秋天的林子里,两头公鹿隔着几十米对峙。先是吼,你一声我一声,能吼上好几分钟;然后并排走一段,各自看对方的体格;只有到这一步还分不出高下,才低头顶角。大多数对峙停在前两步。
一个流行过很久的解释是:它们懂得克制,不然同类之间早就自相残杀完了。这个说法有个致命毛病——一头不肯克制、见谁顶谁的公鹿,在满是克制者的林子里会占尽便宜、留下更多后代。选择记的是「谁留下的后代多」这一本账,「这样对物种好」根本进不了这本账。
所以要换个问法。一种打法划不划算,没法单独回答,得看林子里其他鹿正在用什么打法。
先把它压到路口上:两辆车同时到,都想先过。你可以抢,也可以让。都抢,撞了,两边都要修车;一抢一让,抢的先走,让的多等一会;都让,互相比划几下,各自耽误一点点。
1973 年 Maynard Smith 和 Price 在《自然》上把动物冲突写成了这个形状。他们没有去猜哪种性子好,而是让几种打法在电脑里两两对局:一种一上来就升级到真打、除非对方先退;一种只虚张声势,对方一硬就走;还有几种介于中间。
算下来的结果是这样的:假如林子里全是「让」,混进来一个「抢」的,它每次都赢、一次伤都不用受,抢的比例于是往上走。可抢的多起来之后,抢的主要在跟抢的对上,真打的代价开始实打实落到自己身上;这时候「让」虽然次次拿小头,却不用去医院。两边一拉扯,比例就停在中间某处。
谁多谁吃亏——这就是所谓频率依赖选择:一个做法带来多少后代,随着群体里有多少个体在用它而变。它不是一个固定的数,这一点几乎所有关于「最优」的直觉都是在偷偷假设它是固定的。
停在哪,由两个数拉扯:赢一次能拿到多少,真打一次要赔多少。模型给出的一个预言是,一次真打的代价越高,群体里能维持的「抢」越少。这条在物种之间做干净的检验很难,别当成已经验完的结论;但它至少说明,对峙停在吼叫那一步不需要任何一方「顾全大局」。
EN Two individuals meet over one resource; each can escalate or display and back off. Maynard Smith and Price (1973, Nature) simulated such contests and found the outcome cannot be read off a strategy alone. In a population of pure displayers, an escalator wins every time at no cost, so escalators spread; once escalators are common they mostly meet each other and pay the injury cost, so backing down stops being a loss. The proportion settles somewhere in between. That is frequency-dependent selection: how many offspring a behaviour yields depends on how many others in the population are doing it. Where it settles is pushed by two quantities — the value of winning and the cost of a real fight — and the model predicts that costlier fights leave less room for escalation, though clean comparative tests are hard. Nothing here requires restraint on anyone's part, and the resting point is not the one that is best for the group.
这就换掉了「什么打法最好」这个问题。能问的只剩一个:假设群体里大家都在用某个做法,任何一个少数的新做法混进来,能不能占到便宜?占不到,这个做法就守得住。守得住的做法叫演化稳定策略。
要留意它跟「最好」的差别。全是「让」的群体平均过得最舒服,但它守不住,一个抢的进来就赚翻;真正停住的那个混合点,群体平均收益反而更低。稳定的意思是「挤不进新的」,不是「对大家好」。这句话是这套东西里最容易被讲丢的一句。
野外最干净的例子是美国西部一种小蜥蜴,雄性的喉咙有三种颜色。1996 年 Sinervo 和 Lively 报告:橙喉最凶,占的地盘大,但大到自己看不过来;蓝喉地盘小,看得紧;黄喉不占地盘,长得像雌的,混进别人地盘里偷偷交配。橙喉能从蓝喉手里抢地盘,蓝喉盯得紧、黄喉偷不进去,而橙喉的大地盘正好被黄喉钻空子——石头剪子布,一圈套死。哪一种都能在稀少的时候挤进去,多了又被下一种克,于是三种比例在野外转圈,大约六年一轮。
另一个例子在坦噶尼喀湖:有种鱼靠啃别的鱼身上的鳞片过活,嘴是歪的,一部分个体从猎物右侧下口,另一部分从左侧。1993 年 Hori 报告,哪一侧的多,猎物就更提防哪一侧,少的那一侧得手率更高,两者的比例围着一半上下摆。要说清强度:频率依赖这部分站得住,但后来对「左右是不是两个界限分明的类型、是不是简单遗传」有不少质疑——在大样本里,嘴的偏折更像连续分布。这一半别当定论。
整体强度也说清楚:这类模型多,野外能干净测出来的少——模型说的是什么条件下什么守得住,不等于某个动物就在跑它。
