• • •
ARTHUR C. Clarke was my favorite writer when I was a teenager. His visionary approach to the world around us, and his imaginary futures, were constructed in great detail, and they often pointed to the most positive aspects of humanity. While he is most famous for 2001: A Space Odyssey, he wrote dozens of books and many more essays and short stories.
In one of his other well-known books, Rendezvous with Rama, which was published in 1973, a large cylindrical starship flies through our solar system, and a team of astronauts is sent to unravel its secrets. That starship, Rama, was initially thought to be an asteroid; however, it is detected by a series of automated telescopes known as SPACEGUARD. These had been put into place after a meteor smashed into northern Italy in 2077, leading humanity to create a sort of early warning system for any objects that might cross Earth’s path and potentially threaten our well-being.
When, a number of years after the book’s publication, a project like this was actually proposed, its creators paid homage to Clarke and named it the Spaceguard Survey. It uses a variety of discovery methods, including automatic detection of objects in space. In addition to being a real-life incarnation of science fiction, it is also the vanguard of a whole new way of doing science: automatically. One portion of this program, known as Spacewatch, uses automated image processing to detect what are termed NEOs, or near earth objects. In 1992, it was responsible for the first automated discovery of a comet, which now goes by the unwieldy name of C/1992 J1.
Automated science is being done in fields from biology to astronomy to theoretical mathematics, all using computers to make new discoveries and identify new facts. We have projects that find new species by looking at the genetic sequences in water scooped out of the oceans, and ones that allow chemists to automatically discover ways to synthesize specific chemicals. There is even a Web site named TheoryMine to which anyone can go and get a novel mathematical theorem created by a sophisticated computer program that automatically generates a mathematical proof and have it named after you or a loved one. This is automated discovery combined with vanity plates.
But that’s not the only big thing going on right now: There is a movement known as citizen science. Everyday individuals are becoming involved in actual scientific discovery. How is this possible? The principle is rather elegant. While computers are good at lots of things, from adding numbers to counting words in a document, they are often very bad at many simple things: We are still way ahead of computers in labeling photographs or even reading fuzzy or blurred text. This computer limitation, in addition to providing a stumbling block to any robots whose route to global domination relies on caption creation, has created an entirely new field of computer science known as human computation: Simple tasks are given to lots of people to perform, often either for a small amount of money or because someone has cleverly hidden the task in a game. One of the most well-known examples of these are the distorted words we often have to read correctly11 in order to prove our humanity to a Web site. Rather than simply being an inconvenience, they are now being exploited to actually help digitize such works as the New York Times archives. By pairing a distorted known word with one that computers are unable to decipher, everyday users who can read these words are helping bring newspapers and books into digital formats.
Scientists are beginning to use this sort of human computation. These researchers are relying on citizen scientists to help them look through large amounts of data, most of which is too difficult for a computer (or even a single person) to comb through. One example is Galaxy Zoo, in which scientists gave participants pictures of galaxies to help classify them. The participants weren’t experts; they were interested individuals who participated in a minutes-long tutorial and were simply interested in space, or wanted to help scientific progress.
Several intrepid scientists turned a fiendishly difficult problem—how to predict what shapes proteins will fold into based on their chemical makeup—into a game. They found that the best players of a simple online game known as Foldit are actually better than our best computers.
Our pattern-detection abilities, and other quirks of how our brains work, still give us the lead in many tasks that are required for new knowledge. So, in truth, the facts that are changing are not changing simply without the involvement of the general population. We are part of the scientific process now. Each of us, not just scientists, inventors, or even explorers, are able to be a part of the process of creating knowledge.
We are living during an auspicious time with respect to knowledge: For the first time, not only has massive computational power allowed for much information related to scientific discovery to be digitized—the amount of scientific data available online for analysis is simply staggering—but discoveries are actually occurring automatically through computational discovery, and also in a widely distributed fashion, through citizen science.
These combined forces, in addition to changing how new scientific knowledge is generated, have enabled us to obtain massive amounts of data on the properties of scientific discoveries and how they are found. This has led to what I, along with one of my collaborators, Nicholas Christakis, have taken to calling eurekometrics12. Eurekometrics is about studying scientific discoveries themselves. More traditional scientometric approaches that use citations are still very important. They can teach us how scientists collaborate, measure the impact of scientific research, and chart how scientific knowledge grows, but they often tell us nothing about the content of the discoveries themselves, or their properties. For example, rather than looking at the properties of the articles appearing in plant biology journals, we can instead look at the properties of the plant species that have been discovered.
One simple example of eurekometrics—and one that I was involved in—is examining how discoveries become more difficult over time.
• • •
IF you look back in history you can get the impression that scientific discoveries used to be easy. Galileo rolled objects down slopes; Robert Hooke played with a spring to learn about elasticity; Isaac Newton poked around his own eye with a darning needle to understand color perception. It took creativity and knowledge (and perhaps a lack of squeamishness or regard for one’s own well-being) to ask the right questions, but the experiments themselves could be very simple. Today, if you want to make a discovery in physics, it helps to be part of a ten-thousand-member team that runs a multibillion-dollar atom smasher. It takes ever more money, more effort, and more people to find out new things.
