The Thing with Feathers (5 page)

Starling flocks certainly seem to arise from a vibrant, intricately choreographed essence of life itself, a force that defies understanding. How can a hundred thousand birds zip around
at 30 miles per hour, each mere inches from the next, and maintain a cohesive flock while constantly shifting direction? The more you think about it, the more it boggles the mind.

Scientists call this
collective behavior
: essentially, a bunch of individuals acting as a crowd. In this particular case, that behavior is self-organizing, which makes starlings interesting, because many things in this universe tend toward disorder; if a glass shatters on the floor, its pieces won’t spontaneously rearrange themselves. Logic and experience would seem to imply that groups of starlings must be influenced by some outside force to stay organized, a force that counteracts the second law of thermodynamics—that physical systems tend toward chaos.

But funny things can happen when individuals behave as a group. Other examples of self-organization include snowflake crystals, our economy, the Internet, and even the evolution of life as we know it. Hundreds of scientific papers, theses, and books have tried to incorporate the same concept into such varied topics as language development and traffic jams. Sometimes organization results from many individuals making small, separate decisions—a sort of “ground-up” mechanism, as opposed to the “top-down” approach generally associated with the imposition of order. Starling flocks might be more spontaneous than they seem.

Economist Jeffrey Goldstein took this thinking a step further in 1999 when he attempted to define
emergence
, a loose property by which complex systems form strictly through simple interactions. Emergence applies to termite mounds, hurricanes, art, rock concerts, financial markets, and religion. Although these systems represent vastly different realms, Goldstein believes they share enough similar characteristics to be considered together; he defined emergence as “the arising of novel and coherent structures, patterns and properties during
the process of self-organization in complex systems.” Starling flocks, which coalesce from the aether, fit this definition perfectly.

Emergence has become a hot catchphrase over the past decade (popular author Steven Johnson wrote a book about it,
Emergence: The Connected Lives of Ants, Brains, Cities, and Software
), but not everyone believes it is entirely useful. One of the idea’s most vocal critics is biologist Peter Corning, who published a paper in the journal
Complexity
that called emergence elusive, ambiguous, and “a venerable concept in search of a theory.” He pointed out that a game of chess could be considered an emergent system because the complex game results from a few simple rules. But rules don’t cause anything; they merely describe relationships. Though chess may seem to be spontaneously organized—a ray of logic in a chaotic universe—that’s only because it is affected by two players who are channeling their energies into it; knowing the rules won’t help you predict the outcome of the game.

Starlings must obey physical rules to avoid collisions within their flock, no question. But are the birds chess pieces, beholden to some greater force? Or does flock behavior arise from within? If we knew their rules, could we predict the mesmerizing antics of a hundred thousand birds swirling together through the Irish dusk?

—

IN 1970, BRITISH MATHEMATICIAN
John Conway devised a simple exercise that he dubbed the Game of Life. On a grid with no boundaries, some squares are filled in to start the game. Then, depending on circumstances, each square lives or dies from one generation to the next.

Conway set just two rules. If an occupied square has two or
three neighbors (of eight adjacent squares, including diagonals), it will stay occupied; otherwise, it dies. And if an unoccupied square has exactly three neighbors, it will spontaneously become occupied.

People quickly realized that Conway’s “game” was quite interesting. It mimics populations. It produces immensely complex patterns. It can form self-replicating structures that kill their parents. It can, theoretically, perform any algorithmic calculation possible on a modern computer within its single grid. In a two-dimensional, cellular sense, the Game of Life offers fascinating parallels to real life, and it is incredibly simple. Depending on the pattern you start with on the infinite checkerboard, vastly different results are possible; some populations die out after many generations, some skyrocket, but most eventually stabilize at some equilibrium. Tiny adjustments have huge consequences. Occupy five neighboring squares in a certain way, and they eventually increase to a population of 116 and stabilize after 1,103 generations. Move a couple of the initial squares over, though, and they stay forever static. Nudge them again, and they become a unified glider that shoots itself across the board into infinite space.

