Most of the time is spent testing for artifacts. That’s the pain. To make sure it’s not you or car exhaust from the street, or contamination from the air, or something it picked up in Antarctica.
—Stanford chemist Richard Zare, on experiments to detect organic compounds on the now-famous meteorite from Mars
In 1996 Stanford and NASA scientists announced that they had found evidence of ancient life on Mars. They based their conclusions on markings found on a rock they believe was kicked off Mars by a meteor that crashed into the red planet millions of years ago, flinging tons of rock and dust into space. This particular rock, they believe, orbited the Sun as a chunk of space debris for about 16 million years before landing on a blue ice field in the South Pole, where a sharp-eyed Antarctic researcher out for a joyride on her snowmobile spotted it, picked it up, and brought it home.
How do they know that the curious tube-shaped forms that landed in their lab were fossils of ancient life, rather than dried-up mud cracks, as one researcher put it? The simple answer is, they don’t—at least, not yet. But they’re working on it (along with dozens of other teams of researchers who would love nothing better than to prove them wrong).
The more complicated but truer answer is that facts rarely present themselves cleaned up and alone, ready to be admired and fussed over. Instead, nature bestows her blessings buried in mountains of garbage, and scientists rarely know what they have their hands on until they’ve sifted through the mess, laboriously, patiently, piece by piece.
The astronomers who found the first planets around stars other than our Sun faced similar problems. The announcements over the past few years about discoveries of new planets did not mean anyone had actually sighted a planet; they meant that astronomers had seen some unusual wiggles in the position of a star that indicated the star was being pulled off course by some unseen companion. When they saw the wiggle, they did not say: “Eureka! A planet!” Instead, they probably said something closer to: “Oops. Something must be wrong with our experiment.”
And so it goes with almost any important discovery you can name. A few years back, an astronomer announced that he had seen the face of God—or the first wrinkles in space-time traveling to us from the creation of the universe. Another group of researchers reported they had found the so-called top quark—the last known member of the family of fundamental particles to be pinned down.
The data these scientists were looking at hardly spelled out a clear signature: Most of the time, they were looking at strings of numbers (what else is there, you might ask?). From a bunch of digital signals, they read patterns like tea leaves, telling us that there’s more dark matter in the universe than we thought, or that a certain gene sits on a certain place on a chromosome, or that an asteroid is hurtling toward Earth. In fact, what they really “see” is an immense amount of static noise in which is buried a signal. Maybe.
Filtering out the static is a process central to both science and human perception (which are partners, after all, in much the same enterprise). You can’t perceive anything if you can’t block most of the information that comes your way. You can’t hear the voice on the phone if other people are shouting in the background; you couldn’t see anything at all if your iris didn’t shut out most of the light—allowing only a tiny bit to sneak in through the pupil.
Scientists, in turn, have to shield their experiments from extraneous influences. Slices of the Mars rock are examined in a vacuum so that no earthly critters creep in and muddy the results. Particle physicists bury their detectors underground and cover them with tons of shielding so that cosmic rays don’t make stray tracks that could be mistaken for new kinds of particles.
The astronomers may well have it worst of all. For one thing, noise drowns out most of their potential observing time. At night the star-spangled sky glitters like a black velvet showcase for multicolored diamonds—at least on a dear night far away from smog and city lights. During the day, however, there’s absolutely no sign of those glittering prizes.
Where do the stars spend their days? They’re up there, of course, decorating the sky as always, but you can’t see them in the glare of the light from the Sun. You can’t see the stars in the day for the same reason you can’t hear a whisper in a noisy restaurant—the insistent Sun shouts them out.
Even at night, there’s a lot of noise in the sky that can make it hard for astronomers to see the stars; there’s light from the Moon and the city; there’s heat from the telescope itself (many are refrigerated); there’s wind that stirs up the images and makes them as blurry as a penny on the bottom of a pool. A lot of the art in astronomy (and other sciences) is figuring out how to get around the noise without losing the signal.
To some extent, the noise problem is merely an artifact of choice and circumstance. Consider the pile of dust and dirt accumulating under the refrigerator. The crumbs were recently part of the cake you had for breakfast; the cat hair was part of the animal’s fur; the leaf belonged to a tree; and the paper clip found its way to the floor from something you opened in the mail. You don’t consider any of these things as candidates for the trash heap until they wind up in the “wrong” place.
