In 2009, the American Agency DARPA launched a competition to assess whether social networks and crowdsourcing could facilitate the identification of ten huge red balloons moored at different locations across America. All ten balloons were identified by the winning team within eight hours, fifty-two minutes and forty-one seconds.
When DARPA and all interested parties had recovered from the sheer surprise at the swiftness with which the competition had been won, most social media experts incorrectly assumed that the competition had been won through the superior number of participants that the winning team had managed to recruit relative to the other teams. The capability of social networks to draw large numbers of people had ensured that the balloons were identified in a few hours, so it was assumed. But subsequent announcements by DARPA underscored that winning was more brain than downright brawn or numbers. There was, in fact, a remarkable method in winning.
In all, DARPA reported, between fifty and a hundred ‘serious teams’ actively participated, from a total of four-thousand teams that had shown interest. It was also estimated that around 350,000 people had participated on competition day. The competition’s results were declared as:
|Place||Team Name||No. of Balloons||Time|
|1||MIT Red Balloon Challenge Team||10||6:52:41pm|
|2||GTRI “I Spy a Red Balloon” Team||9||6:59:11pm|
|3||Christian Rodriguez and Tara Chang (Red Balloon Race)||8||6:52:54pm|
|4||Dude It’s a Balloon||8||7:42:41pm|
|6||Army of Eyes Mutual Mobile||7||4:33:20pm|
|10||iSchools DARPA Challenge Team||6||6:13:08pm|
Two months after the competition, DARPA organised a Computer-Supported Cooperative Work Conference1 in Savannah, Georgia, to discuss the outcome and insights into the competition.
The conference brought together three of the competing teams; the winning team from the Massachusetts Institute of Technology (MIT), the second-ranked team from Georgia Tech Research Institute (GTRI) and the tenth-placed team from iSchools; an alliance of five different American universities (Pennsylvania State University, University of Illinois at Urbana-Champaign, University of Pittsburgh, Syracuse University, and University of North Carolina at Chapel Hill).
The objective of the conference was to compare and contrast the three different strategies in order to discern what could have led to such rapid location of all ten balloons.
THE TENTH-PLACED TEAM
The iSchools team had utilised a number of different strategies. The five universities forming the team collated various student and alumni mailing lists and social media groups and invited each person on these lists and groups to serve as an observer out of loyalty towards the respective university. However, iSchools reported that few invitees bothered to participate and only one balloon was identified through this strategy.
The second strategy bordered on outright eavesdropping and was made up of two distinct sub-strategies. The first sub-strategy was based on the discernible assumption that all competing teams would be using social networks to track the balloons. Consequently, iSchools set up a team of ‘human analysts’ to monitor various information channels (e.g. Facebook, Twitter and websites) that the other competing teams may be using to register sightings.
The human analysts would ‘overhear’ any reported sighting and rapidly evaluate the validity of the sighting, largely based on the preceding reputation of the reporting source. However, a second level of confirmation was pursued through the recruited team members who may have been positioned close to the sighting. When a reported sighting was not located close to a known team member, new observers from organisations located within the vicinity of the sighting were rapidly invited to confirm the report.
It is impressive how most of the other teams were initially not equipped to safeguard the information they were gathering, particularly across an exposed medium such as cyberspace. However, eventually, most teams became aware of this eavesdropping over the day and even started to post deliberate false reports of sightings to throw the competition off any potential scent.
The GTRI team’s strategy was largely based on the altruism of participating members in combination with wide-scale Internet promotion across social media.
Three weeks before the start of the competition, the team set up a number of online marketing initiatives, including a dedicated website and a Facebook group. The objective was to create a higher degree of visibility than the competition so as to increase the probability that during the competition anyone seeing a balloon would report the sightings to GTRI.
In marketing terms, the GTRI team were seeking to nurture ‘brand association’ by creating a deep-seated notion in people’s mind about who to contact upon spotting an eight-foot, bright red, weather balloon. This included solid search engine rank optimisation so that the team’s website would always appear relatively higher in any search related to the competition.
Brand association is often deliberately linked to some positive aspect so that consumers relate the specific brand with an upbeat viewpoint. In this respect, GTRI announced that they would donate any competition winnings to charity, so any reported sighting would have a motivating benevolent cause. They also assumed that this charitable perspective would also reduce false submissions. However, the validity of this hypothesis has never been confirmed.
Unfortunately, none of these marketing strategies appear to have created significant motivation, making it all the more admirable that the GTRI team, with just 1,400 registered participants, ended in second place, just one sighting short of the winning team.
THE WINNING STRATEGY
The winning MIT team admitted that they had learned about the competition just a few days before the balloons’ deployment, much as DARPA had aspired. Consequently, the team had taken an erudite decision to exploit a strategy that balanced identification speed (gained through having as many active participants as possible) against breadth (having the same participants distributed across the widest possible coverage of the United States’ topography). This balanced approach contrasts sharply with the use of, say, a mailing list of a specific institution (e.g. a University), even if the mailing list contains hundreds, possibly thousands of names. Such a list would very likely contain clusters of names belonging to potential participants who live within a relatively limited geographic area in the vicinity of the institution and lack critical wide-area spread.
