The Social Media Tag Challenge: Crowdscanner describes how they won

Post by: Kim Stephens

On March 31st, the US State Department sponsored a game called  “Tag Challenge” that took social media monitoring to a new level.  It was designed by graduate students from six countries, “…the result of a series of conferences on social media and transatlantic security.”

They constructed a task that would be impossible for one person to complete: find 5 “jewel thieves”  in 5 cities across the globe in one day, photograph them, and upload the image.  The winning team, an MIT affiliated group which dubbed themselves “Crowdscanner,” was only able to find 3 of the 5 individuals, however, much was learned about how loosely connected distributed networks can be incentivized to solve a problem.

“The project demonstrates the international reach of social media and its potential for cross-border cooperation,” said project organizer Joshua deLara. “Here’s a remarkable fact: a team organized by individuals in the U.S., the U.K and the United Arab Emirates was able to locate an individual in Slovakia in under eight hours based only on a photograph.”

I had the pleasure of interviewing one of the Crowdscanner team leaders, Dr. Manuel Cebrian of the University of California, San Diego (who also led a team that won the DARPA Red Balloon Challenge in 2009). What stood out to me from our conversation was his emphasis on their incentive structure versus the social media tools. The networking tools were simply the means to the end, but the structure of the reward incentive, which was born out by strong micro-economic theory, was absolutely fundamental to their success.

Another interesting component to the challenge was the interaction between the competing teams, which I found in background information provided by Dr. Cebrian.  Some rival teams actually attacked Crowdscanner on twitter with tweets questioning their competence and encouraging people not to support them. As the challenge period came to a close, these attacks became increasingly desperate–even mentioning that Crowdscanner was not from DC and therefore shouldn’t win. That team emphasized that they were “playing for charity,” which the Crowdscanner team noted “…even though it was clearly not in line with their vitriolic attitude towards us.”

How this competing team used twitter to find information also provides a lesson:

[The other team’s] strategy for spreading awareness consisted of their Twitter account… surfing trending hashtags, and tweet-spamming many individuals, social, governmental and private organizations in the target cities, often with an explicit plea for a retweet. The vast majority of these were ignored and, we believe, reduced their credibility.

Q: What does this challenge tell us about incentives and social mobilization? 

We used an incentive scheme that is designed to encourage two things simultaneously: (1) reporting to us if you found a target; (2) helping recruit other people to search for the target. Here’s how we described it: If we win, you will receive $500 if you upload an image of a suspect that is accepted by the challenge organizers. If a friend you invited using your individualized referral link uploads an acceptable image of a suspect, YOU also get $100. Furthermore, recruiters of the first 2000 recruits who signed up by referral get $1 for each recruit they refer to sign up with us (using the individualized referral link). See their webpage for more info on the design.

Graphic by Crowdscanner

The incentive to refer others is significant, since otherwise, you would actually rather keep the information to yourself, rather than inform your friends, since they would essentially compete with you over the prize. But by paying you for referring them also, the incentives change fundamentally.

Q: What tools were you using to monitor twitter?

Monitoring twitter was the smallest component. In fact,  monitoring  was the easy part, since the data is there to be sorted and analyzed. The biggest challenge was finding the non-twitter data: we had to infer how information was spread.

Q: Why did you all succeed?

We were able to succeed by leveraging a combination of social media and traditional media, and by building up a reputation as a credible, reliable team. Some competitors focused purely on social media, almost using Twitter exclusively to spread their message. This is not enough, as they became perceived as spammers. We were more selective in our Tweets and social media strategy, and I believe this gave us an edge.

Q: Do you think this model could work for finding real “jewel thieves” or high target terrorism suspects? 

Ransoms are complicated incentives. With traditional ransoms, once you have the information you have no incentive to recruit people to help you. Why would you team up?  So the question becomes, how can you structure it so that people are not greedy? We used the same incentive structure for the balloon challenge. These micro-economic models [and the way we employed them] demonstrate that people do recruit their friends, but only if they are provided the right incentive.  If you spread the word, then you get the money.

Q: So, why aren’t organizations using this distributed network model?

Centralized systems are inefficient but they are predictable. In a distributed system you have high efficiency but also have high unpredictability.

Gathering evidence is easy, doing justice is hard. We need to have models that make sense of the data. But currently,  we don’t have this kind of training. It is a new science: “network science” at most, a 10 year-old discipline, and only a few people that can make sense of it. It will take a while for us to be able to use these tools in any concerted way.

Related articles

• Crowdscanner: Viral social media challenges can be solved ‘within the hour’ (zdnet.com)
• Humanitarian Response in the Age of Collaboration and Networked Intelligence, by Gislio Olafasson http://dl.dropbox.com/u/13821490/ISCRAM2012-Humanitarian%20Response%20in%20the%20Age%20of%20Mass%20Collaboration%20and%20Networked%20Intelligence-V2.pdf