The Numerati

by Stephen Baker

  Isn’t it strange, I say to Andresen as we make our way back from Norman’s corral, that we’re working to monitor the health of cows before we get around to people?

  “Cows don’t care much about privacy,” Andresen says. He’s wearing a brown outback hat to keep the sun off his fair face, which he says burns easily. And he’s wearing socks under his sandals, the way northern Europeans do. “If they care,” he adds, “they don’t let on.”

  Then again, even if cows like Norman knew and cared about privacy, would it be utterly foolish of them to sacrifice a dose of it for medical monitoring? Consider yourself. If the medical industry came up with a system like Andresen’s—presumably one that didn’t involve clamping a fistula onto your stomach—would you sign up? This is the kind of question we’re likely to be facing as sensors and computers and wireless networks grow ever stronger and cheaper. Forget about the rest of the herd, they’ll say. We can create a custom service just for you—provided you fork over your data.

  Already, auto insurers are up to something very similar. In Britain, Norwich Union offers special rates to drivers who agree to place a black box full of recording instruments inside their car. This way, the company can monitor driving behavior and offer further discounts to drivers who keep the speed down and stay clear of high-risk roads and neighborhoods. In other words, the insurers are analyzing not just the drivers’ profiles or records, but also their behavior.

  This already happens in a rudimentary way in health insurance. Smokers often pay higher premiums, for example. But imagine how much more sophisticated this model could become if we were wired with sensors. New insurance markets could take shape, all of them feeding on the fluctuating signals streaming in from our bodies. In this world, buying health insurance could start to feel like taking out a mortgage. Here’s the choice: Lock in a fixed rate, and the company insures you no matter what. But it costs a mint. For lower premiums, you might opt for a floating rate, with premiums that rise and fall with your health risk. Those who play these numbers shrewdly come out on top in medicine, replicating their success in finance. They’re data masters. The rest of us underwrite their winnings.

  I can just imagine angrily calling the help line when I see on an insurance bill that my rate edged up a few dollars despite the impressive drop in cholesterol. “But your blood alcohol went above the threshold six times,” comes a voice from a distant country. He turns a deaf ear to my arguments that red wine is part of the regimen . . .

  We’re sitting at a table in Andresen’s computer lab, which is chock-a-block with components and circuitry and blinking lights. Andresen, his hat hanging behind him from a loop around his neck, is up at the whiteboard, describing a type of cow known as a “dark cutter.” These are a scourge. Somewhere along the line—no one knows exactly how or when—dark cutters appear to have experienced some sort of trauma. As a result, their meat is bluish instead of red. It appears to be emptied of its blood. No more T-bones, porterhouse steaks, or top-of-the-line filets mignons from these beasts. The sole option at the slaughterhouse is to grind dark cutters into cheap hamburger. Each one represents lost money.

  Now let’s imagine that a few years from now, Andresen, Warren, and their team have successfully stitched together their network of cows. A certain number of those animals, inevitably, will turn out to be dark cutters. Researchers will have at their fingertips the lifetime record of every one of these cows, each wag of its head, each snooze in the shade. They’ll be able to feed this data through their computers and search for patterns. Do the dark cutters have anything in common? Did they suffer jolts or get too cold on the road trip to the feedlot? Did they sleep less than the others or eat at a different rate? It’s all guesswork at this point. But what they discover could eventually lead to adjustments in the way cows are raised or transported. Perhaps certain practices handed down by generations of cowboys will have to be dropped—and replaced by science.

  What if the data shows that from their earliest days as calves, dark cutters behave differently? Researchers could turn this knowledge into a predictive tool. This would give them a behavioral profile, etched in math, of a dark cutter. In the same way, each calf could conceivably be scored on the likelihood that it would grow up to produce bad steak. What then? Would ranchers cut their losses and send high-risk calves to be slaughtered right away?

  Such questions are at least a few years away—at least for the cows in Kansas. Building a bovine network is an immensely complicated undertaking. Getting all the sensors to work in sync is a slog, and each one presents its own challenges. The heartbeat, for example, is hard to distinguish from the sounds of fluids and gases going about their noisy business inside the animal. Radio signals struggle to escape thick walls of beef. Batteries wear down and die. Then there are networking conundrums. How do you update software on a thousand head of cattle, or protect them—heaven forbid—from hackers? Still, those are technical issues. Many of them resemble the challenges that engineers mastered in the cell phone industry. If the economic payoff is big enough, they’ll work through them.

  And when they do, the focus will inevitably move to implanting the devices in us. Governments looking to cut health-care spending will certainly be interested. Electronics companies, as Eric Dishman will attest, view health surveillance as a mouthwatering market. And for the insurance industry, the more information they have about us, the better they’ll be able to calculate risk and create a host of new personalized services. Put those groups together, and you have a mighty coalition. Who knows? Given the potential health benefits, many of us may well cheer them on.

