Do Navigation Apps Think We’re Stupid?
As a hamburger enthusiast, I often need directions to some burger joint I’ve never tried. Recently, my phone’s instructions sent me toward the on-ramp for the interstate. Then the app urged me, in 500 feet, to merge onto the freeway. By that time, though, what else could I have done? Did the app imagine that I might get confused, and turn around instead?
Mapping software is incredible. Having instant access to every storefront, building, park, and transit stop on every street, almost anywhere in the world, has changed my life as much as any other single innovation of the cellphone age. But also, mapping software is a little weird. Seemingly random places show up as landmarks in my neighborhood: a Bitcoin ATM, a nearby hotel I’ll never stay at. And when I need directions, my app likes to tell me things that no one ever needs to know, such as when to merge onto the freeway from an on-ramp. Why is it so obtuse? Or perhaps the better question is: What makes the software think that I’m obtuse?
Simply put, the maps don’t see the world the way the people who use them do. In the data that underlie a digital map, a road network is represented as a bunch of lines. Those lines have a beginning and an end. Seth Spielman, a geographer who worked for a time as a data scientist on Apple Maps, explained to me that a driver often gets instructions from the app at transition points between those segments. When I turn onto the ramp, then merge onto the freeway, I’ve driven through a pair of segments—and from the map’s perspective, I am thus in need of extra guidance. But I don’t feel that need at all. From my perspective, just a single phrase—Get on the freeway—would suffice.
[Read: The existentialism of GPS]
That mismatch of advice is a problem that digital maps have created for themselves. If you started driving before the age of GPS-enabled, app-driven smartphones, you may remember what a traditional road map looked like: lines crossing other lines. A freeway ramp or cloverleaf might be shown in abstract. You’d see how roads connected, and then you’d navigate them on your own.
Maps are always simplifications. But now they simplify a lot less than they used to. “The way the real world is digitally represented creates all of these trivial intersections,” Spielman said. That explains why a digital map might instruct you to “continue” down a straight road: If the street name changes, then, from the map’s perspective, you’ve just exited one road and entered another. Don’t do anything, your onboard flight computer says. Stay on track by going forward.
The odds of your getting these unhelpful tips goes up in concert with the maps’ precision. Apple and Google have surveyed the world in more granular detail than has ever been produced in human history. Camera-topped cars—and sometimes bikes or pedestrians—have captured views of countless streets. Individual buildings, road lanes, and turn signals, along with bike lanes, park paths, and transit lines, are included in the data sets. Apple Maps displays detailed facades of landmarks like Radio City Music Hall. It shows the actual sizes and location of trees in some cities.
All those data points make the mapping apps delightful, even when you aren’t using them for navigation. But their sheer exhaustiveness has a downside: It leads to what might be called map-splaining. Spielman showed me a satellite image of the intersection of Arapahoe Avenue and 28th Street, in Boulder, Colorado. It’s just a normal intersection of two thoroughfares. In the old days, a map would have depicted it as two lines intersecting; a driver who arrived there would surely not have been confused. But Apple and Google have collected enough data to represent this junction in all of its constituent parts.
The maps know that one road is five lanes wide and the other six; both have medians. They understand that right turns between the streets can be accomplished via dedicated merge lanes that skip the red light. They appreciate that two lanes allow left turns between each of these streets, facilitated by a left-turn-arrow traffic signal. Having all this information helps the maps give their step-by-step instructions: Take the first turn lane from northbound 28th Street, then a quick right into the parking lot for Flatiron Coffee. That level of precision may be convenient for some drivers, but it comes at the price of breaking down the built environment into lots of extra segments and transitions that may trigger the display of useless routing information. Perhaps the software should just be telling you to “go past the light and make a left.”
[Read: How Google builds its maps]
Apple Maps has tried to make its guidance feel more natural, in part by using common, human-sounding phrases. For example: “Go past the light and make a left.” This language is intended to replace now-familiar and robotic phrasings such as In 300 yards, turn left. Google Maps is also trying not to be so tortuous or wordy. The software breaks down each route into multiple maneuvers, David Cronin, a senior director on the Google Maps design team, told me. Then it decides which and how many maneuvers a driver or pedestrian needs, how to describe those maneuvers, and what sort of visual and auditory information would best illustrate them. The goal, Cronin said, is to “provide clear and unambiguous instructions without being too verbose.”