There is an old riddle that vividly demonstrates just how noise can interfere with thinking, even when that noise is information. Imagine you are a bus driver. At the first stop of the day, nine passengers get on your bus. At the second stop, two people get off. At the third stop, four people get off, but three new people get on. What color are the bus driver’s eyes?*
Noise, in other words, is whatever you don’t want to be where it is—whether it’s the conversation in the background or the weeds in the garden. It’s what you need to get rid of to see what we want to see, to learn what we need to learn.
These things we dismiss as noise often have a great deal to tell us, however. Both scientists and artists learn to pay attention to the crumbs that other people are about to sweep under the rug. They learn to be good noticers. The same could be said about good teachers, good parents, effective politicians.
And certainly inventors. Saccharin was discovered more than a hundred years ago when a scientist doing some chemistry experiments stopped during his research to have dinner. He noticed that his dinner tasted unnaturally sweet and that he had a strange white powder on his hands. The powder made his food—and also his fingers—taste sweet. He paid attention and gave the world a nocalorie substance hundreds of times sweeter than sugar.
Post-it Notes were invented in much the same way. A chemist investigating ways to make a better glue found instead a glue so ineffective it wouldn’t permanently attach to anything. Instead of throwing the invention in the garbage, the chemist saw that it would work perfectly in a different context—sticky notepaper that could be peeled off without leaving a trace.
In fact, any signal takes its meaning from context; in a different context, the same message can have no meaning at all. If you send someone a message in code, but they have no way to decode it, your message has no more information than total nonsense. A tree falling in the forest may make a sound even if no one’s around to hear it, but it doesn’t convey a signal. Conversely, a scene devoid of signals to one person may contain a wealth of information to others. Think of Sherlock Holmes.
In The Collapse of Chaos, Stewart and Cohen point out that no signal has inherent meaning—outside the context of whatever hears it, or sees it, or decodes it. A compact disc, for example, may contain all the information necessary to play a piece of music. But without a CD player, it is only a pretty silvery disk—suitable for playing Frisbee, perhaps, but not much more. In the same way, a strand of DNA contains no message in the absence of molecules that can read the genetic code. “The number 911 has no inherent meaning,” they write. “In the context of the U.S. telephone system, it means ‘emergency’; in the context of a lottery it may mean ‘you lose’; and in the context of housing it means that you live on a fairly long street.”
Even DNA can transmit a different signal in different contexts, thereby producing very different results. “The caterpillar has the same DNA as the butterfly,” the authors remind us, “the maggot has the same DNA as the fly, the human embryo the same DNA as the grandmother she eventually becomes....”
As scientists know all too well, it’s easy to mistake a signal for noise, and vice versa. It happens all the time. Yesterday’s newfound planet is exposed as a glitch in the instrument that took its picture. That newly discovered particle turns out to be a cosmic ray. It happens the other way, too—when the background noise turns out to be a new particle, or the wiggle that was dismissed as an aberration turns out to be a planet.
Harvard science historian Gerald Holton poses the question this way:
How to determine which of all possible demonstrable events are indications of scientifically usable phenomena; which of them are really connected to the fixed regularities of nature, and which are merely passing phantoms, clouds with ever-changing form never twice the same, and thus reflecting only ephemeral concatenations? We might call it the problem of telling the difference between signal and noise.
One night recently, I joined astronomers Sallie Baliunas and Chris Shelton on top of Mount Wilson in the San Gabriel Mountains for a night of observing. The hundred-inch Hooker telescope, which dominates the site, is practically a shrine in astronomy. It is the place where Edwin Hubble first saw that our Milky Way galaxy was not alone in the universe, but in the company of billions of similar “island universes.” The Hooker also saw the telltale stretching of starlight that reveals our universe is still expanding from a central point in space and time, the explosive origins of everything—the so-called big bang.
During the 1980s the Hooker was mothballed. It stood as a historical curiosity, a World War I-era telescope with a mirror fashioned from French wine-bottle glass at the same factory that made the mirrors for the Hall of Versailles for Louis XTV. In 1995, however, it got a new set of optics that give it one of the sharpest views in the Northern Hemisphere. And, surprisingly, the clear air makes it one of the best observing spots in the world. (Ironically, the same inversion layer that traps the smog and suffocates downtown Los Angeles allows Mount Wilson to rise above it all for an exceptionally clear view.)
Harvard’s Baliunas is using the Hooker (among other things) to look for Earthlike planets around other stars. The night I was there, she and Shelton—who built the optics—were taking the system out for a test run. And first on their list to look at was a special star that they wanted to examine because it played a central role in a possibly important discovery. Baliunas had learned that morning that another astronomer thought the star might harbor an unseen planet.