Until recently, no one actually tried to measure the increasing difficulty of discovery. It certainly seems to be getting harder, but how much harder? How fast does it change?
I approached this question in a eurekometric frame of mind and looked at three specific areas of science: mammal species, asteroids, and chemical elements. These areas have two primary benefits: In addition to being from different fields of science, many have clear discovery data going back hundreds of years. The first mammals discovered after the creation of the classification system developed by Carl Linnaeus date to the 1760s. The first asteroid discovered, Ceres, was found in 1801 (and was actually large enough to be thought a planet). And the first modern chemical element discovered (I ignored such elements as lead and gold, which have been known since ancient times) was phosphorus, in 1669.
As I began thinking about how to understand how discoveries get harder, I settled on size. I assumed that size is a good proxy for how easy it is to discover something: The smaller a creature or asteroid is, the harder it is to discover; in chemistry, the reverse is true, and the largest elements are the hardest to create and detect, so I used inverse size. Based on this, I plotted the average ease of discovery over time.
What I found, using this simple proxy for difficulty, was a clear pattern of how discoveries occur: Each dataset adhered to a curve with the same basic shape. In every case, the ease of discovery went down, and in ev
ery case, it was an exponential decay.
What this means is that the ease of discovery doesn’t drop by the same amount every year—it declines by the same fraction each year, a sort of reverse compound interest. For example, the sizes of asteroids discovered annually get 2.5 percent smaller each year. In the first few years, the ease of discovery drops off quickly; after early researchers pick the low-hanging fruit, it continues to “decay” for a long time, becoming slightly harder without ever quite becoming impossible.
There is even a similarity in one view of medicine. As Tyler Cowen, an economist at George Mason University, has noted, if you tally the number of major advances, or definitive moments, in modern medicine (as chronicled by James Le Fanu) in each decade of the middle of the twentieth century, you get an eventual decline: “In the 1940s there are six such moments,13 seven moments in the 1950s, six moments in the 1960s, a moment in 1970 and 1971 each, and from 1973 [to] 1998, a twenty-five-year period, there are only seven moments in total.”
But here’s the wonderful thing: The output of discovery keeps marching on in each of the areas I examined. We keep on discovering more asteroids, new mammals, and increasingly exotic chemical elements, even as each successive discovery becomes harder. These all occur at different rates—we find asteroids much faster than new types of mammal, for example—but we aren’t at the end of discovery in any of these areas. In fact, I only know of one area where scientific research has exhausted all discoveries: the “field” of the discovery of new major internal organs.
The trajectory of discovery in human anatomy began in prehistoric times with the discoveries of hearts and lungs, the organs that are rather hard to miss, especially after seeing your colleague disemboweled by a mastodon. This initial flowering of discovery was followed by that of more subtle organs, such as the pituitary gland. But in 1880, a Swedish medical student named Ivar Sandström14 discovered the parathyroid gland, and the final major internal organ in humans was discovered. That was it. The field’s age of discovery was over.
But science as a whole proceeds apace. We pour in more effort, more manpower, and greater ingenuity into further discovery and knowledge. A simple example is one of the first quantities to be studied in the field of scientometrics: the number of scientists over time. The first U.S. PhDs were granted by Yale University in 1861. Since that time the number of scientists in the United States and throughout the world has increased rapidly. For example, the membership of scientists in the Federation of American Societies for Experimental Biology increased from fewer than five hundred in the year 1920 to well over fifty thousand by the late 1990s. This hundredfold increase is extremely rapid in a period of less than eighty years and is indicative of the increase in scientific power through sheer numbers of scientists.
In fact, if you uttered the statement15 “Eighty percent of all the scientists who have ever lived are alive today” nearly anytime in the past three hundred years, you’d be right. This has allowed more research to be done by larger scientific teams. Not only that, but higher-impact research is done by teams with many more scientists. Of course, growth like this is not sustainable—a long exponential increase in the number of scientists means that at some point the number of scientists would need to exceed the number of humans on Earth. While this may almost be the case on Krypton, Superman’s home planet, where the whole population seems to consist entirely of scientists, I don’t see this happening on Earth anytime soon. But this rapid growth demonstrates that scientific discovery is by no means anywhere near finished.
For example, in pharmaceutical research, drug companies counter the decreasing number of drugs created per dollar spent by pouring more money into drug discovery. As science grows exponentially more difficult in some areas, affordable technology often proceeds along a similar curve: an exponential increase in computer processing power means that problems once considered hard, such as visualizing fractals, proving certain mathematical theorems, or simulating entire populations, can now be done quite easily. Some scientists arrive at new discoveries without a significant investment in resources by becoming more clever and innovative. When Stanley Milgram did his famous “six degrees of separation” experiment, the one that showed that everyone on Earth was much more closely linked than we imagined, he did it by using not much more than postcards and stamps.
