The Next Frontier in Crowdsourcing: Your Smartphone
Researchers have shown that the swipe-to-unlock action can be replaced with useful data collection or micro-tasks.
Crowdsourcing can be the best way to collect useful data or distribute small tasks.
Rather than swiping the screen or entering a passcode to unlock the smartphone in my hand, I have to tell it how energetic the people around me are feeling by tapping one of four icons. I’m the only one here, and the one that best fits my actual energy level, to be honest, is a figure lying down and emitting a trail of z’s.
I’m trying out an Android app called Twitch. Created by Stanford researchers, it asks you to complete a few simple tasks—contributing information, as with the reported energy levels, or performing simple tasks like ranking images or structuring data extracted from Wikipedia pages—each time you unlock your phone. The information collected by apps like Twitch could be useful to academics, market researchers, or local businesses. Such software could also provide a low-cost way to perform useful work that can easily be broken up into pieces and fed to millions of devices.
Twitch is one of several projects exploring crowdsourcing via the lock screen. Plenty of people already contribute freely to crowdsourcing websites like Wikipedia and Quora or paid services like Amazon’s Mechanical Turk, and the sustained popularity of traffic app Waze shows that people are willing to contribute to a common cause from their handsets if it provides a timely, helpful result.
There are certainly enough smartphones with lock screens ready to be harnessed. According to data from market researcher comScore, 160 million people in the U.S.—or 67 percent of cell phone users—have smartphones, and nearly 52 percent of these run Google’s Android OS, which allows apps like Twitch to replace the standard lock screen.
Michael Bernstein, an assistant professor at Stanford working on Twitch, sees this kind of bit-by-bit mass data collection—while you’re waiting for an elevator or sitting in a boring meeting—as a way to get around the time requirements of some current crowdsourcing efforts. “Many people wish they could help but simply don’t make it a priority,” he says.
In a study conducted by the Stanford researchers, 82 Twitch users completed 19 tasks per day over a three-week period. Participants were asked how many people were nearby, how they were dressed, and how energetic or lethargic they were—a novel mobile measure of public activity. The researchers found that the tasks weren’t more time-consuming or distracting than the basic slide-to-unlock gesture—the median time to complete each task was 1.6 seconds, while the unlock gesture was 1.4 seconds.
“All we did was replace that gesture that makes sure you’re paying attention with something else that makes sure you’re paying attention and also happens to contribute to some global goal,” Bernstein says.
Those who participated were not offered any payment for their efforts, but the app told them how many people nearby chose the same answer as they did. That may not be enough of an incentive to persuade people to participate, though. A little more than half of the study participants used the app, on average, for 32 days, but nearly half of the study’s participants uninstalled the app within a day.
Researchers at the University of Toronto developed a similar app but tried paying people to use it.
Researchers gave 10 study participants $20 apiece to spend two weeks using an Android app called Slide to X that presented them with one of several different tasks to perform to unlock the phone: a standard swipe, a multiple-choice question about the user’s health or recent activities, or an easy math question designed to determine whether the user was actually providing thoughtful answers or just tapping on one at random. The app also collected data like the number of times a user unlocked his phone, and where he was and what time it was when he unlocked it.
Paper coauthor Khai Truong, then an associate professor at the University of Toronto who studies ways to make mobile computing more useful, says people in the study answered an average of 50 questions per day apiece, or 772 total over the full two weeks, and that they did seem to be engaging with the questions rather than absentmindedly tapping the screen. The compensation worked out to a bit less than 3 cents per question, and all but one participant said they’d be willing to continue the study for the same amount of time and compensation. “It’s a good indication that this kind of interface doesn’t require a lot of financial compensation for people to want to use it,” Truong says.
Truong, now an associate professor at UNC Charlotte, is working on an application that anyone can use for simple, nonprofit data collection by asking multiple-choice questions (although rather than paying users, he imagines those using the app to gather data will make donations to a user’s charity of choice). He says this could be especially useful for gathering information that can’t be picked up while sitting in front of a computer—such as how noisy different parts of a city are at different times of day.
Bernstein and the Twitch group are also prepping a new version of their app that asks users to help structure the Web by checking the accuracy of facts posed by University of Washington-developed information extraction software Reverb’s analysis of Wikipedia text. The app is expected to be available in a month or two.
AI is here. Will you lead or follow?
Join us at EmTech Digital 2019.