The problem was, a star with a planet orbiting it would send almost exactly the same signal as a star that had a small companion—that is, a double-star system. Planets are seen by the way they tug on the stars they orbit, moving the stars slightly from their normal paths. In other words, the signal that the astronomer who found the planet saw could have actually denoted a double star. Double stars are quite common, and therefore not anywhere near as important as planets, which are rare (at least as for as we know). If the star with the planet was actually a double star, then the planet would be an illusion. And only the Hooker could focus the star sharply enough to find out.
Shelton burped the telescope to get out any ambient air, then homed in on the star. To everyone’s delight and astonishment, it appeared to be a double!—two stars piled up on top of each other like traffic signals. That meant no planet had been discovered. The signal meant a double star: bad news for the astronomer, but a good night for Baliunas and Shelton.
To make sure, however, they had to look at another star they knew was a single—to rule out any influence of the optical system itself. “You don’t know anything until you compare and contrast,” says Shelton. If the other star looked double, too, then it would mean that “noise” had crept into the system. If the other star looked like a dot, or a single, then it meant that the first star really was a double, and they’d made an important discovery.
To make a long story short, there was a lot of exuberance in the dome that night; the second star looked like a dot, and there were cheers and high fives all around. Then, just as suddenly, the dot cloned itself into two. The optical system was shaky. They had not made a discovery after all—except the all-important discovery that the system wasn’t working quite right.
Particle physicist Leon Lederman had a similar experience with a particle called the upsilon that soon became known as the “Oops, Leon.” Again, it was noise masquerading as signal. Later, the upsilon was pinned down with a better set of data.*
The discovery of antimatter is a good example of the opposite process—finding a real signal in what appears to be noise. A theorist discovered antimatter as a minus sign in an equation; an experimentalist saw it as a cosmic ray track that curved the wrong way. What’s remarkable is that neither gave in to the temptation to dismiss such an unexpected discovery as a mere aberration.
Complicating the noise issue is the fact that one person’s noise is another person’s signal. I was struck by this on a visit to Kitt Peak in Arizona, where the National Science Foundation sponsors some state-of-the-art telescopes. As we went from one to another, it became clear that the signals one astronomer was seeking were considered so much noise in the eyes of another. At the solar telescope, for example, a group of researchers studying Earth’s atmosphere was trying to get rid of telltale lines in the Sun’s spectrum that indicated the presence of various elements on the Sun’s surface. Since they were interested in elements on Earth, they had to get rid of “noise” from the Sun. Astronomers studying the Sun, however, have the opposite problem: for them, signals coming from Earth’s atmosphere interfere with what they wish to find out. Or as Kitt Peak astronomer Richard Green said to me: “One person’s floor is another person’s ceiling.”
Luckily, there are a whole range of tools to take care of the noise problem, some of them mathematical and scientific, some of them built into the human perceptual system. In fact, the ability to shut out noise can be an important sign that a newborn is developing properly. Years ago I had the pleasure of watching esteemed Harvard pediatrician Berry Brazelton going through his standard exam of a very young baby. He shone a light in the baby’s eye. The baby winced. He shone the light again, and the baby didn’t respond at all. Next, he rang a bell in the baby’s ear. The baby startled. He did it again. This time, no response.
All this indicated that the baby was developing normally, he said—and learning how to shut out “noise” is a major part of that effort. In everyday life, our brains continually ignore all kinds of similar stimuli: the nose that’s always protruding into your field of view; the feeling of clothes on your back or rings on your fingers; the sounds of the refrigerator or the air-conditioning system.
Another way to get rid of noise is to wear blinders—literally, or figuratively. When you want to see clearly over the glare of streets or water, you don Polaroid sunglasses, which selectively filter out only those light waves coming at you from a horizontal plane (that is, glare). (You can play signal-noise games with Polaroid glasses by rotating them ninety degrees; now the horizontal surfaces such as streets will light up, but the vertical surfaces such as windows go dark.)
Some detectors designed to track down elusive subatomic particles are almost all filters. Elaborate triggering systems operated by computers automatically toss out more than 99 percent of the data retrieved, because particle accelerators simply create too much stuff for anyone to look at. The triggering systems screen every “event” for noise and throw most of them away.
To avoid throwing out the baby with the bathwater, however, researchers need to know everything there is to know about the types of noise likely to mess up their experiments. This means they have to become experts in the irrelevant, connoisseurs of distraction.