When one area of research becomes difficult, the scientists in that field either rise to the challenge by investing greater effort or shift their focus of inquiry. Physicists have moved into biology, raising new questions that no one had thought to ask before. Mathematicians and computer scientists have turned their formulas and algorithms to the social sciences and unleashed basic new discoveries about the way societies operate. Or scientists figure out ways to make the hard questions much easier, whether by importing techniques from other areas or inventing new methods.
However it happens, scientific discovery marches forward. We are in an exceptional time, when the number of scientists is growing rapidly and consists of the majority of scientists who have ever lived. We have massive collaborative projects, from the Manhattan Project to particle accelerators, that have and are unearthing secrets of our cosmos. Yet, while this era of big science has allowed for the shockingly fast accumulation of knowledge, this growth of science is not unexpected. Both the growth of scientific facts itself as well as what allows discovery and innovation to plow ahead can be explained by scientometrics.
However, just as scientometrics can explain the growth of scientific knowledge, it can also explain how facts are overturned.
CHAPTER 3
The Asymptote of Truth
ANYONE who was alive during the Late Cretaceous, between about one hundred million and sixty-five million years ago, would have seen many familiar creatures: the fearsome tyrannosaurs; duck-billed hadrosaurs; numerous birds; and even a selection of small mammals. If you ventured into the sea, you would have seen a wide variety of marine creatures. Among these were animals known as coelacanths (pronounced “SEE-luh-canths”)—gray, thick-lipped fish, they were one of the more hideous creatures of this period. A large meteor struck the Earth at the end of the Cretaceous period around what is today Mexico’s Yucatán peninsula, numerous volcanoes erupted that deposited ash across the sky, and the planet was plunged into a cataclysmic climate shift that caused the extinction of all of these creatures. In addition to wiping out our beloved dinosaurs, this massive extinction included the coelacanth, the ugly stepsister to nearly everything else that lived in the ocean.
But in 1938, on the eve of World War II, this changed. While the cause of the massive extinction at the boundary between the Cretaceous and Tertiary periods wasn’t yet known, the numerous extinct species from that time had already been chronicled. From the Tyrannosaurus rex to the coelacanth, what was lost in the past had been well studied. At the time, Marjorie Courtenay-Latimer, a young woman in South Africa,1 was curator of a small museum in the town of East London, not far from Cape Town. She had befriended a fisherman in the area, and he would periodically show his catch to her, allowing her to add any possible finds to the museum’s collection.
One winter day at the end of 1938, when Courtenay-Latimer went to the dock to check out the fisherman’s haul, she noticed a strange fin poking out. When she excavated it from the rest of the pile she discovered what she described as “the most beautiful fish I had ever seen, five feet long, and a pale mauve blue with iridescent silver markings.” No doubt this was the scientist in her speaking, with the thrill of seeing something possibly new. Anyone else would have seen a terribly ugly, oily, and foul-smelling fish.
That was definitely the assessment of a taxi driver when Courtenay-Latimer attempted to catch a ride back to the museum with her large, stinking find. But her hunch that this fish was important was verified when she scoured her books back at the museum and identified her catch as the long-lost coelacanth. Somehow, Courtenay-Latimer was astonished to discover, this
fish had survived in the Indian Ocean unchanged by evolutionary pressures for tens of millions of years. This was confirmed by a professor at a nearby university, who upon seeing sketches of her find sent the telegram MOST IMPORTANT PRESERVE SKELETON AND GILLS = FISH DESCRIBED.
It took another decade and a half, and the offer of a large reward, for a second coelacanth specimen to be discovered. This one was found off the Comoros islands, between Madagascar and the African mainland. But the impossible had been done: A supposedly extinct species had been discovered alive and well.
The coelacanth is an example of what are known as Lazarus taxa:2 living things that are presumed long extinct until contrary evidence is discovered. Of course, predicting whether a single extinct species will one day be rediscovered living in some corner of the planet is nearly impossible. But if we look at large groups of species we sometimes can determine, in aggregate, how many species might actually not be extinct after all, and how often facts are incorrect and need to be overturned.
In 2010, two biologists at the University of Queensland in Australia3 tabulated all the mammals that have very likely gone extinct in the past five hundred years. This yielded a list of 187 species. Then they checked to see how many were eventually recategorized as nonextinct. The answer: More than a third of all mammals that allegedly were lost to time in the past five hundred years have since been rediscovered.
This sort of large-scale analysis is not just for understanding the nature of Lazarus taxa. It can be extended more generally to enable us to understand the entire edifice of science and how we overturn long-held scientific beliefs. By looking at how science changes overall, we can see the patterns in how scientific knowledge is revised. And it can lead us to measuring the half-life of facts.
The Half-Life of Facts Page 3
