Evolution works in mysterious ways

But not unreasonable or magical.

The catch of the week is "On the stimulus situation releasing the begging response in the newly-hatched herring gull chick (Larus Argentatus Argentatus Pont.)" by Tinberger and Perdeck. It was published in the journal Behavior in 1950, so it is not precisely recent, but it is seminal work regarding how evolution works. The authors studied how gull chicks peck at their parents' beaks 'to let them know' that they are hungry. At least that is what it might seem like to a human observer. What is happening in reality is that the chicks have an innate reflex to peck at the red patch on the adult's beak, but when they are young they are pretty bad at aiming so often they grab the regurgitated food in the adult's beak close to the patch instead of actually hitting the red patch. Even more fascinating, the parents typically don't initiate the feeding process; instead they have a reflex to regurgitate when the chicks start pecking at their beaks. The authors went on to investigate the process by fabricating mock heads and beaks with different shapes and colors and discovered that color contrast between the beak and patch is more important than the actual red color, and that the shape of the beak is also important. With this knowledge, they created 'supernormal' stimuli, elongated beaks with high contrast patches that the poor chicks went crazy after.

I see two relevant lessons in this study. First, evolution did not create a direct 'feed the young' connection; instead those chicks that peck more tend to get more food and have a higher probability of surviving and those parents that respond to the pecking have a higher probability of their genes surviving. The same phenomenon happens in other situations and can be bad. For example, when a CEO makes it to the top of the company that person (and others!) might assume that because she made it to the top, her behavior and decision-making process are inherently good without considering that they might have been appropriate only in a particular situation – or worse, they did not make a difference and are just tagging along. Machine Learning algorithms do this often as well, they might learn to classify pictures of wolves based on whether there is snow present in the picture or not as opposed to actual characteristics of the wolves. This of course does not affect only CEOs; any human who gets positive reinforcement without understanding why the behavior was optimal is at risk.

The second is that we can create supernormal stimuli, things that we really did not evolve to experience and can be dangerous. Obvious cases are calorie-rich food, certain classes of drugs such as cocaine and methamphetamine, pornography, etc. A less obvious but more often pernicious case is personalized advertising that a person gets on the web based on everything internet companies know about that person. Internet companies don’t exploit us on purpose, it just happens to be the behavior that allows their survival. The problem with supernormal stimuli is that we don’t have natural defense mechanisms against them, so we have to always be on the watch.

You can get the document at JSTOR for free, but you need to get an account.
Alternatively, I posted 3 pages with the discussion and summary of the paper here (fair use policy).