You might recall from chapter 4 the pile of refrigerator parts that lay stacked behind the Middleton Theater’s concession stand for months, until one day, Jerry, the regional supervisor, showed up and told Alex and the manager to get rid of them. The manager loaded them up and took them home, leaving Alex to wonder why they were purchased in the first place if they were just going to lie around a theater lobby. Wasn’t the goal to actually fix a broken refrigerator? Perhaps the manager inherited both his procrastination and impulsivity—buying expensive parts and piling them up for months certainly counts in our book as “impulsive”—from his parents. About half the variability in behavioral traits such as procrastination, conscientiousness, or impulsivity is genetically inherited. We doubt, however, that the knowledge of how to fix a refrigerator is genetic. Maybe the manager’s dad taught him how to fix refrigerators, or maybe the manager was trying to find the time to read a repair manual and figure things out on his own—a form of “individual learning” but with help. Alex never asked him.
On a deserted island, though, even with some prior training, the manager would have had to build a refrigerator from scratch. From Homer’s Odyssey, to William Golding’s Lord of the Flies, to the 1960s’ sitcom Gilligan’s Island, Western culture loves to put people on deserted islands and see what happens. Without fail, technology plays a big role in how the inhabitants fare. It has to. Unlike on the original 1960s’ television show, shipwrecked passengers on a twenty-first-century Gilligan’s Island would have swam ashore with more than just accumulated knowledge and a couple-dozen trunks filled with money, stock certificates, and designer clothes. Our modern castaways would have grabbed their smartphones, and it’s a good bet that at least some of the devices would have survived the trip ashore. No better way to learn how to build a refrigerator than to look it up on your smartphone—at least until you run out of power.
Think about something else, however: because about a quarter-million active patents relate to smartphones, the shipwrecked group would have brought ashore with them the work of millions of highly technical people, if we assume several people behind each patent. Back on the US mainland, about forty thousand software patents are issued every year. Sorting out all the possible patent infringements, two law professors have calculated, could provide continuous full-time work for two million patent attorneys. This doesn’t even begin to account for the millions of people who have populated the Internet with their ideas, views, wisdom, and whatever else is posted there, including hundreds of sites detailing how to build a refrigerator.
The point is, ideas require people to create them—and manage them. Will the island group produce any ideas of its own? On Gilligan’s Island, the Professor invented everything the group needed—except a reliable boat, of course. He made a lie detector out of coconut shells. He used tree sap to seal raincoats together into a hot-air balloon. He made nitroglycerine from rocks and papaya seeds. But had the Professor died, most of his genius would have gone with him. Future generations of Gilligan’s islanders would have plunged deeper and deeper into technological darkness. Fortunately, as documented in a 1978 television movie, they were all rescued before this happened.
Gilligan’s Island is a great introduction to a prominent theory put forward by Joe Henrich about how cultural knowledge accumulates over time. Using the island of Tasmania as a case study, Henrich started with the assumption that each person learns from an expert in the population. For a young child, those experts are one’s parents, and then later a learner may focus attention on other people in the community who appear to be more successful and/or knowledgeable than the parents about certain tasks. Under this “Tasmania hypothesis,” just how skilled or knowledgeable each learner becomes will depend on certain variables and probabilities: how easily an expert can be identified in the first place, how exceptional the expert is in teaching a skill, how accurate the learning process is, and how exceptional the learner is in understanding the knowledge being transmitted.
Under reasonable assumptions, a typical learner usually will not become as good as the expert is, although occasionally a student by chance will surpass the teacher in terms of skill. After Gilligan learned from the Professor how to make use of tree sap, for example, he concocted glue with exceptional adhesive properties that was crucial for building the island telephone. The idea is that with larger populations, there is a better chance that at least one of the many Gilligans will surpass the Professor in terms of mastering some task. At some critical population size, that probability is high enough to expect that in every generation, at least one Gilligan will surpass the Professor, raising the level in every generation. This helps explain why IQ scores have been rising over the last century.
