggressive left fewer descendants. Nothing here runs "for the good of the group" — what operates is individual descendant counts. The strongest comparative leg is the bonobo, which split from chimpanzees roughly one to two million years ago with no human involvement, is markedly less aggressive, and shows some paedomorphic cranial and dental shifts; a 2012 paper framed it as a naturally occurring case. Why it remains a hypothesis: with dogs you can name the selector and the trait, while here the selector is inferred; the trait list looks partly assembled after the fact; and genomic searches for overlap between human and domesticate selection targets remain statistically contested. One thing must be stated plainly: the hypothesis compares species across hundreds of thousands of years. Any reading that shifts it onto differences between human populations has no evidence behind it and no bearing on what the hypothesis explains — the two scales simply do not match.