The Evolution of Behavior

The sentence "why does it do that" is really four questions that cannot stand in for each other

Back to the bird outside your window. Four kinds of answer, all of them true.

One: what is happening inside its body right now. Days lengthen, light signals reach the brain, a cascade of hormones follows, and the brain regions that control song swell during the breeding season and shrink again afterwards — something first measured in canaries.

Two: how this particular bird came to sing this way. Who it heard as a chick, during which stretch of time, and how long it practised afterwards.

Three: whether individuals who sing like this leave more offspring or fewer. This is the slot that goes crooked most easily. Do not say it sings in order to attract a mate — that is not how the ledger works. What actually happens: males hatched each year vary in when and how hard they sing, some of them end up holding better territories, pairing up and fledging young, and their variants make up a slightly larger share of the next generation. Nobody wants anything. Some simply leave more descendants.

Four: when this behavior appeared in the lineage. Whether close relatives do anything like it, and on which branch it first showed up.

In 1963 Niko Tinbergen wrote these four questions out as a checklist, and it became the basic discipline of the field. Its value is not that there are four boxes. Its value is the pressure it puts on you to say first which box you are answering.

An honest note: why the pre-dawn chorus is so dense is still contested — too dark to hunt insects, cool morning air carries sound further, having the energy to sing then advertises that last night went well. The third cell is the hardest of the four to nail down, because a plausible story is cheap.

"Is this instinct or is it learned" is a question that yields nothing

The white-crowned sparrow is a common North American bird. Raise a newly hatched male alone, so that he never hears another of his own kind, and he still sings as an adult. What comes out is blurred and impoverished, but it is still recognizably a sparrow's song and would not be mistaken for another species. So "learns nothing, sings nothing" is wrong: he already carries a rough sketch.

The decisive step comes next. Play him recordings of his own species between roughly day ten and day fifty after hatching, and as an adult he sings the song well; play them outside that stretch and nothing takes. And he is picky — recordings of other species during that same window are largely refused. He carries not only a sketch but a filter, which already knows which kind of sound is worth keeping.

Then it gets harder. Having heard is not enough. In 1965 someone did the following: let a chick hear a normal tutor, then deafen him before he first sang. He could not produce the song. He needs to hear his own voice and tune it, pass after pass, against the remembered model.

So answer the question: what fraction of this song is innate? The template is built in, the content is heard, the tuning runs on listening to itself. The question yields nothing, rather like asking what share of a biscuit's shape is flour. In 1953 someone put this into a fierce critique, and the innate-versus-learned split has not been usable in its naive form since.

One more result dissolves even the window. That "day ten to day fifty" figure came from tape recordings. In 1984 researchers swapped in a live tutor — a real bird next door, visible and able to interact. The birds learned past day fifty, and some learned the song of an entirely different species. The window is not a welded door. It is closer to a cost curve: past that stretch, the same input needs a stronger social signal behind it to move anything.

Learning itself is a fussy eater — some things stick in one trial, others will not stick in a hundred

A beautifully built experiment from 1966. Rats were given water of two kinds: one sweetened, the other rigged so that licking the spout set off a flashing lamp and a clicking relay — call it bright-noisy water. The consequence came in two kinds too: half the animals were made nauseous afterwards, half got a mild shock.

The result was not that everyone learned. It crossed over. Sweet water plus nausea, and the rat avoids the taste from then on; sweet water plus shock, and it drinks on as before. Bright-noisy water plus shock, and it avoids the lamp and the click; bright-noisy water plus nausea, and it drinks on as before.

Same animal, same business of pairing two events — and whether it can be learned depends on which two. What is more, taste and nausea link up across tens of minutes, from a single trial. Under the dominant learning theory of the time that should not have happened: that theory held any cue pairs with any consequence about equally, and what mattered was how many repetitions and how long the gap.

You have this too. Throw up once after eating something spoiled and you can avoid that taste for years; get badly startled while a song happens to be playing and you do not usually end up afraid of the song.

Nobody designed this for the rat. Read it this way instead: ancestors varied in which things linked easily to which, and those whose sense of taste linked fast to a bad gut died slightly less often from eating the wrong thing, so that configuration slowly grew as a share of later generations.

Imprinting is the same thing wearing a different face. In its first hours out of the shell a greylag gosling fixes on the first roughly suitable moving object it sees and follows it — usually its mother, sometimes, in experiments, a person. That is not learning especially fast. It is being narrowly sensitive, to a narrow class of input, inside a narrow stretch of time.

Put these two together and the instinct-or-learned split collapses completely. They are not two drawers. What gets passed on is never a particular action but a whole set of biases about when, toward what, and how easily learning happens. Behavior looks flexible; the fussiness underneath it is hard.