The flip side of the Tasmania hypothesis is that cumulative knowledge will be lost when a population goes through a bottleneck. Prehistoric Tasmania, for instance, fell backward technologically when its population abruptly declined. Technology, though, is not the only thing subject to population bottlenecks. Among Polynesian languages across the Pacific, smaller island populations have experienced higher rates of word loss over the centuries, whereas larger populations have seen higher rates of word gain.
The idea that more people means more ideas leads some to take an optimistic view of world population growth, which may reach ten to eleven billion by 2100. Think about it: more people, more ideas. Ideas in the “ether” are a product of our collective brains, linked together through social and technological communication networks. One case study showed that seventy-five years of population growth in a Kenyan town brought about the ingenuity to farm again on what had been barren hillsides, thereby increasing family incomes. As the world’s population becomes urban—from an eighth in 1900, to half in 2008, to two-thirds by 2050—there is even more optimism, at least in some quarters.
This optimism is built on two trends. First, population growth is slower in cities, as parents increasingly invest in the education of their children rather than simply having more kids, as has long been a priority in agricultural societies (more kids, more workers). In chapter 6, we mentioned how the total fertility rate has declined globally, but it has declined even further—to below 2 percent—in urban centers. Second, even as population growth has slowed in dense urban settings, the production of ideas, innovation, and information has grown exponentially. Why? Because in large societies, innovation amounts to more than just one idea per person. Instead, it’s a dynamic process of idea exchange within networks of people. The pace of urban life grows superlinearly with population size, including the total number of social contacts and volume of communication activity. Both the gross domestic product and number of patented inventions in a city, for example, grow in proportion to its population size raised to an exponent between 1.2 and 1.3. In other words, if population grew from a hundred thousand to five hundred thousand, we’d see the other number jump from three million to twenty-five million.
These are among the “scaling laws” of cities; nevertheless, the communication of ideas still takes place on a human scale. As we know from the television shows Friends and Sex and the City, people living in urban areas hang out in peer groups. Cities effect little measurable change in embeddedness of social networks—that is, how likely an individual’s contacts are also connected with each other. What matters is the effective cultural population size, which is the number of people actually sharing information. The productivity of a dense population still boils down to its smaller groups and the fluidity of membership.
For good reason, many managers consider eight people with diverse skills as making an excellent team. A psychology experiment at the University of Arizona showed that a group of eight people, working together on a complex computer game, performed better than an individual or four-person group. A group of sixteen did no better—maybe even slightly worse—than the group of eight. Even per capita, performance was best in the eight-person teams. Gilligan’s Island had seven people with different strengths, including one all-around expert. The Professor oftentimes got help from Gilligan and the Skipper, and sometimes from Ginger, Mary Ann, and (rarely) the Howells.
In science, the goal is to increase the collective IQ, so to speak, which is why famous simultaneous discoveries are not random coincidences. For example, Charles Darwin and Alfred Wallace, during two weeks in 1858, apparently “discovered” evolution independently. Nowadays, individual geniuses are more difficult to identify because large research teams are the norm. Although it’s an outlier, a 2015 physics article set a record with over five thousand authors, and it’s not unusual to see articles with fifty or more contributors. As the number of researchers grows—currently at about 5 percent annually—the number of scientific research papers grows as well. The best estimate is about 4 percent per year since 1965, over which time tens of millions of peer-reviewed papers have been published. Those papers collectively cite about a billion references—a number also growing about 5 percent annually. Such growth rates may seem modest, but they are exponential, so the quantities are nearly doubling every decade and a half.
None of this explosion in verbiage is restricted to science, of course, as the sheer number of English words published in books has been growing exponentially for centuries, from millions of words in books in 1700 to trillions per year in the twenty-first century. And these books contain only a few percentage points of the verbiage recorded online. By 2007, humans had already stored two trillion gigabits, with the volume of digital data doubling about every three years. By 2016, more than ten times that much information was stored worldwide—that’s the number twenty followed by twenty-one zeros’ worth of information. That’s a lot.
All this is reminiscent of Alvin Toffler’s classic book from 1970, Future Shock, which predicted the consequences of “information overload.” Forty-five years later, historian Abby Smith Rumsey argued that vast amounts of digital information hinder our collective capacity for forgetting, which is an important behavioral trait that clears away informational clutter, making room for creative thought. As we saw in previous chapters, oral transmission of a story prunes away superfluous details, rendering it more learnable and relevant. In contrast, a viral video gets copied identically millions of times without being streamlined by the transmission process, and actually accumulates more junk in the form of comments and metadata.
