On humid nights in the Middleton Theater lobby, against the sounds of the ice machine, an old video game in the corner, and the muffled voices of whoever was starring in that month’s movie, the manager would tell Alex stories about his life. He said that decades before, an ex-girlfriend tried to have him killed. Two black Cadillacs with no license plates followed him in his Corvette to his house one night, then circled the block several times before leaving. A few days after that, newspapers reported a man shot in a Corvette in a mall parking lot in Madison. “They got the wrong man and never bothered me again.”
That’s what he said, anyway. As the movies changed, the manager told a lot of stories like this, and in each one his safety was in peril. In one story he was holding off a roomful of electricians with a hammer because he hadn’t joined their union. In another, he faced an angry sergeant after he screwed up an infantry drill. One day he reported how the pastor of his Methodist church abruptly told him never to show up again. The manager never told Alex why.
These stories were 100 percent unique to the manager, transmitted to Alex and perhaps a few other employees. We wouldn’t expect another person to have the same stories. Unlike the manager’s stories, traditional folktales often experience a great deal of transmission, with little change, over long periods of time. As we will see in chapter 4, this was the case with Little Red Riding Hood, which was told and retold by parents to children over several thousand years. There was some change in the tale, but it was so slow that the variations—a tiger rather than a wolf in the Asian version, for example—can still be identified as versions of the same tale. Similarly, the fairy tale Snow White has numerous variations but is still recognizable. For instance, in the Irish version—known as Lasair Gheug—it’s a little trout as opposed to a mirror that tells the Evil Queen stepmother, much to her anger, that she is not the most beautiful woman ever in Ireland.
The key to whether something counts as a traditional tale versus a story told in a theater lobby resides in transmission, which in the case of folktales must be highly accurate, otherwise errors and/or embellishments would quickly render the story unrecognizable. There are remarkable illustrations of this kind of fidelity among storytellers across the world. In Rajasthan, the region of northwest India where semifeudal rule lasted into the mid-twentieth century, a caste of bhopa storytellers has told the same epic poems for centuries. Told and retold by these Rajasthani bards, the stories are incredibly long and long lived. For example, the massive eighth-century BC Mahabharata epic, which details the Kurukshetra War, is a thousand stanzas long and over six times the length of the Bible. Another epic told by a bhopa, when written down by an observer, ran over six hundred pages of text. The story, about a pastoralist who elopes with a personified goddess, and sets off a caste war that kills him and twenty-two of his brothers, later avenged by his son, probably mythologizes an intercaste blood feud from the distant past. It might take a month of eight-hour sessions to recount one of these epics, but they are so accurately retold that the Scottish historian William Dalrymple found that the version he heard late in the twentieth century differed only by a couple turns of phrase from a version recorded three decades earlier by a Cambridge scholar.
Such long-term accuracy of transmission requires long, often-hereditary apprenticeship. As fathers taught their sons to be bhopas, they had them learn ten lines rote per day by age four. This is what cultural evolutionists Rob Boyd and Pete Richerson refer to as guided variation—cultural recipes passed down through lines of descent, which effectively act as cultural agents across the generations. These recipes survive only if they are transmittable and capable of being replicated. Teaching is the cultural “bottleneck” of each generation, which is why language itself is—or has been, at least—shaped by its learnability. Although the innate capability of humans to learn a language is extraordinary and unique—a child learns several hundred words by the end of year two, or about what a specially educated chimp can learn in a lifetime—the iterated learning process shapes the language itself to become more learnable and compositional over generations.
A way to do this—relevant also to artificial intelligence, as we’ll see later—is to become compositional, or made up of interchangeable parts. Nicaraguan Sign Language (NSL) originated with the first special education programs in Nicaragua in the late 1970s and now is used by nearly a thousand deaf signers. Inventing the language themselves, the first users of NSL—adults who understood Spanish—expressed their signs holistically. To represent a cat that swallowed a bowling ball walking down a street—in a story that study participants were asked to narrate—signers would wiggle their hands across the body from left to right. Subsequent generations of NSL users, however, expressed their signs sequentially. The same cat would be expressed in two separate signs: first a circling motion to indicate the wobbling, followed by a flattened hand sweeping from left to right to indicate it was walking. By these later generations, NSL had become compositional, made up of parts that could be interchanged like words. Cultural transmission had shaped the sign language.
