Last night I attended a phenomenal meetup sponsored by Lukas Biewald’s Dolores Labs, which builds tools to help companies use Amazon’s Mechanical Turk (AMT) more effectively. (For those unfamiliar with Amazon’s “human intelligence task” marketplace, read my earlier post on AMT.) People are using the system for things like analyzing language, training machines, crowdsourcing robotics, and conducting research in economics and psychology (it turns out to be a great way to find cheap research subjects).
Basic tasks pay $.01-.03, and can be completed fast enough to earn a worker between $1.20 and $3.50 an hour. While most workers come from the US (signing up requires a US bank account), and use the Turk to supplement their income, Lilly Irani‘s surveys found that 15% of workers depend on their AMT income to get by. She argues that the AMT interface makes it hard to see the cloud as individual people with hopes and dreams and a need to earn more than $2 an hour with no benefits. Sharon Chiarella, VP of AMT, mentioned that system only has 200,000 workers. That’s only .02% of internet users. An influx of quality workers could make MTurk a lot more useful to developers, academics, and other people with huge task volumes.
Here’s where Samasource might fill an obvious need. It makes little sense for American workers to spend much time on the Turk; the tasks are not high enough in value to get done here, where per-capita income averages $21,500 (vs. $450 to 900 in low-income countries like Kenya). So why don’t we outsource these types of task to people who have no other sources of income, like refugees or prisoners? The nice thing about those two populations is that computers and internet access, as well as benefits like health care, are often provided by social service agencies or non-profits.
Samasource wrote a funding proposal (rejected recently by USAID) to use two existing IT centers built by CARE International in Dadaab, a refugee camp with a quarter of a million Somalis living in a facility that’s 270% over capacity. Overcrowding makes other kinds of livelihood development difficult; many traditional activities must compete with basic needs for resources. A recent Oxfam analysis found that 50% of the population has access to less than 13 liters of water per person per day, and that one reason for this shortage is use of water for livelihoods activities (ice-making, butchers, hotels, and brick-making, for example).
Refugees represent a vibrant pool of untapped talent. Current livelihood programs are limited by resource constraints and infrastructural challenges associated with creating and transporting products. Remote work is a viable alternative; by training refugees, many of whom have a high school or college education, in things like task completion, we could solve the need for more workers on the Turk and give some of the poorest people on earth a chance to earn a living.
Here are our current obstacles to testing our model in Dadaab:
1. AMT is currently set up so that each worker needs his own US bank account to complete tasks. Samasource can act as an aggregator and collect payments on behalf of lots of workers, but we’d get shut down if we tried this right now on the Turk.
2. Bandwidth is sometimes inadequate in East Africa because the region doesn’t have reliable fiber. We need to test using AMT in Dadaab’s IT centers before it’s clear this will work (since the IT labs were previously used for online university courses, I’m guessing it will).
3. The skill sets of refugees in Dadaab are highly variable, so we may need to invest in finding and training people on task completion before we start seeing good results. The nice thing about AMT is that it can be used as a training platform (requesters have the option to reject tasks that are not completed properly, creating a good feedback loop).