To achieve that goal, map designers must sometimes intervene and tell the software to ignore portions of its data set. “We recently made a change that stops giving people directions when they need to proceed straight through a traffic circle,” Cronin said. In general, though, map-direction algorithms are made to be as broadly applicable as possible. Apple handles route instructions differently for urban versus rural roads, and for highways versus local streets, but its overall approach is broadly similar across its 30 countries and regions. Google does a bit of place-by-place fine-tuning, Cronin said; “there are always tensions to respond to.” In India and Southeast Asia, for example, Google Maps provides different routes for two-wheeled vehicles, given their ability to traverse narrower streets than cars.
The data that allow the mapping apps to be so powerful, if also sometimes wonky, are constantly in flux. Google makes 50 million edits to its map per day, according to Cronin, adjusting details such as how roads are classified, where they join, which are closed due to construction, and so forth. All those changes may affect the quality of the apps’ instructions, and their propensity to map-splain, in ways that the designers cannot necessarily predict.
They also shape which points of interest will appear on maps. Both Apple and Google will try to show you businesses that are relevant to your current location. These may, at times, seem pretty random: a Lululemon, for example, or a barbeque place. The apps rely on popularity in deciding what to surface—they keep track of all the spots users tap on or route to most often. Spielman told me that, at one point, this criterion caused Apple Maps to show an excess of pizzerias and Chinese takeout restaurants by default, because so many people were tapping on them to order food.
Google, which knows where you live if you give it a home address, might show different points of interest—hotels, perhaps—if you’re looking at the map of someplace far away. Apple avoids this use of people’s data, making its results more private but also more uniform. Both companies make use of information about how people (or at least their smartphones) traverse space to inform their guidance. These data might be used to evaluate current traffic conditions, for example. Spielman suggested that if a jogger ran across a given street, Apple Maps might be nudged to suggest that crossing at that intersection is more efficient than doing so in other places. Likewise, if someone tapped absentmindedly on a bunch of different bars while waiting for an Uber, those bars might start popping up for other people, on the theory that they’re popular.
Popularity also has a way of building on itself. Spielman told me that tech companies sometimes buy or scrape data to get business locations. Data for chains, such as big-box stores and fast-food restaurants, tend to be easier to find and more standardized than information for smaller businesses, giving the chains a boost on maps. Cronin disputed this account. “Our aim is to create a digital representation of the real world, and that real world includes a range of businesses and places,” he said, adding that local proprietors and other people can add places to the map. Apple also allows businesses to submit their information to its map. But once a destination has become a point of interest, people may be more inclined to get directions to it, reinforcing its position. Google also puts sponsored points of interest on maps. Cronin explained that those are marked differently—with a rounded square instead of a round pin—but I hadn’t noticed that distinction until he pointed it out.
[Read: Would you drive an extra five minutes to save the planet?]
The growth and spread of mapping data may have some other, occult effects. Cronin said that Google Maps improves people’s confidence in moving about the world. But Sara Fabrikant, a geographer at the University of Zurich, told me that this very confidence may be undermining humans’ ability to self-orient. When the system fails—say, if your phone dies or you otherwise can’t get a signal—the effects of getting “lost” are graver than they were before: It leads to confusion and delay, she told me, and eventually the loss of confidence in one’s capacity to navigate.
The technology companies hope that any social or cognitive downsides of mapping apps could be remedied by better features in the apps themselves. Cronin acknowledged that the maps may inhibit people from exploring, and in that way learning more about the world around them. But he said that new technologies, such as an augmented-reality street view with superimposed walking directions, could encourage pedestrians to way-find in the actual world, looking at their phone for guidance only when they need to. Google is also testing the idea of showing detailed previews of the end of a route, so drivers can work out ahead of time where they might look for parking, for example. Cronin suggested that this approach might support the skill of spatial planning. Apple, meanwhile, hopes that calling out waypoints, showing a user which way to go, and teaching them how to do it counts as its own form of geographical education.
But new features may just as well encourage more complacency. “I think most people are just conditioned by the apps and accept how they work and thus don’t complain,” Spielman said. Because, on the whole, what is there to complain about? Mapping apps and the turn-by-turn instructions they provide are fantastic, and their quirks are easily forgotten. After spending so many years being told to merge onto a freeway when, as a driver, I could do literally nothing else, I’d eventually stopped hearing it. Map-splaining is just another part of driving, hiding in the background. Now I’m at the stoplight for the freeway on-ramp; now I’m turning left; now I’m getting on the freeway; now I’m on the freeway. Me and my map app, there’s nowhere we can’t go.