Over just the last few years, the wait times in between assigning a task to the crowd and a cloud laborer actually getting started on it have shrunk from hours to minutes to seconds. Now research from MIT and Adobe's Advanced Technology Labs claims to have reduced the wait to milliseconds, creating for all practical purposes, a real-time crowd on demand.

In a paper published in the proceedings of the recent Collective Intelligence 2012 conference, MIT's Michael S. Bernstein, David R. Karger, Robert C. Miller and Adobe's Joel Brandt built on their previous research in which they used Amazon's Mechanical Turk to create on-demand synchronous crowds working on a task. To create the desired crowd for those earlier experiments, they introduced the retainer model of recruitment "which pays workers a small amount to come quickly when asked, and a set of empirically derived design guidelines for its use."

The team managed to refine the system to be able to reduce the desired crowd's response time to as little as two seconds. This brief video explains a little more of how it works:

 That brings me to the team's new research, which optimizes the creation of on-demand crowds to occur essentially in real-time -- as little as 500 milliseconds. They did so by proposing three improvements to the retainer model:

 First, retainer subscriptions allow workers to sign up for push notifications for recruitment, which reduces the length of time it takes to recruit new workers onto retainer. Second, combining retainer pools across requesters leads to both cost and wait time improvements. Large retainer pools can then be made more effective by using task routing to connect appropriate workers to the tasks that need them. Third, a precruitment strategy recalls workers from retainer a few moments before a task is expected to arrive, dramatically lowering response time. 

 The full process and complicated algorithms underlying all the above work are laid out in the team's full paper from CI 2012. 

In addition to breaking through the previous two-second barrier, the team claims their optimization will also minimize costs and the retainer model will make workers more engaged and have to spend less time looking for work. 

It will be interesting to see if the use of real-time crowds catches on, opening up new avenues for startups and smaller operations to compete on-demand.

 - Eric Mack is a contributing editor for Crowdsourcing.org. He also currently contributes to CNET. In the past, his work has been featured by NPR, Wired, the New York Times and other outlets. You can contact him at eric.mack@crowdsourcing.org. Find him on Twitter and Google+. Also be sure to follow Crowdsourcing.org on Twitter.