The concept of the pedometer is largely credited to Leonardo da Vinci, who built a number of distance-measuring devices to aid in his cartography. In one of his sketches he added a pendulum to an early version of a surveyors wheel in order to explicitly track the number of steps taken by Roman soldiers (Figure 2-1). Pedometers became somewhat popular with European watchmakers in the 1700s and were of particular interest to Thomas Jefferson and James Madison, both of whom documented their interest in various letters. In one of those letters, Jefferson describes a pedometer watch he found in Paris in 1786:

Figure 2-1. Sketch of Da Vinci’s pedometer

Pedometers became somewhat popular in the United States in the mid-1900s. In 1938, they were marketed as toys, and the Hike-O-Meter was a premium prize you could order from Wheaties cereal (Figure 2-2). Similar aluminum pedometers were sold in the United States throughout the 20th century, but they didn’t attain widespread use until the early 1960s, in Japan. In the early 1960s, Dr. Yoshiro Hatano was doing research on the walking activities of Japanese people and found that the average person walked about 3,500 to 5,000 steps per day, but in order to be healthy, they should walk 10,000 steps each day to burn 20% of their caloric intake. In 1960, Dr. Hatano started selling a pedometer called the Manpo-kei (Figure 2-3) which translates to “10,000 steps meter” and it was hugely successful. It’s estimated that even today Japanese households average 3.1 pedometers. Digital pedometers started popping up in the mid-1980s in Japan and slowly made their way to America shortly after.^[4] In the 1990s, they were somewhat common and even began to show up on (Japanese-made) digital watches.

Figure 2-2. Wheaties Hike-O-Meter

Figure 2-3. Manpo-Meter

As popular as early digital watch pedometers were, they lacked one very important function, data logging. If you wanted to have any awareness of your progress over time, you had to write it down somewhere, which greatly limited the usefulness of the technology. However, in 2006 Nike introduced the Nike+iPod Sports Kit, which included a sensor that was built to fit into certain Nike gym shoes and wirelessly connected to iPods or iPhones to log your data. The Nike+ kit wasn’t meant to track your steps all day; it was only intended for working out, but the logging functionality turned out to be crucial to the adoption of the technology and ended up leading to more widespread production and use of fitness trackers.

Connecting the Dots

Connected fitness trackers really began to pick up steam in 2012 with the introduction of the Fitbit One (Figure 2-4), a fob-looking tracker that users clipped on to their pants, bra, or shirt. The default step goal for the Fitbit One was Dr. Hatano’s 10,000 number. It had about a 10-day battery life and tracked steps, flights of stairs, and sleep (it came with a sleeve that you could put it in for sleep tracking). The Fitbit One was my first fitness tracker; everybody in my studio got them, and I was really excited about it because this was the first time I was going to be able to break that barrier between the physical and digital world! I was used to being able to access this type of behavioral information in the more conventional (screen-based) products I worked on, and it had been incredibly useful. At the time, I was working on a big project for Scholastic that would tie together a bunch of data to target reading levels of students and push them to pick new books that would help them grow as a reader. Thus, I was understandably enthralled with the possibilities that encoding physical data represented.

Figure 2-4. Fitbit One

In reality the physical barrier crossover wasn’t the tectonic shift that I had been hoping for: the data I could get from the tracker was aggregate and limited, but it did give far more insight into the black box of physical activity than we had before. Even at this low fidelity, activity and other metrics were interesting to see. As it turns out, I was pretty sedentary, and I really had no idea, I immediately began walking more and taking the steps instead of the elevator at work. In addition to the basic input-and-output type correlations (I feel better if I walk more than 11,000 steps), one of the more interesting outcomes of continued use of these devices is knowing what certain things feel like in the context of your life. When you walk 10,000 steps for the first, second, third, twentieth time, and so on, you begin to get an idea as to what that day looks like as a whole within the very specific context of your life. You understand what changes you need to make to your lifestyle to achieve 10,000 steps per day. This is experiential knowledge, and it’s different from explicit knowledge because it must be experienced. It’s also different from tacit knowledge because it’s not universal (like riding a bike). It really hinges on combining the repeated tacit experience of walking 10,000 steps and combining that information with the observation of lifestyle changes to your personal situation to take those 10,000 steps.

