Thursday, July 12, 2012

From Crowdsourcing to Crowdseeding.

I just returned from a workshop at the Free University of Berlin. Together with fellow-authors a book is put together about "ICT in Areas of Limited Statehood". The book deals with how new technology (phones, satellites, mapping, etc.) can be used and enable us to learn about and solve the world's big problems -- especially developmental ones. This was the second time we met as a group (last time was about a year ago: post here) so the feeling-uncomfortable-around-each-other was gone, and we had a really good time, also socially. It is also a particularly great group, which includes some of the big names when it comes to ICT, and some super-impressive people that work on the implementation side (the key people of Ushahidi, Russia Helpmap, and Map Kibera).

I'm writing a chapter based on our experiences with Voix des Kivus: our cellphone-based system that we implemented in Eastern Congo to learn about local level events (more here). To obtain high-quality (representative) data in real-time we implemented Voix des Kivus with a crowd-seeding system in contrast to the-know-so-often-used-term crowd-sourcing? Why?

Under crowd-sourcing, in contrast to outsourcing, a task is not given to a specific group of people but to an undefined public. This has two major benefit: 1. it's relatively cheaper to collect large amounts of data, and 2. One can make use of the "wisdom of the crowd": the idea is that a group of people is smarter than one.

I presented Voix des Kivus on the second day of the workshop so during day 1 I was thinking in what way crowd-seeding is different from crowd-sourcing, and how to convincingly bring this across. You're ready? Please have a look at the picture below. A+B+C+D is the whole population of people (for example all the people in South Kivu, Congo where Voix des Kivus works). "Knowledge" stands for people knowing about the task send into the crowd. "Means" stands for whether or not the people can undertake the task. In the Congo the idea was to make use of the Congolese crowd. However, given the conditions in Eastern Congo a large part of the population won't know about the Voix des Kivus project. As result we're confined to the first row in the table only. Moreover, in the Congo few people have cell-phones and even less have also phonecredit and thus we are left also only with the left column. That is, crowd-sourcing would make use of the subset of the population: only A. In the Congo this is a very small part of the whole population (A+B+C+D).

Under crowd-seeding we select the reporters and visit them, make them aware the project (in our case Voix des Kivus), and we give them phones with phonecredit (we 'seed' phones into the crowd). As a result, crowd-seeding increases the crowd from A to A+B+C+D. That's the first benefit: a larger population. Moreover, Voix des Kivus took a random subset of A+B+C+D to be reporter: as a result i that the data that we receive is representative -- again in contrast to a crowd-sourcing system where its likely a particular type of people send in information (indeed a forthcoming article in Political Analysis by Adam Berinsky, Gregory Huber and Gabriel Lenz investigating the crowd-workers of Amazon's Mechanical Turk finds exactly that). Finally, under a crowd-seeding system one knows the people that provide information. One builds up a long-term relationship with the information providers and as a result the incentives to lie decrease (important in the Congo where people might want to overstate suffering and hardship in order to receive aid) and it makes it more difficult for the bad guys to hack the system and provide misinformation.

The picture above is made by Patrick Meier (friend, founder of Ushahidi, etc.). He wrote about the presentation here.

Now moving from a crowd-sourcing system to a crowd-seeding system is not all honky-dory and this is what we learned from our experiences with Voix des Kivus in Eastern Congo. Macartan and I wrote our reflexions up in a short document that we think is important reading material for any person trying to implement a phone-based project to collect information, especially if they are academics and work in security-sensitive areas. Macartan presented this document earlier this year at the University of Bristish Columbia. Please find the document here:

No comments:

Post a Comment