The Numerati

by Stephen Baker

  Like energy. Takriti doesn’t like to broadcast it, but he left Big Blue in 1999 for Houston, where he worked for Enron. Back then, Enron was not only innovating the kind of corporate fraud that would lead to its collapse. It also ran a world-class mathematics laboratory. The entire world, as Enron saw it (and was soon to demonstrate all too vividly) was awash in uncertainty. People had trillions of dollars riding on chance. If you looked at weather as a topsy-turvy market, for example, theme parks were betting on sunshine, farmers on rain. Enron’s math team could calculate the weather risks and then develop indexes and financial options for cold fronts and dog days. Everyone could hedge the weather, and Enron would turn this into a business. Given enough mathematicians, it seemed, every chancy element in the world could eventually be quantified, modeled, and turned into a financial instrument.

  Takriti’s stock soared at Enron. And when IBM called in late 2000, they offered him the top job in stochastic analysis. Takriti jumped. He got out of Houston, it turned out, barely a year before Enron’s collapse. His new focus at IBM would be every bit as hard to quantify and predict as flash floods in the Mojave Desert or the looming corporate bankruptcy in Houston. Takriti would be modeling human workers.

  I tell Takriti that being modeled doesn’t sound like much fun. I picture an all-knowing boss anticipating my every move, perhaps sending me an e-mail with the simple message, “No!” before I even get up my nerve to ask for a raise. But Takriti focuses on the positive. Imagine that your boss finally recognizes your strengths, he says—maybe ones that are hidden even to you. Then he “puts you into situations where you will thrive.”

  If your performance is stellar, companies eventually could wield your mathematical model as a specimen of workplace DNA. And they could use it, in a sense, to clone you. Imagine, says Aleksandra Mojsilovic, one of Takriti’s modelers, that the company has a superior worker named Joe Smith. Management could use two or three others just like him, or even a dozen. Once the company has built rich mathematical profiles of their employees, it shouldn’t be too hard to sift through them to identify the experiences or routines that make Joe Smith so good. “If you had the full employment history, you could even compute the steps to become a Joe Smith,” she says. Most of this, of course, would involve training programs, not genetic manipulation. And the real Joe Smith may have intuitive smarts or a knack for design that just cannot be replicated. “I’m not saying you can re-create a scientist, or a painter, or a musician,” Mojsilovic says. “But there are a lot of job roles that are really commodities.” And if people turn out to be poorly designed for these jobs, they’ll be reconfigured, first mathematically and then in life.

  When Samer Takriti sits down to define one of his colleagues in symbols, he looks to economists and industrial engineers for guidance. They’ve been modeling complex systems for decades. Economically, he regards us as components in a labor market. Our value rises and falls with demand. In that sense, we fit into the financial equations developed on Wall Street. And when we’re employed, what do we do? We work with colleagues to build things and create value. So, boiled down to numbers, we share at least a few mathematical properties with the components that are unloaded every day at IBM’s huge microprocessor factory up the road, in Fishkill, New York. Look at us one way, and we’re stocks. Change the perspective, and we’re machine parts.

  Of course, this isn’t entirely fair. We’re more than stocks and parts, quite a bit more. Takriti is the first to admit it. It’s because we’re so different—so hard to predict—that Takriti needs a team of 40 Ph.D.’s, from data miners to linguists, to decode our behavior and our traits. They catalog what they find—each of our gestures, each of our skills—into symbols that a computer can digest. “Everything must be turned into numbers,” Takriti says.

  One of Takriti’s challenges is to help IBM develop a taxonomy of the skills of its 300,000 employees. On its balance sheet, IBM lays out the value of many other assets, from supercomputers to swiveling Aeron desk chairs. When strategists at the company are figuring out whether to sell a division or invest more money, they pore over these figures. They sketch out rosy and grim scenarios. They do the numbers.