  LODGED SOMEWHERE on Microsoft’s massive computers are a host of e-mails my mother sent to my Hotmail account over the years. She averaged about three a week. A historian or a sociologist could look through them and study the patterns of an American couple early in this century marching purposefully into extreme old age. The e-mails report on Thursday night dinners with the grandchildren and dog walks in the rain. She writes about her activities on the vestry at church and about my father’s latest letters to the editor at the Oregonian, deploring the treatment of prisoners at Guantánamo Bay. In a couple of the e-mails she writes about taking a cab to a health center in Portland where she and my father participated in a study on aging and cognition, or as she put it matter-of-factly, senility.

  Now I find myself at the Oregon Center for Aging and Technology, or Orcatech, a hulking state-of-the-art health center on the banks of the Willamette River. Inside, legions of Portland’s elders labor on long rows of treadmills. A café in the spacious sunlit lobby sells expensive lattes. Outside, a ski lift hoists doctors and patients in a gleaming glass pod to a complex of hospitals at the top of a hill. This is where my parents came to lend their brains to science. But in the future, I realize, as I talk to researchers, the elderly can save the cab fare. With the spread of sensors, the cognitive laboratory moves into our homes, where analysts will keep tabs on the working of our brain by tracking the patterns of our daily activities. Nearly everything we do—if studied in meticulous detail—provides a glimpse inside our head. I hear this from researchers constantly. Whether they’re discussing the changing pattern of steps on the magic carpet or the adherence to a pharmaceutical regimen, they add, “This also gives us a good cognitive read.” It’s like a two-for-one sale. Test anything, and you get brain results as a bonus. In this type of analysis, a long string of e-mails, the kind my mother sent to me, would qualify as Exhibit A.

  How do analysts correlate irregular footsteps and typos with dementia? The research starts in the laboratory, with the sensors nowhere in site. Micha Pavel, a Czech-born mathematician, explains how he runs seniors through a series of drills over time to test their memory. For each person, he draws up a model of working memory. Like an actuarial chart, it predicts the “survival” of each piece of information. In some, the lines are fairly flat. The memory’s holding. In others, it curves sharply. If you look at each forgotten fact as a death, as Pavel does, some
of these people are hosting full-blown epidemics. “We try to assess the probability that an item will be lost,” he says. “Is it a function of time or intervening events?” In most cases, he says, it’s new events that push out memories, as if each person has a limited storage space, what he calls a “memory buffer.” Naturally, the people to worry about are those whose buffers are shrinking. Once Pavel has this memory data, the next step is to study the rest of these people’s lives—their mouse clicks, word choice, sleeping patterns—and to draw correlations with what’s happening inside their head. The work has barely begun. But studying their written words is a natural place to start.

  For centuries, people have been scrutinizing letters for insights about their loved ones. When handwriting deteriorated and non sequiturs popped up, they had reason to worry. I certainly fretted as I saw my mother’s e-mails grow shorter and less regular. And when I began to see typos in messages from this former legal secretary, I was alarmed. But this was late in the process, only a year or two before she died. Could a rigorous statistical analysis of her typing patterns, sentences, and word choice have pointed to problems years or even a decade earlier? In cases of early detection, doctors can quickly start medicines and therapies to forestall or slow the deterioration. Dishman, meanwhile, is working on a host of technologies to help Alzheimer’s sufferers cope. One is a phone prompt. When a friend or family member calls, the person’s photo and name pop up on a screen, along with some details, such as the last time that person called.

  I walk into the model house in the Orcatech lab. It’s strewn with gadgets that they’re testing in the homes of scores of Portland’s elderly. In one corner is a sensor-wired bed, like the one the little dog jumped on. Lying on the floor is a cane with a boxlike appendage at the bottom. It’s designed to measure how much the user leans on it—signaling the possible weakening of that person’s legs. Amid these gadgets in the model house are two computers, which I think of as the brain and a separate nerve center. One computer sits hidden in a closet. It picks up all the wireless signals from sensors around the house and relays them back to Orcatech. The other PC sits in plain sight. It’s outfitted with a host of games, along with standard word processing and e-mail programs. Every interaction with the computer, every keystroke and click of the mouse, sends back details on cognitive trends. Researchers are still at the early stages now, trying to build a baseline for each user. But within a couple of years, says Pavel, “we’re going to be measuring the motor speed, the keyboard interactions, and the complexity of the words they generate.”

  One model for this analysis is a study of the writings of the prizewinning British novelist Iris Murdoch, who died of Alzheimer’s disease in 1999. Murdoch left behind decades of written manuscripts, a treasure trove for cognitive researchers, at University College London. They studied her word choice at different points in her long career and found that in her last novel, Jackson’s Dilemma, published in 1995, she used a simpler and less varied vocabulary than in earlier works. In fact, they saw that her language followed a curve. It grew more complex from her first novel, Under the Net, to one written at the height of her career, The Sea, the Sea, before falling off at the end. The scary thing (from my perspective, at least) is that with advanced statistical analysis of different writings, from blog posts to e-mails, researchers (or even employers) may pick up the downward trend of our cognitive skills long before we even suspect it.