Consider the plight of the team of researchers who finally saw the first wrinkles in space-time coming at us from 10 (or so) billion years ago, the fossil imprints of the conditions at the origins of the universe.
The fact that seeing such a thing was even possible grew out of a famous signal-noise confusion. At about the same time that theoretical physicists had figured out that our universe could have been created in a violent explosion, various astronomers had been puzzled by an unexpected hiss in the sky. Thinking it was noise, the researchers figured it was the result of a creaky instrument or some other interfering source—perhaps bird droppings in the antenna.
The theorists, meanwhile, had figured out that if the universe really did start with a big bang, it should be possible to detect traces of leftover radiation from the explosion; it should still pervade the sky, coming at us from the origin, now vastly stretched and cooled over the course of its lo-billion-year history.
The punch line, of course, is that the “noise” bothering the astronomers turned out to be that very leftover radiation from the big bang. But that was only the beginning. As soon as people realized that they were reading messages from the origins of the universe, they set about trying to decode them—in particular, to find the seeds of structure in the universe, the tiny lumps in space-time that eventually grew into the great strings and clusters of galaxies that drape the darkness like garlands. The structure had to come from somewhere, so researchers set out to see if they could perceive it in the leftover radiation.
Berkeley astronomer George Smoot, who led the effort, describes the task this way: “We were looking for tiny variations ... something less than one part in a hundred thousand—that is something like trying to spot a dust mote lying on a vast, smooth surface like a skating rink. And, just like a skating rink, there would be many irregularities on the surface that had nothing to do with those we sought.”
Their satellite, the Cosmic Background Explorer, or COBE, was able to pick up the signals, but it also picked up enormous amounts of static noise. This noise could come from stray heat sources or magnetic radiation or artifacts in the software analysis or a host of other things.
“It is difficult to convey how obsessed we were with trying to eliminate these errors,” writes Smoot. “I had started writing a list of potential things that could fool us back in 1974. Ever since, I had continually updated the list, adding new candidates....”
A new and better COBE is already in the works. When the astronomers go back for a second look, they’ll have to be on guard against a vast array of distractions. Foremost among them is the not so obvious fact that in order to get a good look at the cosmic background, you need to get rid of everything in the foreground. That includes signals that come from the instruments, from the Earth and Moon, from our own galaxy and other galaxies; it includes signal-absorbing dust inside our galaxy and beyond it. It includes the motions of objects as they sweep past each other in small eddies and large currents.
“There are many sources of confusion,” says UCLA astronomer Ned Wright, who also worked on the project—the motion of Earth through the solar system, the motion of the solar system through the galaxy, the motion of the galaxy through the local cluster. “We don’t know the orbital velocity of the solar system through the Milky Way,” he says. “We don’t know the velocity of the Milky Way through the local group.”
In effect, these cosmic background explorers are trying to pick up a whisper in the roar of a crowd. Particle physicists face much the same problem. “It’s not clear that they know how to handle the backgrounds,” said Fermilab’s Lederman, who won a Nobel Prize for the discovery of a particle so elusive it has been called (quite accurately) a spinning nothing. “When you’re dealing with complicated backgrounds, it’s not a numerical question. It’s a visceral question. How do you feel about these backgrounds? How well do you understand these backgrounds?”
There are other ways to deal with unwanted information, to zoom in on what you really want to know. Microscopes do this by magnification, effectively pushing all extraneous matter out of the field of view. Telescopes zero in on certain targets to the exclusion of others. For every decision to focus on one thing, a piece of context is lost. These trade-offs are inevitable. In a way, it’s like getting lost in thought, so concentrated on some delicious idea that the rest of the world simply fades away.
Astronomer Vera Rubin saw something no one else had ever seen recently because she spent several years looking in-depth at a galaxy that others had only skimmed. What she found was completely unexpected: stars rotating in opposite directions within the same galaxy. She compared her way of looking at galaxies with Georgia O’Keeffe’s description of looking at flowers: “Nobody sees a flower,” said O’Keeffe. “We haven’t time, and to see takes time, like to have a friend takes time.”
Astronomers who need to map the locations of millions of galaxies to make global sky surveys could never pick out Rubin’s unusual specimen. “If you look at one million galaxies, you’re going to miss the freaks,” says astronomer Richard Green.