In the case of my day, I knew that I had to walk to work, take a little walk at lunch, and walk my dog for a little longer than usual if I wanted to hit my 10,000 step goal every day. For a lot of people, gaining this experiential knowledge—which took maybe a week to achieve—was the primary benefit of owning one of these early fitness trackers, and once that information is more encoded into your lifestyle and daily routine, the usefulness of the tracker begins to decline. When you know what you need to do to achieve this goal and you do it repeatedly, the specific measurements of this information become less and less useful to you. As the device provides less and less value, you get to a point at which the benefits outweigh the costs of upkeep. With the Fitbit One, upkeep meant charging it every 10 days or so, and always remembering to take it out of your pants when you were doing the laundry and clipping it to yourself every morning. For multiple members of my studio and other friends who bought the device, this was all they needed and they slowly began falling out of rotation after one or two charge cycles. For the people who did stick with the device, the emphasis switched over to the more social aspects like competing with friends.

After the commercial success of the Fitbit One, there was a flood of new fitness trackers. One big lesson everyone seemed to learn was that the pager-style form factor of the Fitbit One was inconvenient, and so it was decided to move the pedometers to the wrist, which might be the worst place on your body to measure your steps. The primary emphasis of steps, (specifically, Hatano’s 10,000 steps goal) never went away. A major highlight of this first group of devices was Nike’s FuelBand (Figure 2-5). The FuelBand was aimed at more athletic individuals and still tracked the dubious life-based metrics like steps and an estimated calorie burn, but the interesting part of the FuelBand ecosystem was how much emphasis that Nike placed on Fuel Points (Figure 2-6), a proprietary unit of measurement that attempted to describe overall energy expenditure. Nike’s Fuel Point system was an attempt to even out the inherent flaws of using the step count as the metric that activity goals are based on, as opposed to using step count and then displaying calorie burn as a secondary metric.

Figure 2-5. Nike Fuelband

Figure 2-6. Nike FuelBand application screenshots

The problem with basing your activity goal on overall steps is that all steps are not created equal. 10,000 steps for someone who weighs 150 pounds is very different in terms of energy expenditure than it is for someone who weighs 250 pounds, as is 100 steps down hill as opposed to 100 steps up a steep incline, or the difference between running as opposed to casual walking. These are all factors that overall steps don’t account for, but there’s also the issue of vigorous athletic activity that must be measured by an accelerometer and gyroscope that’s worn on your wrist. I would not want to be on the data team that was tasked with translating the complex movements of someone playing basketball into an accurate step count. Even if they perfected those algorithms, what use would the output be when a leisurely hour-long stroll would produce the same measurement? Most trackers picked up this slack by estimating how many calories one burned, but the difference between the FuelBand and everyone else was that Nike used Fuel Points to integrate energy expenditure into the primary metric that people used to measure their activity.

Each new device in this first group had its own small advancements or distinctions. The Misfit Shine was appealing because it used a watch battery that lasted forever and didn’t look like a rubber band. Fitbit Flex was like the Fitbit One except it was worn on the wrist; the Jawbone UP24 band was designed by Yves Behar and had a better designed interface; and the Withings Activité (Figure 2-7) was really cool because it looked like a normal watch with an analog interface. All of these devices were slight improvements layered on top of the same underlying functionality, a gyroscope and accelerometer that helped you count to 10,000 steps. The canary in the 10,000-step coal mine came from the Chinese firm Xiaomi when it released its flagship Mi Band (Figure 2-8). The Mi Band fits in with this group really well: it tracks steps and sleep with surprising accuracy, lets you compete against friends, has a decent battery life, syncs with a phone, and even vibrates when getting a call. The reason the Mi Band is interesting is because you can get one for about $10 where the other devices in this category cost at least eight times as much. The fact of the matter is that the sensors and underlying reporting have become so cheap that to stay in business, you need to go beyond reporting of basic activity.