Without the kind of vetting that has long typified cultural transmission, culture is bound to accumulate a lot of junk. As technologies crank up the volume, informational junk accumulates on our devices as well as in our texts and videos. A cleaning app can easily wipe a gigabyte of junk data off your smartphone every week—equivalent to a truckful of books or the ancient Library of Alexandria. All junk.
Junk, however, is still part of evolution. Much of human DNA, for instance, appears to be junk, meaning that it lacks any observable function. That doesn’t mean that it might serve a function; it just means we haven’t found one for it. On an evolutionary timescale, natural selection tends to clear away junk about as fast as it’s produced by random mutation. Similarly, in early cultural evolution, the size of the community was the constraint on how much junk information might tag along with the more useful and learnable information.
Once we started storing information outside ourselves, though, the constraints were lifted. In this century, there has been virtually no constraint because storage and processing have kept up with the explosion of information. In 2015, the computer giant Intel celebrated a half century of correctness of a 1965 prediction by its employee, Gordon Moore, that computing power per dollar cost would increase exponentially, doubling every two years. This subsequently became known as Moore’s law. Just to think how far this could go in theory, César Hidalgo of the MIT Media Lab calculated the ultimate information capacity of Earth—human, biological, and technological—as being about 1056 bits (8 bits equal 1 byte). Again, that’s a lot. In 2017, the largest computer in the world—in China—held about 1015 bits, which, compared to the “planetary hard drive,” is like the diameter of a single electron compared to that of a large galaxy. Hidalgo calculated the planetary hard drive has used up 1044 bits out of its total capacity, which is like having traveled half a centimeter on a drive from Boston to Seattle.
Storage has increased in science too, not just in terms of computer data but also in terms of space in peer-reviewed journals. After the online, open-access (meaning that its content is free) journal PLOS ONE began publishing in 2006, it doubled in size every year for six years in a row in terms of the number of papers that appeared. As more reviewers have joined the effort, PLOS ONE can maintain a rigorous review process even as it publishes about a hundred papers per day. This high-volume model has spread in reputable ways—some prestigious print journals such as Science and Nature have online journals as well—but it has also given rise to hundreds of predatory open-access journals that will publish, for a hefty fee, almost any academic study on almost any subject.
Like the headwaters of a river branching into the uplands, new niches of science spring up in increasingly specialist soil. As a headline in the satirical newspaper the Onion read, “Scientists Make Discovery about World’s Silt Deposits but Understand If You Aren’t Interested in That.” In 2016, there were twenty peer-reviewed papers published on the niche subject of the neutron activation analysis of archaeological pottery. If you are a specialist with an interest in that topic, that’s manageable, but that manageability does not extend to the eight hundred or so papers published that year on neutron activation analysis generally, much less the several thousand papers on archaeology. If you worked on something more mainstream, such as obesity—as we do—you have over twenty thousand papers to deal with from 2016 alone. This is beyond human capacity to absorb, which means science may be reinventing the wheel more often: more stuff, but less change.
Let’s take a closer look at the diminishing returns effect between the bulk volume of what gets published and amount of new information created. There is a general law of diminishing returns in how the size of vocabulary—the number of different words—increases only with the square root of the raw number of words published: if you increase the number of books by a hundredfold, you increase the vocabulary contained in those books only tenfold. This effect extends to citations one makes to the work of others. Nowadays, most citations are to works that are only a few years old. As the volume of scientific papers grows exponentially, this window of several years represents a smaller and smaller fraction of the total scientific literature being produced. If the volume of papers doubles every decade—a 7 percent annual growth rate—a bibliography today covers only half as much of the expanding field as the same-size bibliography a decade before.
