I've been thinking about crowdsourcing's application to development for a while now. Samasource is the most popular such system, but others exist (MobiWorks being the one I've interacted with the most). I've always had core issues with these systems, which have led to some epic debates (primarily with Prayag Narula of MobiWorks) in the iSchool's development seminar. My basic issues (that I can't take credit for) are simple:
1) The market will evaporate; the growth rate of tasks/requesters is much less than the growth rate of Internet users/workers.
2) Outsourcing doesn't actually benefit the workers/communities much; they don't learn useful skills.
3) There's a "race to the bottom"; increased competition over lower paying tasks will just lead to no payments at all. The tasks can't get more complicated, as cross-cultural training is too difficult.
4) The price on Mechanical Turk (the most common crowdsourcing engine) is already at the low end of the Indian living wage, meaning we're already at the bottom and there's not even a short-term win available.
I don't want to go into the specifics of these arguments. Existing solutions address some of these complaints, e.g. Samasource utilizes their non-profit status to pull in jobs at higher wages, limiting the impact of issue three. MobiWorks is discussing accreditation systems that can alleviate issue two.
As a researcher, I decided that I needed to get more data to really analyze this situation. Towards this end, I interviewed a rural Business Process Outsourcing (BPO) office in Karnataka, India, that is doing mechanical turk jobs. The office has 60 employees, with ~15 doing mechanical turk work. The MTurk work started a few months ago, and the manager is quite excited about the technology. The results were amazing, and highlighted how little we actually understand this market. To lead, the interview transcription is available here. This interview was around an hour long, and conducted a month ago. It was transcribed on Mechanical Turk, so please excuse the errors. I tried to anonymize anything that could be problematic.
The finding are numerous. Firstly, though I'm fairly certain he was exaggerating, the manager claimed a daily wage of 20USD per turk-worker per day. This is huge, and well above the ~4 USD/day he was paying his normal BPO (call center) workers. The justification was that, if you avoid the really cheap tasks on turk, you can make a lot of money. This was really surprising. I had expected a slog through low-wage tasks for 8 hours a day. Instead, workers were doing interesting, highly demanding, intellectual tasks like content authoring. That's game changing.
Unfortunately, that's just part of the story. The content authoring was (seemingly) mostly for content farms like Direct Media, basically spamming search engines. In fact, the vast majority of their high-paying tasks were spam: spamming Craigslist, search engines, or installing browser toolbars. Apparently you can make 20USD/Day doing that. I shouldn't be surprised, but I am. Is there a future for this type of work? Is outsourcing the spamming industry really an economic model?
However, the workers seem to enjoy that work. I wasn't able to push too hard on the manager in terms of the ethics of the situation, but it's hard to argue with success. They're working, earning a lot, learning valuable skills, and enjoying their life. You can't ask for much more "Development" than that. Of course, these workers are primarily the Indian middle-class, so one could easily argue that this is not going to impact the poorest of the poor. True or not, this is more impact than most ICTD projects, and I find it hard to argue with.
I also learned about some new issues for Turk-farms in India. I'm very surprised that you cannot post tasks from India. MTurk explicitly disallows everything but outsourcing. Why? It's probably due to practical issues (e.g., Banking) rather than any imperialist bent to Amazon.com, but it is quite unfortunate. You have to believe that there would be more value if workers were not forced to cross cultural boundaries to complete work. There's clearly a lot Amazon could do to improve Turking for Indian workers. Allowing for multiple worker accounts to share one bank account would be an easy win for BPO operators. The manager argued that allowing for users to explicitly "work together", rather than alone, on tasks improves value for both workers and requesters.
There's a lot more to this interview, and I'm likely to take up a few blog posts discussing it.
In the end, this interview improved my view of crowdsourcing for development. It hasn't resolved all of my issues, but rather showed benefits I hadn't anticipated. Crowdsourcing, as demonstrated by Mechanical Turk, is clearly an immature system that has a lot of growing to do. However, this seems to be (I think) a promising way to prove the livelihood of workers in low-income parts of the world. In the least, it's not obviously doomed. Doing it right, in terms of both economics and ethics, is going to be difficult. However, that's just life.