I have recently began a series of posts to discuss my colleague Kevin Laland’s recent book on cultural evolution, Darwin’s Unfinished Symphony: How Culture Made the Human Mind. Last time we talked about just how different — quantitatively, for sure — human cognition is from anything else known in the animal world. This doesn’t make us the result of magic, of course, but it does mean that biologists, anthropologists, and cognitive scientists are faced with a rather unique challenge if they really want to understand what makes us humans.
Chapters 2 and 3 of the book, with which this post is concerned, deal with a widespread mechanism by which both humans and other species learn: copying other members of the same species. Chapter 2 aims at establishing that copying is, indeed, a very common learning strategy, while chapter 3 deals with the basic question of why, and under what circumstances, copying is an adaptive strategy (it isn’t always, so this is not a trivial question). There is far too much material for me to do a section-by-section commentary, so I will focus on a number of things that struck me as particularly interesting.
Rattus norvegicus, the brown rat, is not Norwegian, but of Chinese origins. It is also one of the animal species best adapted at living with humans, despite our systematic attempts to exterminate what we regard as a pest. Already Darwin had noted that the brown rat is so successful because it is very good at copying what works from other members of its species, as well as at avoiding behaviors that turn out to be lethal. Interestingly, rats have been able to adapt, both genetically, but — more importantly for our purposes here — behaviorally, even to new poisons that act slowly, designed to stretch the time between cause and effect, thus, in theory, precluding other rats from learning to stay away from poisoned foods.
Turns out, however, that the strategy adopted by these rodents is rather more canny: they don’t avoid dangerous foodstuff, they actively seek the sort of foods eaten by fellow rats who thrive. In other words, populations of Rattus norvegicus have developed “dietary traditions” in order to stay a step or two ahead of their human nemesis.
Rats are certainly not the only animals to have developed cultural traditions of sorts. Laland details several studies that, beginning back in the 1970s and ‘80s, have clearly shown the phenomenon to be common among apes and monkeys. Interestingly, we don’t really understand the adaptive value — if any — of some of these behaviors. For instance, at least three populations of orangutans are known to blow raspberries as they go to sleep. Nobody knows why. Or consider this: in the Lomas Barbudas reserve, in Costa Rica, pairs of monkeys simultaneously insert their fingers into each other’s nostrils, remaining in such odd position and swaying as if in trance, for several minutes. Go figure.
Learning from your social environment can be maladaptive, if one does not pay attention. For instance, blue tits and great tits birds often forage in mixed species groups. However, the blue tits eat twigs that are high on trees, while great tits feed mostly on the ground or on lower branches. Experimenters shifted things around so that young birds of one species would be reared by parents of the other. They observed that the animals adopted the foraging behavior of the rearing species, which sometimes was not exactly a great idea: some great tits attempted to forage hanging upside down from branches, in the manner typical of blue tits. And they kept falling off!
In order to copy, apparently, one does not have to be particular bright:
“The ubiquitous influence of social learning in nature is beautifully illustrated by the example of mate-choice copying, where an animal’s choice of partner is shaped by the mating decisions of other, same-sex individuals. This form of copying is extremely widespread, with examples known among insects, fishes, birds, and mammals, including humans. The fact that animals do not require a big brain to copy could not be more clearly demonstrated than by the tendency of tiny female fruit flies to select male flies that other females have chosen as mates.” (p. 41)
Another area where copying is fundamental is anti-predator behavior. Obviously, learning by trial and error in that case could easily be fatal, so animals tend to learn predator avoidance by watching what their conspecifics do. Interestingly, at least in some cases, fear of predators does not seem to be innate, but acquired, as demonstrated by the fact that rhesus monkeys reared in captivity are not afraid of snakes, while they wild counterparts certainly are. It’s also fascinating that experiments show that fear of predators is learned very quickly and lasts for a long time, while monkeys tend not to develop long lasting fears in response to stimuli that are not actually threatening. This, as Kevin puts it, is not only efficient in terms of predator avoidance, but also precludes the acquisition of potentially time wasting “superstitions,” i.e., fears of things that are not dangerous. If only humans were as good at avoiding superstition as monkeys…
So, copying as a learning strategy is widespread, and does not require large brains, only a cognitive system sophisticated enough to be capable of associative learning.
That said, for a long time biologists have struggled with why, exactly, copying is so widespread in nature. Even though the answer seems intuitive, mathematical models have repeatedly faced researchers with what is known as Rogers’ paradox, named after University of Utah anthropologist Alan Rogers: such models seem to show that copying is just as likely to lead to learning maladaptive, or outmoded, behaviors as adaptive ones. What gives? That is the topic of chapter 3 of the book.