  Eric Dishman’s team is searching for similar clues in speech and social interactions. In 300 homes in Oregon, they’re piecing together models of people’s relationships. How often do they make phone calls, and to how many people? How often do friends visit? For each person, they’re turning this data into a score, a so-called “social health index.” If people’s indexes fall, it’s an indication that something has changed—perhaps a deepening of dementia. They’re also testing a subject’s responses to familiar voices on the phone. Usually, people recognize voices of close friends and family members instantly. If there’s a lengthening pause before that recognition, they want to capture it. “We’re looking at milliseconds of difference,” he says. “That might be an alert that there’s some kind of trouble.”

  This analysis can get complicated. Picture yourself as one of Dishman’s subjects. You’re watching a basketball game on TV. It’s overtime between Dallas and San Antonio. Tony Parker drives the lane and is fouled. The phone rings and you reach for it.

  “Hello.”

  “Hi.”

  That’s Nowitzki’s sixth foul. He’s screaming at the ref.

  “HELLO?” comes your sister’s voice from the phone.

  “Who’s this?” you say absently, watching the replay.

  “Me, you idiot . . .”

  Plenty of things besides the early signs of dementia can interfere with our thought processes. Music, anger, and sleepiness all throw us off our stride, as do cocktails. In time, behavioral scientists will try to incorporate these distractions into their models. The only way to do it, as you might have guessed, is by learning even more about us.

  Dishman is working on one gadget that takes this to an extreme. It’s a wearable device to help people deal with fits of hyperaggressive anger. This involves a heart monitor connected to a souped-up cell phone. A user, Dishman says, starts by filling out an electronic form listing “the people most likely to stress him out.” His boss might be one. Next he lists the places that cause stress. “I might be relaxed at the pool,” Dishman says, “but more stressed out at Intel.” Finally, the user makes his electronic calendar available. By combining all of this with the location data from the cell phone, the key pieces are in place: where the person is, who he’s with and what he’s up to. So if the heart starts racing, Dishman says, “the system can look at the calendar and say, ‘Oh my God, that’s a stressful meeting because Kevin’s in the room. We’ve got to get him out of there!” At that point, the phone rings. The user answers and is prompted by the computer, according to the most dire scenario, to leave the room. Then he’s led through a series of questions. “Are you clenching your jaw?” it asks. “Your fists?” It recommends taking some slow breaths or getting a drink of cool water.

  For now, the Intel team is trying out this technology on students who don’t have hyperaggressive disorder. It gives them cues for managing stress. “I’ve analyzed your calendar,” Dishman says in his machine voice. “And I see you have meetings every half-hour, back to back, for 12 hours. Five of those meetings are with people you’ve flagged as the most stressful, including your boss and this asshole at work. Do you want to give yourself a break?” Ultimately, he says, this type of data, from our schedules to the people most likely to fill us with rage, should merge with our other health information, including our genetic particulars, the medications we’re on, and the feeds streaming from dozens of sensors. He sees each one of us eventually building what he calls a “dashboard to manage our health and wellness.” This will be the control panel for our lives. There is almost nothing about us that these dashboards will not be eager to learn. And naturally they’ll incorporate all of the trends and medical insights gathered from the dashboards of everyone else.

  So here’s a question. Let’s assume that some form of Dishman’s vision takes shape, and each one of us builds a far more detailed repository than we have now of medical and personal data. Who do we share it with? “This comes up in every study we do,” he says. “How do you help an elder with Alzheimer’s, who’s not computer savvy, decide who gets the data and who doesn’t? It’s a huge design problem. You go into one house, and they say, ‘Anyone can have this data.’ You go into another, and they say, ‘My son can have the data on finances, my daughter can have the data on health, and my other son who I’m pissed at can’t have any data about anything.’” Coming up with laws and technologies to help people wisely manage the privacy of medical records is a challenge every bit as daunting as predicting Alzheimer’s or short-circuiting a hyperaggressive fit.

  As Dishman goes on
about privacy, I’m pondering my own medical secrets. Who should I share them with? Then I wonder, more to the point, how eager am I to learn about them myself? As research advances into patterns of disease, behavior, and genetics, we’re going to be bombarded with loads of statistical projections about every conceivable malady. Say you learn that you have a 20 percent risk of going blind in old age from macular degeneration. You can lower the odds or delay its onset, you read, by altering your diet, quitting smoking, and taking some pills. Do you change your life to respond to that risk? How about a 7 percent chance of having a stroke in the next ten years? How seriously would you take that? What if your risk in some disease is 8 percent and the national average for your age group is 6 percent? Is that worth paying attention to? Our medical charts, covered with numbers and probabilities, will start to look like the scorecard of a Las Vegas bookie. We’ll be awash in the odds of our own demise.

  Here’s my prediction. As the analysis of our medical data grows, new types of consultancies are going to process these reports for us. Picture a company, SpareMeTheDetails.com. They’ll add up all of our reports and give us a life prescription, a combination of medicines, dietary tips, even exercise regimens on the magic carpet, all of them designed to keep various scourges at bay for as long as possible. This process, of course, will involve sophisticated algorithms based on probability, which will create loads of work for the Numerati. But the point is that many of us, in an age of exploding medical data and analysis, may be happy to pay for the privilege of remaining, to one degree or another, in the dark.