Still another noise-elimination tactic is to make the signal itself louder and clearer, so it can stand out from the fray. Astronomers and particle physicists use photo multipliers to do this, politicians use megaphones, the rest of us wear hearing aids or eyeglasses. The Hooker telescope’s new optics act like a set of powerful spectacles that focus the light from a star so brightly that it stands out clearly even above the noisy skies of Los Angeles.
Smoot knew that when the signals came from the cosmic background, they wouldn’t be obvious or leap out, but slowly become apparent as they gained more and more strength. The weak signal, he explains, would get stronger by repetition, “like the ever-darkening mark left by repeatedly rubbing a pencil lightly over a piece of paper.”
Another noise-reduction tactic is to make a very restricted, focused search for exactly what you are looking for—the complement of wearing blinders. When you tune into one radio station, you also tune the others out. COBE went actively hunting for a particular kind of signal. Our eyes and ears do much the same thing as we choose to actively listen or look at one thing and ignore the rest. “Sense organs do not passively accept incoming data; they go fishing for it,” Cohen and Stewart point out—noting that more neural connections run from the brain to the ear than the other way around, and 10 percent of optic nerve fibers go the “wrong” way.
There are also many purely mathematical ways to get rid of noise and focus on signals. For example, one can assume that anything random is “noise” and eliminate it. Of course, that means you need to know what “random” is. Another way is to pick up only what changes, ignoring the constants in the same way as your brain ignores the motion created by the turning of your head or the image of your nose in your field of view. Computers can do this rather easily.
One very “hot” development in this field is wavelet analysis. Princeton’s Ingrid Daubechies—one of the leaders in this approach to signal processing—calls wavelets “mathematical microscopes.” They’re squiggly shapes that can be superimposed on confusing signals, working like zoom lenses to zero in on just the interesting part of a picture, for example, while keeping the overall landscape in view.
Essentially, wavelets are atoms of information, and they come in a vocabulary of types, like the table of elements. Each wavelet represents a slightly different shape. From this basic vocabulary, you can build just about anything—just as you can build anything in the universe from the hundred-odd elements. Also, just as you can deconstruct anything in the universe into a hundred-odd elements, so you can deconstruct (almost) any signal into wavelets.
The Navy has found wavelet analysis useful for finding enemy submarines in the noisy environment of the ocean; astronomers used them to pick out a subcluster of galaxies in a galactic group; the FBI is using them to store and retrieve fingerprints more efficiently; electronics engineers are using them to create TV signals that work both on the new high-definition TVs and older sets.
One major benefit of wavelet analysis is that wavelets allow you to look at either the big picture or the local picture without losing information about the other. (A little like watching a sailboat race with binoculars, zooming in to see an individual sail number, but not losing your view of the other boats in the race.)
Even noise itself can be put to work enhancing signals. Called “stochastic resonance,” this method makes use of the fact that sometimes random background fluctuations can boost weak signals to the point where they can be heard. An extremely rough analogy is a marble, say, sitting in one of those “cups” in an egg crate. Say you get a “signal” only when the marble hops from one cup to another. But in a quiet environment, the marble doesn’t have quite enough energy to make that leap. If you put the egg crate in your car and drive over a rough road, however, you may find it much easier to get the marble to jump into another cup. The random shaking of the ride is enough to help boost the marble over the barrier—making your “signal” visible.
All of these signal-processing techniques are used to sharpen otherwise ambiguous signals in a wide variety of settings, from CAT scans of the human body, to sorting out social or political trends, to finding new planets. And all of them in the end remind us of the power of invariants—the things that do not change, no matter what. Only by getting rid of the noise, the distractions of intruding signals or warped frames of reference,* distorting effects of scale or the artifacts of measurement, can we come to know what the natural world is really like.
When Shelton and Baliunas wanted to check whether their double star was really a double star or simply a glitch in the system, they moved to a different star to see if things would change. When Smoot was seeking to verify that he’d really found the primordial wrinkles in space-time, one of the most convincing clues was the fact that the wiggles he saw were scale invariant—meaning that each patch of sky revealed the same spectrum of wrinkles, from smallest to largest.
That’s exactly how the search for truth is supposed to work. You see something and then you try everything you can think of to make it go away; you turn it upside down and inside out, and push on it from every possible angle. If it’s still there, maybe you’ve got something.
“You have to kick the measurement every way you can to see what shakes loose,” says Baliunas—in a fitting metaphor for an astronomer who likes to soup up jalopies in her spare time. In other words, you do everything possible to eliminate all sources of possible error and confusion that could obscure what you’re trying to see. Then you’re ready to look for the truths that can be found in the mathematical art of pattern perception.