But here’s the interesting thing: because of the rapidly expanding number of articles and books being produced, the odds of having your work cited increases, especially if it’s only a few years old. It used to be, as Derek de la Sola first noticed in 1965, that most scientific papers were never cited—science’s coldhearted way of shedding junk. In 1980, about 30 percent of all published research went uncited, but by 2015, only about 10 percent of papers went uncited—a fraction that is still declining. Inflation of one’s citation numbers is such that hundreds, if not thousands, of living scientists now have a higher h-index than Darwin does (the h-index measures how many publications a researcher has with at least that number of citations—for example, an h-index of sixteen means that sixteen different papers have at least sixteen citations). And of course, the best way to generate higher citation counts is to cite your own work as often as possible, even if it has nothing to do with what you’re currently writing about.
Borrowing from Lewis Carroll’s Through the Looking Glass, this is the Red Queen effect, meaning that researchers need to do more and more just to stay in one place. Increasing competition to get into the top journals is ferocious, and it’s not surprising that there have been some high-profile retractions in the past decade for papers that faked data. One, on the attractiveness of Jamaican dancing, made the cover of Nature in 2005. After the senior author later found out that the lead author had faked the data, it took him a long struggle to have the paper retracted, which did not happen until December 2013 with a published two-sentence retraction letter in Nature. It seems the publisher was more interested in heavy citation counts than in whether the citations were to admittedly bogus science. In 2015, a Science paper faked data on how door-to-door canvassers can influence people’s opinions about personal stereotypes. Interestingly, the hypothesis was validated by a genuine study by different scientists the following year.
Dutch psychology professor Diederik Stapel, who was suspended from Tilburg University in 2011, started out by manipulating experimental psychology results and in later years simply faked his data to achieve the kind of “results” that prominent journals love. As a senior scientist, Stapel was at the center of his network of junior coauthors, who were apparently unaware of the fraudulent data. Citations of Stapel’s work have naturally plummeted, but his closest coauthors seem to have gone down with him. Guilty or not, reputation matters in science.
With all the junk, inflation, and fraud, can we still find and build on the best ideas, like on Gilligan’s Island? Without selection, evolutionary drift fills the void: a lot of activity, signifying little. Random drift is variation and transmission, but with little or no selection. The rash of fake news online may come to mind, whereas another well-studied example is first names, formerly traditional as we described in chapter 1, but now subject to random copying and thus evolutionary drift. With random drift in an expanding volume, even though more new ideas are generated, change in the most common ideas can paradoxically grind to a halt. This is because invented ideas, which start from obscurity, are less likely in an exponentially growing corpus to drift up into the popular conversation. Expanding populations have more ideas, but those ideas then have more trouble making it into common vocabulary because the most popular ideas are surfing the wave to ever-increasing popularity.
It’s the Red Queen effect on steroids. Paradoxically, exponential growth can make it look like the winners under drift were selected. Think of a “top two hundred” list of the most popular ideas or topics in some genre. It could be Twitter celebrities, trending scientific topics, or just the most popular English words. There is turnover on that list, with new entries replacing others on the top two hundred. Under random drift, there is usually continual turnover on this list, as new entries become highly popular just by the luck of random drift. When the corpus is growing exponentially, however, the rich-get-richer effect at the top keeps raising the bar. In 1970, the top 1 percent of five-year-old research papers had only fifty citations each, but by 2005, the top 1 percent of five-year-old papers had over a hundred citations. As the top 1 percent starts to age, turnover at the top slows down. The same thing is found among English words. In books from the early 1800s, there was a turnover of about seven or eight words per year among the top two hundred words. By 1900, the turnover was only two or three words, and by 2000, the turnover was only one word per year among the top two hundred. More volume, less variety, everywhere you look.
These exponential trends won’t continue forever, however. Although the ultimate planetary hard drive may be huge, Moore’s law for computers has finally begun to level off, at least for the time being, as transistor size would need to get smaller than ten nanometers for the doubling law to continue. Moore’s law is finally entering into a third phase of the classic S curve cycle of invention, innovation, and leveling off. As capacity levels off, will stronger selection return, surfacing all the good ideas again and shedding junk? Will more people mean more good ideas again? Maybe, but given the vastness of the informational universe, humans will need help. That help—the new S curve—is artificial intelligence. Selection will be possible again when machines are the ones vetting the information, but it won’t be the same, as we’ll see in the next chapter.