The team also recognised that merely having an extensive list of participants was not sufficient per se. The likelihood of success was highly dependent upon having each and every participant being motivated and incentivised towards proactive participation. Such an objective could only be achieved through the one essential factor that possesses universal appeal; financial gain.
Shortly before the launch of the balloons, the MIT Red Balloon Challenge Team’s website announced that if successful, the $40,000 prize money would be used to reward participants:
“We are giving $2,000 per balloon to the first person to send us the correct coordinates, but that’s not all – we are also giving $1,000 to the person who invited them. Then we are giving $500 to whoever invited the inviter, and $250 to whoever invited them, and so on.”
This multi-level marketing technique (aka referral or pyramid marketing) was intended to incentivise the recruitment of a widespread chain of participants, each of whom aspired to win the prize money, but concurrently, themselves contributing other participants. Moreover, each new recruit was less of a direct competitor for the prize money than a ‘cooperating partner’ in a mutual win-win scenario.
MIT recognised that it was necessary to nurture two distinct types of incentives within the team; one dedicated upon achieving the set objective and yet another focused towards recruiting other team members. Given this distinction between goal and recruitment incentive, it should not come as a surprise to note how MIT even recruited participants from outside the United States, who whilst not directly contributing towards the spotting of the balloons, provided essential social media recruitment to their respective circle of friends within the country.
In scientific terms, this recursive strategy is an inverse variation of Kleinberg and Raghavan’s Query Incentive Network; combining the promise of individual reward (across many potential levels) with the unquestionable power of social networks to connect people.
The striking success of this technique can be gauged from the fact that just thirty-six before the launch of the balloons, the MIT team consisted of just four initial names, but by December 5th this number had expanded to an impressive 5,000 participants.
Much like the GTRI team, MIT recognised the need to appeal to the altruism of some participants and announced that any money remaining at the end of the competition would go towards charity. In a digital sense, the MIT team was ingeniously serving both an altruistic God and a reward-seeking mammon.
The hasty, last-minute, formation of a team though, had its downside; MIT were relatively less equipped to handle the significant (often deliberate) misinformation and misdirection that characterised the competition. The MIT team recorded more than two-hundred distinct submissions of balloon sightings, but only between thirty and forty were, in fact, genuine sightings. In contrast to most of the other teams, MIT did not employ a network of ‘trusted humans’ to validate each reported balloon sighting and confirmations were endorsed or rejected using a rather basic, common-sense, approach.
A balloon sighting would, first and foremost, only be considered if multiple similar reports were received. Moreover, given that the balloons were all tethered in public spaces, each reported sighting was expected to differ, even slightly, from any other. The team reasoned that a genuine sighting would inevitably carry a degree, no matter how trivial, of human interpretation, (e.g. actual address, description of location, nearby landmarks, etc.) whereas false reports tended to be highly consistent in their overall description.
Moreover, the team also adopted a strategy of associating the Internet Protocol (IP) address of every submission with the reported location of the balloon. A balloon in Texas, for example, would reasonably not be reported by someone using a Wisconsin IP address and this approach filtered out a high degree of false submissions.
The DARPA Red Balloon Challenge provides quite a number of interesting insights – some patently evident, others less so – yet all holding universal business application. A 2011 article in the journal Communications of the Association for Computing Machinery2 declared that:
‘The Challenge clearly demonstrated the variety, efficiency, and effectiveness of crowd sourcing solutions to a distributed, geo-located, time-urgent problem. The network mobilisation time was far faster than expected by DARPA program managers, requiring days instead of weeks.’
Undoubtedly, the challenge underscored the potential of social networks and crowdsourcing to harness the collective capabilities of large groups of people to gather and share information at a rate and scale that has been unparalleled by conventional mass media. A third recognised measure, accuracy, is, in hindsight, debatable, given the widespread misinformation and information distortion prevalent during the competition.
Notwithstanding the (sometimes deliberate) propagation of disinformation across social media (popularly known as ‘fake news’), such channels are undoubtedly a fertile and flexible source of reliable and rapid crowdsourced information. One need only contrast how the MIT and GTRI teams used social media for rapid communication between their respective participants, whilst the iSchools team merely eavesdropped and used the channel as a ‘data mining source.’ Fundamentally, the real skill lies in discerning which information to trust, and which to discard.
However, the success of social media does not lie in technology alone; the challenge highlighted the need for participant incentive when creating time-critical social mobilisation tasks, and the need that this incentive spans the spectrum from altruism on one side right across outright financial gain at the other end.
So, when people ask us if social media is just hype, and having narrated the story of the ten balloons, we generally seek to contextualise the account within a local context: “…if it takes less than nine hours for ten balloons to be found across the 9.834 million km² of the North American continent, what would it take for any news to travel across the Maltese Islands’ 316 km²? And what if that news affects your organisation’s reputation? Hype or not; wouldn’t you rather be on such networks to be made aware and mitigate any negative outcomes?”
Oh, and as with all good stories, everyone lived happily ever after, except, for the ten red balloons, presumably, which collectively only managed to enjoy the national limelight for seven relatively brief hours.
Hadrian J Sammut is the Chief Officer Advisory and Projects with the local firm of iMovo Limited. He can be contacted on [email protected].