Figure 2-7. Withings Activité Pop

Figure 2-8. Xiaomi Mi Band

Now that the new reality of the fitness tracker market has been accepted, there is a new generation which is centered around heart rate. The two leading devices of this new era are the Fitbit Charge HR (Figure 2-9) and the Jawbone UP3. The Fitbit uses a photoplethysmogram (PPG) sensor to measure your heart activity. This sensor tracks optical changes in the light absorption of the skin, which indicates heart activity. Jawbone’s UP3 uses bioimpedance, which sends a small amount of electricity through the surface layer of the skin and gleans information through changes in the resistance to that electric signal. Though these additional sensors are what’s marketed as the primary selling points of these advanced activity trackers, what will define this generation will come from something else altogether—machine learning.

Figure 2-9. Fitbit Charge HR

Before going into the details of the algorithmic capabilities of these devices, I want to talk about the difference that application of data makes in the context of providing a service. In older economic terms, services are things like education, and commodities tangible items like salt. A popular saying among my more service-oriented friends is, “If you can’t drop it on your foot, it’s a service.” Both the Fitbit Charge HR and the Jawbone UP3 fall in the spectrum of commodity to service, but they’re at opposite ends of this spectrum. The Fitbit uses a “cheerleader” model, in which you’ll get periodic updates (mainly on the device itself) that amount to a contextual check-in—something like, “only 3,000 more steps to go!”—and in this way, the Charge HR isn’t all that different from the 10,000-step generation of trackers.

The Jawbone UP3 (Figure 2-10) stands out from nearly every other tracker on the market by positioning itself as a service that is powered by machine learning. Compared to the Fitbit’s cheerleader model, the UP3’s service is a “coach” model in that it records your information, makes sense of it, and, based on this information, will suggest a change in behavior. Instead of “Only this many more steps,” the UP3 will say, “You did this, it means this, and you should do this.” Internally Jawbone refers to this as the “Track, Understand, and Act” model,^[5] in which, through machine learning, the service can make certain assumptions about your activity, and then via its prescriptive, it can tell you what to do based on the understandings of your activity. And that’s where the value comes from: it’s now a service.

Figure 2-10. Jawbone UP3

Figure 2-11 is an example of how the Track, Understand, and Act model might work. After a night of heavy drinking, the three bioindicators that Jawbone could use to understand my behavior are:

Figure 2-11. Jawbone Smart Coach screen

The understand part of the process then concludes, “Scott was drinking last night,” though it doesn’t expose the understand part to me. The only part that I’m actively aware of is the prescription, which is that I should drink a lot of water and go to sleep earlier. When I’m not drinking, a more typical message might compare my steps to my average, or let me know how different aspects of my behavior are related, such as, “You were awake about 38 minutes earlier than usual this morning. You tend to move less after an early rise. For you, 38 minutes means about 542 fewer steps. Start a new trend. Head out and maintain your 12,731 step average.”

Another thing that is interesting about the UP3 is what it can do, yet doesn’t thus far. If you look at the box of the Jawbone UP3, there are three things in the specifications section that aren’t activated in the software at this time: a heart flux sensor, skin temperature sensor, and galvanic skin response. These things add up to what could basically be used as a rudimentary polygraph test and place the UP3 in the realm of cognitive wearables, which we discuss in Chapter 7. What is most exciting about this scenario is how the cognitive data could be combined with activity data to provide more useful information such as, “Your step average is down by about 700 steps this week. For you, this means about a 28% increase in overall anxiety.”

This might sound like a weird statement, but one of the biggest problems with most fitness trackers is actually their emphasis on steps. There are tons of studies on the benefits of walking more—I am in no way saying that it doesn’t contribute to overall health, and it’s definitely not a bad thing to be doing—but if we take a step back and look at the primary reasons people use fitness trackers, how big of a factor in overall health and weight loss are step counts? According to the National Institutes of Health, about a third of adults in the United States are obese, and according to the Center for Disease Control (CDC), the leading cause of death in the United States is heart disease, which is linked to obesity as well as other causes.^[6] Addressing weight and obesity problems comes down to a simple equation known as energy balance, which refers to the relationship between calories consumed over time and calories burned over time; if you consume more than you burn you gain weight, and if you burn more than you consume, you lose weight.

