Trying to get a large group of people help you in a task is sometimes called "crowdsourcing." There are several accepted strategies for doing this, but a recent study reported in Science by Galeb Pickard and six colleagues at a M.I.T. media laboratory reveals an approach that seemed to work when they had to complete the task in a hurry. As they point out, gettting crowdsourcing accomplished in a hurry is particularly relevant to such practical common problems as search-and-rescue operations, tracking an outlaw on the run, managing public health responses in a disease outbreak, and even promoting the candidacy of a political candidate.
The spur for this research was a contest sponsored by the Defense Advanced Projects Research Projects (DARPA). No explanation was given for the military advantage of such a research project, but DARPA put up prize money for a nationwide contest that was like a scavenger-hunt. The contest challenge was to be the first research group in the country to find 10 red weather balloons that DARPA had hidden acrosss the country and report the coordinates of the balloon locations.The M.I.T. group chose to use the Internet, particularly Twitter, as a way to mobilize large numbers of people nationwide to help them be the first to find all 10 balloons. The Internet is the obvious way for large numbers of widely dispersed "hunters" to collaborate in reporting their findings, and Twitter provides a condensed communication environment for recruiting helpers. The research strategy challenge was to find a way to incentivize strangers to participate enthusiastically and to share their findings.
The M.I.T. group won the challenge in a little under 9 hours, after they were given time to recruit volunteers via Twitter, which netted 4400 helpers in a little over 36 hours. The M.I.T. collective found all 10 balloons, while the runner-up Georgia Tech group found nine. Two other teams found eight. Over 50 such teams competed, with each using their own Internet-based strategy.
The key to the M.I.T. success was in the stucture of their incentive mechanism. The incentive was the $40,000 prize money, which the researchers pledged to allocate at the rate of $4,000 for each located balloon. Many people who were involved in finding a reporting a balloon got a portion of the prize. $2,000 went to the first person to send in the correct location of a balloon, $1,000 to the person who invited that person to the scavenger hunt team, $500 to the inviter of the inviter, $250 to the person who invited that person, and so on until the budget allocation was reached. Thus, it is seen that rewards went to several people in the success chain, from inviters to the person who finally made the discovey, and everybody could think they had fairly good odds of getting something. The design is one-to-a few to a few more to still more, fanning out into a large network of incentivized collaborators. This mechanism was presumably efficient enough to accomplish the task faster than could be done with other approaches.
Diagram of crowd (or cloud) sourcing and recursive incentivization, showing how people (indicated in black circles) are recruited via Twitter into a success chain where rewards are shared and scaled within that chain (1x is the prize money given to the grand winner, with others recruited within that chain sharing in the budgeted funds in scaled fashion). Other participants (white circles) did not share in the prize because the people they recruited (dashed chains, left incomplete to simplify the drawing) did not help recruit the winner. Yet, even here, the "losers" knew they had a chance to share in prizes and presumably worked hard for that reason.
The researchers point out that this scheme is not entirely original, apparently pre-existing in the form of a known network query model. But what was different here was the huge size of the network tree and the fact that incentives scaled with the size of the tree. "Recursiveness" refers to the fact that the incentive process was repeated in a "self-similar" way, which is math-speak for individual steps or objects having the same properties, but scaled (as in fractals and Mandelbrot sets).
The authors gave a brief discussion of strategies used by other teams who failed to win, yet did rather well in finding balloons. One good strategy involved using a handful of "Twitter celebrities" with a large existing following to get the word out. A large group of helpers signed up to such teams, but the recruiting was not sustained, presumably because of lack of incentive to recruit more helpers. Some teams operated on an altruistic philosophy, with prize money promised to charity. Notably, they did not win.
Not much psychological speculation was presented in the paper. However, it is obvious that selfish interests are strong motivators. Also, a sense of fairness may be involved. People who help create the $2,000 winners for each balloon might feel cheated if they got no reward for helping. Knowing how the game was structured made them aware that they had a chance to win something.
The popular TV show, "Survivor Island," is a case study of the flaws of winner-take-all philosophy. Here, to win the $1 million prize, you have to get everybody voted out of the game. Alliances and cooperation are temporary. At some point, collaborators in a clique start to betray loyalties because those not yet voted "off the island" have become serious competitors. There is no benefit for coming in second or third, etc.
Source: Picard, G. et al. (2011). Time-critical social mobilization. Science 334: 509-512.