Transmission Experiments
To explore this evolution through iterated learning, culture evolutionists have experimented with games that are like the children’s game Chinese whispers, also known as Telephone. In a pioneering experiment, language researchers at the University of Edinburgh used what are known as transmission chains to show how learnability and structure emerges through transmission itself. First, they asked people to watch moving shapes on a computer screen and read their randomly assigned “alien” labels—like “kihemiwi” or “tuge,” for example. The participants were later quizzed on a set of moving shapes for the correct labels. They had seen half the shapes but not the remainder. Their answers were then shown to the next participant, who did the same and passed it on to the next participant. In each round, the image-label pairs were divided randomly into two groups, and the participants trained only on one, with the other half left unseen. In other words, all the labels were undergoing cultural transmission, just never all through one person. In a typical round, a participant would copy the image-label pairs, but with errors and changes. After just several rounds, and without any intentional design, the alien language evolved to become more structured and thus easier to learn.
It’s easy to do your own variations of these transmission chains. Record a passage of speech into your smartphone, for instance, then let another person listen to it. Then wait a minute and ask the person to record the same passage. Then play this new version for someone else, ask that person to record, and so on. The quick evolution of the passage always takes a unique path, but inevitably the passage becomes shorter and easier to learn.
Here’s another example. In a classroom, we hand one player a piece of paper with the following short description, which references the work of Connecticut College’s Joseph Schroeder and his students: “Recently, experimenters found by studying forty-seven captive rats that when they were fed Oreos on a daily basis, they became addicted to the sugar content. ‘The sugar hit them like cocaine,’ said one researcher, who concluded that sugar is a highly addictive substance.” We then give each player a blank piece of paper. With about a dozen participants seated in a semicircle, the first one is asked to read the passage, put the paper down on the floor, and then try to rewrite the passage on a sheet of paper without looking at the master copy. The player then passes the rendition to the next player, who does the same thing. After they finish, we observe how the message changed through variation, transmission, and sorting. In one case, the third participant reduced it to “scientists did experiments feeding Oreos to forty-seven rats, and they became addicted to the sugar.” As we might expect, the message always becomes shorter within the first few students in the chain.
Cultural Attractors
In transforming through iterated learning, a message also retains certain elements and loses others. The number forty-seven in our classroom exercise, for instance, was almost always retained to the end, as were details such as cocaine, Oreos, and sugar. It is thought that minimally counterintuitive elements like these frequently act as cultural attractors—elements that get preferentially retained over generations, whereas other elements get left behind.
If human minds are predisposed to certain types of information—Harvard psychologist Steven Pinker argues that the structure of language reveals how human minds were designed by natural selection—some predispositions will be innate. Almost all languages, for example, have words for the colors black, white, and red, and most have words for green and blue. Thinking of the natural world, it’s pretty clear that having words to describe plants, blood, sky, and sea should have some survival value. When humans and other primates see red, it elicits a hormonal response related to aggression—think of blood, a baboon’s rear end, or blushing skin. In Olympic combat sports such as boxing and judo, the competitor in the red shorts actually has been shown to have a statistical advantage of winning, all else being equal (even controlling for the colors worn by certain countries in the Olympics and so on). Even outside sports, men are perceived as more aggressive when they wear red than when they wear blue or gray.
Researchers have come up with other categories of cultural attractors. There may be emotional bias, as emotional arousal helps people remember experiences. Disgust is also common; the most shared urban legends and news headlines are often sensationally disgusting. But a major category is survival bias, the idea being that humans retain important lessons about the environment, potential threats, and reproductive strategies. Over time, survival information would accumulate in folktales. Little Red Riding Hood and Snow White, for example, contain injunctions such as beware the forest, the unknown, and greed.
Survival bias would obviously be strong in the design of prehistoric technology, which was transmitted across generations as cultural recipes. The recipe for building a canoe is like a story, and one that enhances survival. Ancient teaching of how to build a canoe—from the shape of the bow to the durability of the fiber cords, narrowness of the keel, and quality of the single large tree trunk that was dug out—would obviously have affected success in fishing, warfare, and colonizing other islands. The canoe is also like a story in that certain aspects of the craft could become more prevalent even if they do not affect survival but instead simply are aspects of knowledge that are preferentially passed on from parents to children. At the submerged fishing settlement site of Tybrind Vig, which lies off the coast of Denmark and dates to around 6500 BC, a preserved dugout canoe was found along with intricately shaped and designed wooden paddles, bone fishhooks, and preserved textiles. If the features are inconsequential to survival—say, the designs on the wooden paddles—we might refer to them as stylistic. They have the potential to change quickly compared to changes in functional features—such as the design of the paddle itself—which affect the mortality or reproductive rates of the people using them.