Other animals aren’t the only ones engaging in copying, humans do it too. Experiments and observations in developmental psychology clearly show that children copy behaviors, especially from their care takers, from very early on. Their propensity to do so varies over time, peaks around age four, but never quite disappears. But they don’t copy indiscriminately, instinctively paying attention to behaviors that seem functional, and rapidly discarding others.
Obviously, members of a species cannot learn only by copying each other, or no innovation would ever be introduced, and any significant change in the environment would pose a threat of population extinction. Mathematical models, therefore, predict a mixed evolutionarily stable strategy, where learning by trial and error is in equilibrium with learning by copying. Except for the above mentioned problem posed by Rogers’ paradox. I will skip other fascinating bits of chapter 3 of Kevin’s book in order to focus on how he and his team managed to solve the paradox: by organizing a competition among different learning strategies, implemented by way of computer programs battling each other.
The idea isn’t new. It was successfully implemented, as Laland explains, back in the 1970s by economist Robert Axelrod to solve another famous biological puzzle: the evolution of cooperation. Axelrod organized a competition for the best program to solve the so-called iterative prisoner’s dilemma, a classical situation in game theory where agents have a strong incentive to cheat, even though cooperating would actually be the overall best strategy. Famously, the winning strategy in cooperation games is known as tit-for-tat, and was developed by Anatol Rapaport, a psychologist then at the University of Toronto. The strategy is as simple as it is efficient: when encountering a new agent, act cooperatively on the first round, then do what he does. If he cooperates, keep cooperating; if he cheats, beat the crap out of him. Cooperation, then, becomes a matter of so-called reciprocal altruism: I’m nice to you if you are nice to me, and so long as you will keep being nice to me.
Kevin goes into some detail on how his team — inspired by Axelrod’s success — organized a similar tournament to solve Rogers’ paradox and make progress on the issue of social vs asocial (i.e., solitary) learning. The result was indeed very insightful.
The tournament was structured so that players (i.e., computer program), in each round could implement a mixed strategy composed of three possible moves: INNOVATE (introduce a new behavior), OBSERVE (engage in social learning), or EXPLOIT (implement a previously learned behavior). Obviously, no pure strategy based on a single one of these moves would be adaptive, but which combination turned out to be the winner?
“The first finding that jumped out at us was that it is possible to learn too much! In the tournament, investing lots of time in learning was not at all effective. In fact, we found a strong negative correlation between the proportion of a strategy’s moves that were INNOVATE or OBSERVE, as opposed to EXPLOIT, and how well the strategy performed. Successful strategies spent only a small fraction of their time (5–10%) learning, and the bulk of their time caching in on what they had learned, through playing EXPLOIT.” (p. 66)
“Among the top-performing strategies that progressed to the melee, by and large, the more the strategy learned through OBSERVE rather than INNOVATE, the better it did. However, among the poorer performing strategies we actually witnessed the reverse relationship — the more they copied the worse they did. That told us something very interesting — copying was not universally beneficial. Copying only paid if it was done efficiently.” (p. 67)
Which was a very good clue toward the solution of the riddle posed by Rogers’ paradox. The winning strategy turned out to be one called DISCOUNTMACHINE by its authors, Dan Cownden and Tim Lillicrap, two graduate students from Queens University in Ontario. DISCOUNTMACHINE is so named because it discounts information according to how old it is: the older a behavior is, the more likely it is to be out of step with the always changing environment, and less likely therefore it is to be useful to the agent.
Also, both DISCOUNTMACHINE and all the other top strategies relied mostly on social learning. Indeed, when Kevin and colleagues introduced a “mutant” of the winning strategy that relied solely on innovation, its performance plummeted. Also, the researchers found that learning asocially is a viable strategy — perhaps not surprisingly — only under very extreme circumstances, when the environment changes drastically and rapidly, obviously because no previously developed behavior is then adaptive. Finally, and perhaps a bit more surprisingly, mostly copying was still the winning strategy even when the error rate was very high, up to 70%, further demonstrating the superiority of even badly executed social learning over the asocial option.
“Simple, poorly implemented, and inflexible social learning does not increment biological fitness, but smart, sophisticated, and flexible social learning does.” (p. 72)
One of the conclusions of the study was that, while copying obviously does not discover and introduce new behaviors, it preserves adaptive behaviors beyond the death of the individual that produced the original innovation. This is the first, widespread, step that makes cultural evolution possible.