The Game of Life seems to produce spontaneous order; scatter a few random tiles on a checkerboard, and they emerge into beautifully intricate patterns just by obeying simple rules. Instead of descending into chaos, the Game of Life shows that individual cells can organize themselves into complex structures without an overall plan.

It helped that Conway’s exercise coincided with the rise of microcomputers that could process thousands of generations in a short time. As the power of computers increased, so did the possibilities of modeling ever more complicated groups—like flocks of starlings.

The first computer model of flocking birds was developed in 1986 by Craig Reynolds, a graphics expert in California. Reynolds had been a technical assistant on the 1981 film
Looker
and worked on Disney’s 1982 film
Tron
as a scene programmer. He was frustrated by the challenge of illustrating lifelike swarms of animals. After some tinkering, he discovered that flocks could be simulated using the same principles of Conway’s Game of Life: Instead of designating a leader for others to follow, Reynolds decided to set a couple simple ground rules, sit back, and let a flock form itself.

The result was Boids (pronounced the way someone with a stereotypical Brooklyn accent would say “birds”), a simple program with startlingly complex results. Reynolds created graphic images of individual Boid creatures, represented as tiny triangles, and forced them to: (1) avoid collisions at close range, (2) head in the same average direction as neighbors, and (3) avoid becoming separated from the group. These three rules alone—separation, alignment, and cohesion—produced a convincingly lifelike flock of Boids, which twisted around Reynolds’s computer screen almost exactly like a group of real starlings in midair.

Reynolds added an obstacle to his model, and watched with fascination as the Boids smoothly split around it and regrouped on the other side. He couldn’t believe how realistic the simulation was—it would be perfect for motion pictures. The first movie to showcase the new effect was Tim Burton’s 1992 film
Batman Returns
, which featured bats swarming and penguins marching through Gotham City. Boids and its successors would eventually win Reynolds a 1998 Academy of Motion Picture Arts and Sciences Scientific and Technical Award in recognition of his contributions to behavioral animation.

The implications of Boids reached far beyond motion pictures. Because the program used simple decision-making rules to produce results similar to those of the natural world, it was hailed as an advancement of artificial life and paved the way for subsequent attempts to create artificial intelligence. One of the most wonderful aspects of the model was its unpredictability; despite following straightforward rules, it was impossible to predict the trajectory of the flock of Boids more than a few seconds in advance without programming a prescribed path. The Boids behaved like chess pieces in a game with no players.

After centuries of careful study, we know that particles are generally confined by physical laws. We can predict, for instance, how a gas will behave in certain conditions. But the particles of a gas do not interact except in random collisions; if they were to move with behavioral effects on one another, things would get very complicated very quickly. The world’s most powerful computer cannot always predict the long-term path of three celestial bodies moving within one another’s gravitational fields, even if their initial velocities and directions are finely measured. Neither can we predict the weather more than a few days in advance—there are just too many interacting elements.

And yet Boids elegantly re-created that unpredictability in a simplified, digital environment. The program seemed to mirror real life, but real flocking data would be necessary for scientists to figure out which models best fit reality. Unfortunately, at the time, gathering data on a live starling flock couldn’t be done. Nobody knew how to accurately measure the positions, velocities, and trajectories of thousands of birds simultaneously swirling around the evening sky. A breakthrough was needed, and it would eventually come not from the world of biology but from the world of hard physics.

—

WHEN TENNIS PHENOM SERENA WILLIAMS
cracked a backhand to open the third set of her quarterfinal match against Jennifer Capriati in the 2004 U.S. Open, a line judge ruled the ball good. But the chair umpire, who has ultimate authority, thought it had landed out and overruled the call. Williams threw a tantrum and lost the match. Slow-motion replays later showed that she had been cheated out of the point, and that at least two other balls had been called wrongly against her in the same set. The tournament apologized and yanked the umpire, too late to stop a firestorm of indignant players and fans from voicing their protest. The masses called for change. Of nearly 30,000 respondents to an NBC poll, 82 percent supported the use of instant replays.

Video line calling would be a big step for a traditional sport like tennis, but the technology was already in use elsewhere. Replays were first used to officiate American football games in the 1980s. Cricket and rugby matches followed in 2001, and replays debuted in basketball in 2002. Other sports, including field hockey, baseball, and rodeo, soon embraced the technology. Tennis associations had been testing a system called Hawk-Eye, and within a year of Serena’s controversial loss, the instant replay was approved for professional tournaments.