  But how do they “do the numbers” on you and me? Yes, they know how much we cost. Anything that’s counted in currency fits neatly into their equations. But what do they get for that money? How can that be measured? What’s our potential? Will there be a glut of people like us in the next few years? A shortage? Planners want answers. To carry out these calculations, they have to turn us into something that, like financial instruments, can be measured over time. Picture an average worker in an industry that’s chugging along at its usual pace. At the risk of seeming cold-hearted, let’s give that imaginary laborer a rank based on his current value. Call him a C. If the industry heats up and more such workers are needed, his value rises, maybe to C+ or even B. If he picks up more skills or starts working harder, the same thing happens. His stock rises. But if the industry plunges into recession and companies shut down operations, our worker finds himself in a surplus market. His stock plummets, down to a D or even an F. We’re all too familiar with this dynamic. Workers find jobs in boom times and get laid off in slumps. But often the process has little to do with a worker’s value. In some companies, the last workers hired are the first to get the boot. That rewards longevity, not value. Sometimes it’s the friendly workers who survive, or even workers who have a knee-breaking cousin in the mob. These are metrics a caveman could grasp. The Numerati have a different plan altogether. But how will they calculate our worth? How will they turn us into quantifiable financial instruments?

  The first step is to break us down into little pieces. These are the characteristics we share with others, the bits of us that can be squeezed into columns and assigned numbers. Computers, after all, aren’t yet capable of appreciating us as the integrated and complex beasts that Leo Tolstoy wrote about. You might have the nicest smile on earth and wonderful rapport with colleagues. Maybe you’re mean, or smell like onions. There’s no room, at least in the early versions of IBM’s employee database, for those personal details. Some of them may be crucial. They may represent the real you. But the database understands us largely as a mosaic of résumé items, from job categories to mastery of the computer language C++ to fluency in Mandarin.

  It’s pathetically shallow. Consider what happens when you sit down in a room to, say, hammer out a new marketing campaign with five colleagues. This is life in the analog world. Your brain, by far the most sophisticated computing device known in the universe, processes an astonishing range of data. It perceives a wrinkled nose, a sideways glance, a hint of sarcasm, a flash of disdain. It ties together smells and sounds, and it links them with other memories and lessons from the past. Add up all the words and looks and gestures, and your brain picks up thousands, or even millions, of signals emanating from those five people. In his book Strangers to Ourselves, Timothy Wilson of the University of Virginia notes that as data streams in from our five senses, the brain grapples with more than 11 million disparate pieces of information per second. Today’s computers cannot handle such complexity. IBM’s mathematical system may scan each of us for a mere five or ten data points. I’ve had dogs that dig deeper into human nature. However, once we’re represented as bits of math, the machine can do something superhuman. It can mix and match us in a fraction of a second with a million, or 100 million, others. That scale promises new efficiencies—and insights.

  Imagine what IBM’s bean counters will be able to do once all of the company’s workers are classified by their skills. They’ll start running ever more detailed numbers on workers—just as they do on other investments. They’ll attempt to calculate the financial return for each job category and each skill, whether it’s Java programmers or office managers. They’ll compare productivity in ever-greater detail, worker by worker and region by region. This will help them decide which jobs to send offshore. And they’ll be able to measure productivity based on dozens o
f yardsticks. How productive are workers in your category as they reach ages 45, 50, and 60? Once the company has those numbers, they might be able to calculate not just the present value of workers but also what they’ll be worth down the road.

  This will take some getting used to. To date, we’ve managed our human relations in an old-fashioned economy, one largely lacking in numbers and metrics. For the most part, whether we’re looking for a favor or even a mate, we’ve bartered: here’s what I’ll provide you; here’s what I want in return. Nothing to measure, little to count. For centuries, even commerce worked this way. You won’t give me those two goats for this table? How about if I throw in a hammer? This process is painfully inefficient. Each barter calls for more haggling. Values fluctuate. No surprise, then, that bartering retreated into the hills as soon as societies came up with a numeric symbol for value—money. This was a triumph of the earliest Numerati. It provided a math tool to count and calculate and compare a world of different things. And it eventually led to expanded commerce, global markets, the numbers blazing on the liquid crystal screens of the Tokyo Stock Exchange. Now Takriti and his team are turning us into symbols so that we can take our place in new human markets.

