The Twitch crowdsourcing app in action.

Mar 25 2014

The Story Behind the Twitch App, Stanford’s Crowdsourcing Project

What can smartphones tell us about emotion? Stanford and UC Santa Cruz aim to find out.

What if, instead of merely swiping your thumb across your smartphone's screen or punching in a code, your phone prompted you to fact-check a Wikipedia article or input how many people were near you?

There's plenty of excitement over the opportunities presented by crowdsourcing, but many of the endeavors to use collective brainpower have significant hurdles to overcome. But in comes Twitch, an Android app designed by a team of students and scientists from Stanford University and UC Santa Cruz.

"Not everybody [uses] it, because of time constraints and time commitments," said Rajan Vaish, a visiting researcher at Stanford and a computer science doctoral candidate at UC Santa Cruz.

That problem — how to simplify the crowdsourcing process on an accessible platform — is what sparked the idea that evolved into Twitch.

Vaish and the team, advised by Michael Bernstein, an assistant professor of computer science at Stanford, hit upon a simple action many of us repeat with startling regularity: swiping a smartphone to kick it out of its idle state.

"We're trying to harness that, leveraging people's short bursts of time." Vaish said.

How Twitch Works

Instead of going immediately to the home screen, Twitch users are given a brief prompt to verify a statement's truthfulness, single out a photo from a group from others or provide some information about their surroundings so it can be tied to the users’ current location.

On a campus like UC Santa Cruz, Twitch can gather data to provide professors and administrators with information about student energy levels and dispositions throughout the day and across the campus.

If everyone's grouchy in the library at 10 p.m. on Sundays, it may be indicative of the downside of weekend procrastination — or there could be something else at play. That’s the sort of data Twitch can provide; think real-time traffic reports or weather forecasts.

And using Twitch remains speedy. Unlocking a phone typically takes 1.4 seconds, while responses to Twitch's prompts range between 1.6 and 3 seconds, depending on how much text the prompt contains. Not a terrible trade-off to participate in a global crowdsourcing project.

After an initial trial phase that included 82 people and lasted a little less than a month, the team got plenty of feedback. On the plus side, some of those participants are still using the app, even though their role in the development has ceased. And so far, Twitch has prompted more than 70,000 responses worldwide, with users stretching all the way to parts of Europe and India.

But there are still improvements to be made. A common issue that users have had with Twitch is insufficient motivation. Twitch version 2, currently in the works, will include elements to directly address that issue, Vaish said.

For example, a web application is being built to allow users to see the progress being made because of their contributions. Seeing how those small batches of time combine into, say, an entire article being fact-checked could be the key to retaining users beyond the expiration date of the app's novelty.

With plenty of work still to be done, Vaish isn't looking much beyond version 2, but he is looking forward to eventually making the app's source code available to the public. After being approached by entrepreneurs and researchers about Twitch's possible applications to their respective fields, Vaish said he's excited to see what else the public could build off of the team's work.

<p>Standford HCI Group</p>