Hawk-Eye uses stereoscopic triangulation, a technology that is, in a sense, as old as vision itself. The reason we have depth perception is because of our binocular sight; overlapping images from each eye, slightly offset from each other, are fused together by the brain into one three-dimensional picture. View-Masters and 3-D movies enhance this effect by presenting simultaneous images taken from slightly different angles. Our brain has no trouble aligning the separate perspectives, but
programming a computer to do the same thing is tricky. Hawk-Eye takes images from ten different cameras around a tennis court and fuses them into one model; it decides which pixels represent the tennis ball, flying at more than a hundred miles per hour, and predicts where the ball will land relative to a line on the court. Tests showed that the system is accurate to about one millimeter.

At about the same time that Hawk-Eye’s unblinking lens began to gaze down from the stands at major tennis tournaments, a group of Italian physicists and statisticians were peering into the sky over Rome, trying to make sense of the enormous flocks of starlings that appeared as if by magic each evening—the same flocks artistically photographed by Richard Barnes. They knew that bird flocks had been simulated with endless generations of computer models, most of them minor tweaks of Craig Reynolds’s Boids program from the 1980s, and that nobody had yet compared any of those models with empirical data from a real flock. The best existing set of observations came from a couple dozen fish swimming around a tank, a far cry from thousands of birds in free space.

The Italians, led by physicists Andrea Cavagna and Irene Giardina, recognized a challenge. In 2007, they set up three cameras on a terrace at the Palazzo Massimo, where, more than four centuries earlier, Pope Sixtus V had built a luxurious villa overlooking the imperial Baths of Diocletian, and where, these days, a large flock of starlings likes to roost at an adjacent train station. The cameras were placed fifty meters apart, pointed at the same patch of sky, and set to snap ten simultaneous high-resolution photos per second, using the same stereo photography methods as Hawk-Eye. The researchers started hanging out on the terrace in the evenings, making the museum’s security guards nervous, snapping photos whenever
a group of starlings swirled through the space in view of their cameras.

The difficulty of capturing starling flocks in a 3-D model boiled down to the “matching problem,” which had bottlenecked the entire study of flocks for many years. A computer must accurately match up specks of thousands of overlapping, individual birds between different images, from separate viewpoints and across time. The Italians spent two years developing an algorithm that could analyze photos of starling flocks and match the birds, reporting that it took a combination of “statistical physics, optimization theory, and computer vision techniques.” The breakthrough program could process flocks of up to 8,000 birds with nearly 90 percent accuracy, taking up to two hours to churn through a single eight-second clip. When they fed the program with images from their camera setup, the Italian researchers could visualize starling flocks as nobody had ever done before, with quantitative measurements—a real-life leap of knowledge analogous to Helen Hunt and Bill Paxton’s heroic (though fictional) tornado measurements in the movie
Twister
.

Starling flocks, it turns out, are thinner than you might expect—more like a floppy pancake than a football. The pancake slides around in various directions, shifting its appearance, but generally stays parallel to the ground and maintains a constant proportional shape, no matter the size of the flock. The density of the flock is higher toward its edges—starlings are more tightly packed at the pancake’s fringes than they are in its center. And starling flocks don’t have leaders. When a flock turns, birds fly on equal-radius paths; in other words, they each turn on the same curve at the same speed. In a column of marching soldiers, those on the outside of a turn must march faster to maintain their positions. Starlings don’t
compensate like soldiers do, so birds at the front of the flock end up on the right side after a left turn, those on the right side end up at the back, and those in the back end up on the left side. One potential benefit of this is that no bird must stay in the front position, which, as in a bicycle race, is aerodynamically least efficient and most tiring. Another benefit is that each bird spends the same amount of time on an edge, where there is more risk of being nabbed by a hawk. Because safety from predators is probably a major reason that starlings form flocks at all, any bird forced to spend all of its time on the edge would be less motivated to stay in the group, and the whole arrangement could fall apart.