  Just the way brokers run portfolios of junk bonds or emerging market stocks, the Numerati are dropping us into portfolios of people. It’s happening in industry after industry. Alamo Rent A Car, for example, buys a portfolio of romance-movie lovers from Tacoda and then compares its performance to that of others. If Takriti and his team manage to reduce IBM’s work force into a coherent portfolio of skills—something a computer could understand—IBM could soon deploy its labor in much the way that it manages its financial investments. That’s precisely what Takriti has in mind.

  This has been the case for years in baseball. On my shelf I have a fat encyclopedia listing baseball statistics on every major league player since the 1880s. But now the baseball Numerati are slicing and dicing the data nearly as fast as the Wall Street quants (quantitative analysts, that is). They’re coming up with new metrics—new ways to model the players mathematically. One new statistic is called WARP, “wins above replacement player.” On a quant site called Baseball Prospective, I’m looking at a profile of Carlos Beltran, the elegant switch hitter who patrols center field for the New York Mets. In late 2004, he signed a seven-year contract for $120 million, or an average of $18 million a year. After a stellar 2006 season, his WARP stood at 10.6. That means that if the Mets swapped him for a commodity replacement player making a laughable half million a year, the team would win nearly 11 fewer games in a 162-game season. Each additional victory, by these numbers, would cost the Mets $1.62 million—an affordable price for a rich New York team. But will Beltran’s WARP be nearly as high in 2010, when he’s a creaky 33 years old? Sadly, Baseball Prospective says no. It predicts that Beltran’s WARP by the decade’s end will decline to 3.6 and that he’ll be worth only $5.8 million—far less than he’ll earn.

  Is that calculation right? It’s anybody’s guess. Does the WARP account for Beltran’s intangibles, all those immeasurable qualities—the helpful batting tips he gives a rookie or the way he distracts the other team’s pitcher as he dances off first base? In other words, do the numbers reflect reality in all of its complexity? Often they fall short, even in the statistically exuberant realm of baseball. Pick the wrong numbers, and they can lie. That’s no secret. But just try making that argument to your boss when your numbers slump.

  I confess to Takriti that I find the prospect of being scored like Carlos Beltran a bit unsettling. It’s nice, I’ve found, to live and work off the database, in the foggy barter economy. True, it’s a real headache for bean counters. But the unmeasured universe can be a forgiving place. Smiles, friendships, and even artfully spun yarns all count for something there—maybe a smidgen of job security or even a raise. Workers in a measured workplace are on their own and more likely to rise or fall with their numbers. Precious few of them have seven-year contracts like Carlos Beltran’s. And for these quantified masses, the security of the flock fades away. After all, each lazy or incompetent worker who survives in the mathematically assessed workplace represents a market inefficiency. Once the measurements are in place, these workers will presumably plunge in value or be purged, just like an underperforming stock in a portfolio.

  Think you can manage life in a portfolio of workers? It could be the best thing that ever happened to you. Some stocks shoot through the roof, and certain workers will too. But being dropped into a portfolio is only the beginning. Takriti and his team are already building the next stage, in which we’ll be understood and interpreted in far greater detail.

  ON A NOVEMBER morning in 2006, IBM’s chairman, Sam Palmisano, stepped up to a podium in the Forbidden City, in the heart of Beijing. He had an announcement to make. Palmisano was dressed in his standard business suit and wore his trademark horn-rimmed glasses. But as he made his way to the podium, something about him looked not quite real, more like a cartoon. This was not the real Palmisano, it turned out, but an avatar representing him. And the Forbidden City he visited was a digital simulation. IBM technicians had built it and mounted it within the virtual world called Second Life. Journalists who wanted to hear Palmisano’s announcement grumbled about this venue for weeks. They had to sign up for Second Life and attend as their own avatars.

  By staking IBM’s blue flag in a simulated world, Palmisano was pointing to the company’s future. Already, engineers around the world use computer simulations to design electric turbines and fine-tune the traffic flows in major cities. The way IBM sees it, entire business processes will one day be simulated. Picture managers, their fists grasping joysticks, trying out new industrial approaches and calibrating operations as if they were running their own version of the video game The Sims. If Takriti and his team master their next assignment, the avatars on the screen will be the mathematical models of IBM’s workers.