The best illustration of the relationship of diet and exercise to weight control comes from Aaron E. Carroll of the Indiana University School of Medicine. Carroll states that exercise consumes far fewer calories than people think. To drive this point home, he compares an exercise regimen of 30 minutes of jogging or swimming every day to the equivalent in calorie intake reduction of eliminating two 16-ounce sodas a day,^[7] which is a much more realistic lifestyle change for most people. With this in mind, fitness trackers are only telling half the story of our overall health. The emphasis on physical activity is at best incomplete, but it also can be leading people to believe that they’re doing more to affect their health than they actually are, which can result in lack of noticeable progress and ultimately loss of interest in the device. So, the big question is, what can we do to fix this?

There are a few ways in which people try to track calories. The most popular is manual logging using an application like MyFitnessPal, which has a database of common foods that you can select from and makes calorie tracking relatively easy and integrates with a lot of activity tracker software. The problem with manual logging, though, is that it’s not very sustainable. Compared to the passive fitness trackers that we’re used to wearing, manual logging requires a lot of upkeep because it requires you to pull out your phone and punch in data after every meal. Personally, I’ve never been able to actively log food data for longer than a month, and I’ve heard the same thing from many people, even after seeing positive results.

I’ve always thought of passive calorie counting as a holy grail of wearables and something that wouldn’t happen for a while. I remember sketching out fantasy ideas for sensors that could replace our molar teeth that would be able to monitor our food consumption or wearable cameras that would identify the food we were eating and automatically log it. It wasn’t until recently that I decided to give the controversial GoBe device (Figure 2-12) a shot. GoBe had a very rocky start: the device’s manufacturer claimed it could passively and noninvasively measure calorie intake with its wrist-based device, which launched in 2014 on the crowdfunding website Indiegogo, but almost immediately after raising more than a million dollars, critics began to chime in. The criticism was led by PandoDaily’s James Robinson, who wrote extensively about the dubious claims of the device. But what really stood out to me was when medical professionals began weighing in. The big quote for me was from Michelle MacDonald, a clinical dietician at the National Jewish Health hospital in Denver, who said that the technology is plausible, but we’re not there yet, “but when it does it will be the size of a shoebox...It will come from a big lab, will be huge news, and make a lot of money.” This was enough for me and a lot of other people to completely write off the product, I didn’t think about it again for almost two years.

Figure 2-12. Jawbone Smart Coach screen

I was running some concepts for this chapter by someone who works in the wearables industry, and when I mentioned passive calorie tracking as a pipe dream, he told me to take another look at the GoBe and mentioned that researchers at his company have confirmed that it actually works. Fast forward a week or so. I have it on my wrist, it’s massive and clunky looking, and it takes forever to set up on the application. But after about an hour, I had it monitoring my food intake. Here’s how the mostly automatic calorie tracking interface works: a little while after your body begins metabolizing the food, you open the application and are greeted by a screen that asks you when you ate. The application knows you ate something within a certain time span, and you confirm the more specific time by turning on or off switches marked with 15-minute intervals. This is the first big communication issue: the application immediately gives me a reading of my calories when my body has only begun to metabolize the meal, so if I know for a fact that I had a 400-calorie breakfast, but a half-hour after I eat I see that the GoBe only logged it as 100 calories, it seems wrong to me. Holding off on displaying calorie counts until the meal is digested, or just communicating that it takes a while to process would go a long way for initial trust of the application, because we tend to think of food consumption as a discreet meal that happened at that one time, especially if people are used to the manual calorie-tracking application MyFitnessPal, which shows the full calorie count immediately.