EN The question stops being "which strategy is best" and becomes "which strategy, once common, cannot be invaded by a rare alternative" — an evolutionarily stable strategy. Note the gap from optimal: a population of pure displayers does best on average yet cannot hold, while the mixture that does hold pays worse. Stability means uninvadable, not good for anyone. It can be realised as a mixed population or as individuals randomising. Sinervo and Lively (1996, Nature) described three male throat colours in a side-blotched lizard — aggressive orange holding territories too large to guard, blue guarding small ones closely, yellow female-mimicking sneakers — locked in a rock-paper-scissors loop with morph frequencies cycling over roughly six years, so no resting point exists at all. Hori (1993, Science) reported frequency-dependent oscillation around parity in the left- and right-mouthed forms of a scale-eating cichlid; the frequency dependence holds up, while the claim of two discrete, simply inherited forms has been contested.
养牛的人都知道,留少数几头公牛就够用了,母牛越多,一年能生的小牛越多。可自然界里绝大多数动物的性别比接近一半一半,连那些一头雄性能配许多雌性、大部分雄性一辈子交配不上的物种也是。从「群体多生几只」的角度看,这是极大的浪费。
Fisher 在 1930 年把这笔账写清楚了(更早 Düsing 在 1884 年已写过类似的话)。每个后代都有一父一母,所以全体儿子加起来贡献的孙辈,和全体女儿加起来的一样多。假如群体里女儿偏多,平均下来一个儿子分到的孙辈就比一个女儿多——这时候偏爱生儿子的父母留下更多孙辈,儿子的比例就往上走;反过来同理。两边一推,停在一半。
注意这个停下来的点对群体不划算,它照样停在那儿。原因在「一个打法能不能活下去,看它挡不挡得住少数派」那节里已经说过:选择比的是同一个群体内部谁的后代份额多,不是哪个群体涨得快。这是「为了物种好」那类说法最干净的反例——真按那个逻辑,性别比早该偏向雌性了。
这套逻辑还给出了能去野外量的预言。1967 年 Hamilton 提了这么一种情形:如果一位母亲的儿子们只能在一个封闭的小场子里跟自己的姐妹交配,那么多生儿子等于让儿子互相抢同一批姐妹,只要生够用的儿子就行。寄生蜂正是这样过日子的——把卵产进一个寄主里,孵出来的后代就在这一小块地方配对。1980 年 Werren 在一种寄生蜂上量到:一只母蜂单独占一个寄主时,儿子只占一成上下;同一个寄主上产卵的母蜂越多,儿子的比例越往一半靠,而且是同一只母蜂在不同处境下当场调的。
三节合起来是同一句话:收益不是常数,它随着有多少人在用同一个做法而变。所以「最优」在这里不是一个能单独成立的词,能问的只有挡不挡得住少数派;而挡得住的那个点,跟对大家好不好没有关系。
EN Breeders keep few males, since a population grows faster with mostly females — yet most animals sit near an even sex ratio, including species where most males never mate. Fisher (1930), anticipated by Düsing (1884), gave the account: every offspring has one father and one mother, so all sons together contribute as many grandchildren as all daughters together. If daughters are in surplus, each son averages more grandchildren, parents biased towards sons leave more descendants, and the ratio climbs back — and vice versa. The resting point is wasteful for the group and stays put anyway, because selection compares reproductive shares within a population, not growth rates between populations. Hamilton (1967) added a testable twist: where sons compete only with their brothers for their own sisters, a mother should make just enough sons. Werren (1980) found single-foundress parasitoid wasps producing around a tenth sons, shifting towards parity as more mothers laid in the same host.