Note that equilibrium is only an approximate condition -- there may be directions in gene space in which overall fitness can still increase (even substantially), but it takes time for the random mutational process of evolution to find them. In most directions one would expect to find either only a very small positive (or zero) fitness gradient or a negative gradient, assuming a population that has been genotypically stable for a long time. Recent studies suggest that humans may have experienced rapid evolution in the last 10-50 thousand years due to the advent of agriculture, population growth, etc.
At the end of the article, one of the biologists seems ready to rediscover the Cochran-Harpending hypothesis :-) See also here.
NYTimes: ... It takes just 15 generations under these conditions for the flies to become genetically programmed to learn better. At the beginning of the experiment, the flies take many hours to learn the difference between the normal and quinine-spiked jellies. The fast-learning strain of flies needs less than an hour.
But the flies pay a price for fast learning. Dr. Kawecki and his colleagues pitted smart fly larvae against a different strain of flies, mixing the insects and giving them a meager supply of yeast to see who would survive. The scientists then ran the same experiment, but with the ordinary relatives of the smart flies competing against the new strain. About half the smart flies survived; 80 percent of the ordinary flies did.
Reversing the experiment showed that being smart does not ensure survival. “We took some population of flies and kept them over 30 generations on really poor food so they adapted so they could develop better on it,” Dr. Kawecki said. “And then we asked what happened to the learning ability. It went down.”
The ability to learn does not just harm the flies in their youth, though. In a paper to be published in the journal Evolution, Dr. Kawecki and his colleagues report that their fast-learning flies live on average 15 percent shorter lives than flies that had not experienced selection on the quinine-spiked jelly. Flies that have undergone selection for long life were up to 40 percent worse at learning than ordinary flies.
... “Humans have gone to the extreme,” said Dr. Dukas, both in the ability of our species to learn and in the cost for that ability.
Humans’ oversize brains require 20 percent of all the calories burned at rest. A newborn’s brain is so big that it can create serious risks for mother and child at birth. Yet newborns know so little that they are entirely helpless. It takes many years for humans to learn enough to live on their own.
Dr. Kawecki says it is worth investigating whether humans also pay hidden costs for extreme learning. “We could speculate that some diseases are a byproduct of intelligence,” he said.
The benefits of learning must have been enormous for evolution to have overcome those costs, Dr. Kawecki argues. For many animals, learning mainly offers a benefit in finding food or a mate. But humans also live in complex societies where learning has benefits, as well.
“If you’re using your intelligence to outsmart your group, then there’s an arms race,” Dr. Kawecki said. “So there’s no absolute optimal level. You just have to be smarter than the others.”