The next design flaw in the GoBe application (Figure 2-13) is a little more complicated and took me a while to figure out. I was testing to see if the watch actually worked by meticulously tracking my calories manually and comparing the manual counts to the GoBe. Some big inconsistencies came up. I eat the same meal for breakfast every day, and it’s exactly 400 calories. Some days the GoBe would report my calories at about 424, which is completely acceptable, but other days it would be off by more than 200 calories. The circumstances of my morning didn’t change, but it would consistently be getting the counts wrong, and I couldn’t figure out why, but I had a feeling that it might be in the time intervals that I selected. After a couple days, I realized that if I logged the meal as being consumed within a single 15-minute interval, the device would be surprisingly accurate; if I logged the meal as spanning two 15-minute intervals, it would log almost exactly 200 calories more than I ate. As it turns out, and this was later confirmed by the founder of the company, the time intervals are used not only for time logging, but also inform the calorie counting algorithm. In fact, it seems that the primary reason for the 15-minute based meal logging is to gauge the size of the meal, the logic being that if your meal takes longer, it’s probably a bigger meal, thus higher calorie count. This design flaw gets even worse in higher-calorie meals, if I very quickly eat a 1,200-calorie burger and fries in a single 15-minute interval, and GoBe logs it as 800 calories, there’s no way I’m going to ever trust the accuracy of the device, but if I say it took me a half-hour to eat, it’s almost dead on.

Aside from the two quick fixes of holding off on the calorie reporting until full digestion and just simply asking if it was a big, medium, or smaller meal instead of indirectly asking how long the meal took, there’s also the overall communication of the application that can get in the way. With wearable devices, and specifically health-related wearables, we have this assumption of absolute accuracy, but more often than not they’re just accurate enough to give us an idea of the number of steps we’ve taken or the number of calories we’ve burned or consumed. In the case of the GoBe, the biological indicators it’s working with do not mean the same thing for everyone. The device uses tissue impedance (similar to the Jawbone UP3) and, according to parent company HealBe, can take up to six months for its algorithms to fully learn how your body reacts to food. It simply can’t work perfectly straight out of the box. This initial inaccuracy can present a problem for a $300 device. Is it better to sell the product to people and say, “Just hold on; it’ll work better if you stick with it,” or should they just hope people trust the thing and eventually everything will get better?

Figure 2-13. The GoBe application

Before we get in to the design of fitness trackers, let’s talk about what fitness trackers actually do. Fitness trackers simply digitally encode some form of information that comes from our bodies, and then present that data to us in a meaningful way. More specifically, the fitness tracker records one or more biological indicators (bioindicators), translates that indicator into electrical signals, translates those electrical signals into data that can be stored, converts that data into some metric that is then recorded to some sort of database, and then displays these metrics in a way that possibly makes sense to you. There’s a lot of processing that must happen to get that step count to you at the end of the day, and there’s a lot of calculation and translation from between the bioindicator and data output that affect both the accuracy and communication of the information.

Data to Metrics

The easy part is getting the sensor data. Now that we have our raw movement data streaming from our accelerometer and corrected by our gyroscope, we need to convert that data into something that’s useful to the person wearing the device. Figure 2-14 is my accelerometer data for walking around for about 10 seconds. What information can we assume from this data? Well for one we can see a pretty clear pattern of my steps. Let’s just infer that each one of the higher x axis peaks (bottom line) to the following lower valley is a step because I’m swinging my arms in a repeated pattern, so those are steps (for this data set at least). Given my height, we can set a pretty universal standard for exactly where those peaks should land for walking; given my weight, we can assume how much energy I’m expending to walk and thus infer some sort of calorie expenditure based on that. If I’m running, I’d expect to see the peaks and valleys of this data to be much more different as well as faster. I could then assume from the data a higher rate of caloric burn.

Figure 2-14. Graphed three-axis accelerometer output

There are all kinds of ways to make fitness trackers interesting to users, and now it’s pretty much a necessity if you want to differentiate your tracker from the existing huge field that all basically do the same thing. Your goal could be calorie burn, and that would probably be a lot more indicative of activity than steps alone, so why not have the day start and end when you wake up and go to sleep? Maybe sticking sensors in your ear turns out the be the absolute best place to put a pedometer and heart rate monitor. And what if the entire interface was aural? Throw GPS in there and give people feedback on how boring their day is, based on the fact that they did the exact same thing for the past four days. Trackers based on 10,000 steps have been around for a long time, but 10,000 steps doesn’t mean the same thing for everyone. It’s time to mix it up!