  This process is just beginning, and Takriti already waxes nostalgic for the old days, when it was machines that were modeled. They’re simpler. Machines don’t cheat, feud, pout, develop serious drinking problems, or get depressed. And they don’t come up with great and transformative ideas. Takriti goes on for a while about the maddening randomness of humans.

  I interrupt to ask about the math involved. I point to the whiteboard covered with formulas and notations, some of them snaking up and down to make room for others. “What are you working on here?” (Some of these notations are new to me.)

  He shrugs. Takriti, like many Numerati, tends to downplay the complexity of the formulas he scribbles so effortlessly. He rejects the notion that he and his confreres draw their algorithms and equations from a magical toolbox. Part of this is modesty. But Takriti also has the conviction that even the nitty-gritty of stochastic calculus would be clear to outsiders if we just sat still and paid attention. He starts to explain one of the formulas. Then he stops. It’s the humans that are hard to figure out, he insists. “Math is the easy part.”

  For decades, IBM researchers have been transforming bigger and bigger pieces of the company’s business into math. The science they use, known as operations research, was born during World War II. German submarines, known as U-boats, were attacking convoys and sinking lots of ships. How should the convoys be deployed, the mathematicians were asked, to minimize the damage? Was it better to travel in large groups, escorted by many destroyers? Or would smaller ones be harder for the U-boats to track down?

  Math whizzes at the U.S. Antisubmarine Warfare Operations Research Group (ASWORG) built mathematical representations of the convoys. These were models, and they operated within a set of constraints—conditions imposed by the real world. The ships couldn’t move faster than a certain speed, for example, and they had to carry enough food and fuel to reach their destination. They had to steer clear of icebergs. The mathematicians also had statistics on the U-boats—the size of the fleet, the range of the subs, the deadliness of their missiles. Using this information, they cou
ld model the naval war. Each vessel was linked to others by numbers, by the probability that something good, bad, or indifferent would happen to it. These fleets in the North Atlantic existed in their model as a web of statistical relationships. As the researchers tinkered with the fleet in the model, the odds changed. The ASWORG team was able to calculate that large convoys with big escorts were significantly safer. They determined how far down to send depth charges to inflict the most damage on enemy subs. As the U.S. Navy put these formulas into practice, the attrition of the convoy ships dropped. Shipments reached Britain. By the end of the war, mathematicians were using similar methods to boost the efficiency of anti-aircraft defenses and fuel depots.

  Takriti is telling me about one of the giants of the field, George Dantzig, when he jumps to his feet, reaches high on a shelf for a big textbook, and starts flipping through it. “Dantzig did the mathematics of marriage,” he says. “Maybe it’s something you can use in your dating chapter.” Dantzig, I learn, regarded multiple sexual partners as variables, and he attempted to prove that monogamy—at least from the cool-headed perspective of an operations researcher—yielded better results than polygamy. Takriti doesn’t find the details in the book. Maybe I could find it on the Web, he says. That turns out to be a cinch, though I think it’s safe to say that Dantzig’s study, while fascinating to the Numerati, left the institution of marriage largely untouched.

  But outside of matrimony, his influence is with us every day. In 1947, the Berkeley-trained mathematician came up with the so-called simplex algorithm. An algorithm is nothing more than a recipe—an ordered set of commands. This one was a recipe to guide intelligent decision making. If farmers wanted to know which type of seed to plant in a particular soil, or if steelmakers wondered whether to haul coal on trucks or barges, operation researchers had answers. They just needed the numbers, the constraints, and the goal. Using Dantzig’s algorithm, they could find the point where the objective, whether it involved dollars or tonnage, reached its zenith, its optimum point. Then they could calculate, working backward, how to come up with that result. Known as optimization, this process now guides logistics and planning and network design in much of the modern world. If you want to fly from Los Angeles to New York, Travelocity’s optimization program flips at lightning speed through 10,000 possible routes and finds the one that will make the most sense for you and the most money for Travelocity and its partners (profit is one of its constraints). Military planners optimize helicopter routes over insurgent hotbeds in Iraq. And when you make a call on your cell phone, an optimization program chooses the best pathway of towers to convey the signal.