Social Information Bias
Is there a bias in how social information gets transmitted—not just a bias in what gets transmitted but also in how it gets transmitted? We would guess so from the social brain hypothesis as well as the human preference for stories with gossip and rumor about family, marriage, sex, friendship, betrayal, social status, interpersonal conflict, and deception. Sometimes we get information from people whom we view as role models. This is often referred to as prestige bias—learning from people to whom deference is shown on the basis of their intellect, success, or some other quality or achievement. In traditional societies, these characteristics are frequently possessed by one person, but in a modern world of celebrity endorsements, for example—“I’m not really a doctor, but I play one on TV”—they often aren’t, even though social media amplify prestige bias by extending it well beyond the local community, and allowing people to follow international celebrities and public figures through the same action as friends.
Social identity is an obvious attractor. Psychologists have shown that the more polemical the topic—think climate change or gun control—the more likely people on both sides of an issue are to ignore the evidence and go with their preexisting view—their “inner attractor,” shall we say. On more neutral subjects, however, people are good at letting the evidence guide their decisions. These polemics remind us that perhaps the most important social information is learning how to learn from others, which means learning about cooperation. The benefits of cooperating are obvious at the group level, but they can even be seen for the individual as well, including less violence and shared food, and importantly, also the potential to accumulate cultural knowledge.
How did this cooperation get started in the first place? One theory is kin selection, which refers to preferentially helping those people to whom you are most closely related genetically, but this does not explain why people began to cooperate among societies beyond simply their kin groups. Organized religion is a great illustration, where cooperation is predicated not on kinship but rather on oral and written narratives. We may look for both cultural attractors and survival value in comparing religions. Some researchers maintain that small groups that believe in omniscient, moralistic gods who punish are better able to aggregate into larger societies. It has been shown that more complex societies, and hence religions, including Christianity, Islam, and Hinduism, tend to have more punishing and moralistic gods, but this is a correlation without a causal explanation. We’ll discuss this in more detail in chapter 8.
Leaving aside cause and effect for the time being, the data show a correlation between religious thought, on the one hand, and generosity and honesty, on the other. In a large cross-cultural study of eight small-scale societies, which included foragers, pastoralists, and horticulturalists, people who believed in a moralistic, knowledgeable, and punishing god shared more readily with each other, including with unrelated strangers who shared the same beliefs. Christians and Muslims exhibited more fairness in economic games than those who held local/traditional religious beliefs, although there were no tests of what effect there was on whom they thought they were playing with or giving to. The more moralistic, knowledgeable, and punishing the shared authority figure was, the more money people typically gave to one another. Those groups whose gods had these qualities were in a better position to exchange between unrelated kin groups, and therefore aggregate and expand at the expense of smaller groups that believed in gods without those three qualities.
In the experiment, the participants were asked to give coins to others according to rules they themselves were able to define. Researchers found that generosity toward strangers was significantly higher—by a factor of five in one test—when both sides believed in a moralistic god that punishes bad behavior and knows your thoughts. They checked this by imbuing local gods with greater knowledge (omniscience of one’s thoughts) and/or propensity to punish, but this did not make people more generous. In other words, it had to be a moralistic god that both threatens punishment and knows everything.
What emerged first, generosity or moralistic gods? It’s hard to say; meat exchange is as old as our species, but the oldest gift giving or religion depends on how the evidence is interpreted. Was it religion when cave art and lion-headed figurines were made forty thousand years ago, or engraved lines on a mammalian rib fragment a quarter-million years ago? Also, people can be moralistic and punishing regardless of religion; the Nuer pastoralists, as Evans-Pritchard wrote, held “profound contempt” for people without cattle, exchanged cattle to maintain alliances, and rubbed the ashes along the back of a cow to communicate with the spirits of deceased ancestors.
In any case, knowledge and the ability to punish tend to be characteristics of effective leaders. Think of good schoolteachers: if their authority is undermined, or if the teacher is perceived to have the wrong values (like grading randomly), or is judged to lack knowledgeable, the teacher will be less effective. Before Bad Teacher with Cameron Diaz changed it, there was a general formula for “teacher movies,” such as Stand by Me with Edward James Olmos, where the teacher urges everyone to get along, knows not only the subject but also the students, has high moral standards, and can dish it out when appropriate. Looking ahead, we might ask whether artificial intelligence will ever take on these qualities. In many ways, it already has, as an algorithm can be programmed to be moralistic, knowledgeable, and punishing. Take your typical airline booking site, for example, which punishes you for making changes, knows every route and best connection (knowledgeable), and promises the best price for everyone (moralistic).
Slicing and Dicing in the Digital Age
What does all this research on culture transmission say about the future? How might bias, attractors, and morality interact in an era where cultural transmission is of a different magnitude? Let’s think in terms of advertising slogans. In fact, maybe there’s an idea for a business here, in producing memorable, easily transmittable advertising slogans. The terms “memorable” and “easily transmitted” are important because they lie at the root of what sells and what doesn’t. “Just Do It” and “Where’s the Beef?” didn’t get to be two of the greatest taglines ever by being forgettable. Considering that McDonald’s executives are talking seriously about replacing workers with robots, and combining this with the fact that animal muscle tissue can now be grown in a petri dish from stem cells, we might need to get a job to improve one of their slogans, along the lines of “the all-new, robot-fried, stem-cell-grown McDonald’s cheeseburger in a petri bun.” If they’d hire us, we’d send this message through transmission chains of eight customers to shorten the message and highlight its most attractive aspects through what was retained.
Or maybe we’d simply get our group of participants to start cut and pasting messages electronically. A group of researchers at Facebook considered many cut-and-pasted sentences or passages that appeared on the social media site over a period of years, including this one: “No one should die because they cannot afford health care, and no one should go broke because they get sick. If you agree, post this as your status for the rest of the day.” This phrase, exactly, was copied half a million times over a couple years. If this were one of our Chinese whispers–like experiments, we would expect the passage to become shorter. Here is the difference with copying text, though: the next most popular version, which was copied sixty thousand times, was longer because people inserted “thinks that” after the sharer’s name, and the third most popular version had “we are only as strong as the weakest among us” inserted in the middle. In this social media scenario, cultural evolution reversed its direction by adding words rather than by subtracting them. The opposite can happen as well—the length might be limited to 140 characters or fewer, for example—as social media platforms profoundly affect cultural transmission.
The Facebook team mapped changes in this text between Facebook friends. Not surprisingly, they found that a text’s chances of being replicated were roughly doubled when it included a phrase such as “please post this” or “copy and paste,” and otherwise boosted by phrases such as “see how many people.” But here is another major difference from storytelling: the mutation rate in the text was 11 percent, which means that one in nine users changed it. Now this might be comparable to mutation rates in our Chinese whispers game at the start, but certainly not among our Rajasthani storytellers. Also, that figure of 11 percent was arrived at because the Facebook team identified over a hundred thousand different variants of the text among over a million status updates. So while we might talk about a couple of key changes in Little Red Riding Hood as it spread into China and Britain over many centuries, adapting to different cultures and environments (such as tiger versus wolf), on Facebook the versions changed, splintered, and conformed to different social groups, like those that turned it into a joke—“no one should be without a beer because they cannot afford one”—and others that transformed into an opposing political view—“no one should die because the government is involved with health care.”
One pertinent question comes to mind about text sharing. In our transmission chain experiments, as the participants retrace what happened to the story along the chain, they might share a laugh at a big change because they know how it “should” be corrected. Real cultural transmission is always a group process, with the self-correction feature for group consensus, where learning takes place over a childhood or lifetime. As one bhopa told Dalrymple, “My father used to teach me one story a day, then he would correct me as I recited it back.”
How much we can invest in this knowledge transmission depends on the context, though. In the West, where elite college tuition costs the same as a house, children are a long-term investment. This is different from simple agrarian societies, in which children are a net asset in terms of farm labor, and educating them is faster and more easily done by the parents themselves. As cultural complexity grows, the longer parents need to spend educating the next generation. However much you build up in your lifetime needs to be squeezed through the bottleneck of teaching the next generation. We take a closer look at this in chapter 4.
