Yes, the business case rests strongly on having a big enough market. We think (strong gut feel, but without much data to back it up yet) that there is a very large potential market. It's kind of a "latent" market - it's a way to solve problems that people are not thinking about yet. I think of it like this: before computers became widespread, did people think about using computational tools to solve problems? No, not really. Likewise, I think, for "human computation". The capacity of the human perceptual systems to process input, and to make subtle judgments, is really tremendous, but that has not been harnessed as a resource for problem solving until recently.
There's definitely a market for what I would call "plain vanilla" human computation tasks: text and audio transcription, business listing verification, etc (high volume but cognitively boring stuff). The existence and revenue-generation by MTurk and Crowd Flower, for example, is strong evidence of this. We also think there will be a market for more interesting problem-solving applications in science and engineering research.
Another example: Air Quality Researcher Guy wants to model exposure to indoor and outdoor air pollution across regions in the country; people spend lots of time in their houses, so that constitutes a lot of their exposure (integrating over time); open windows dramatically change the indoor/outdoor air mix. Therefore if we know something about window-opening frequency (and regional demographics, weather, etc) we can correlate that to incidence of respiratory illnesses, for example. Soooo...we sample images from Google Streetview and have people tell us if windows are open! (simplified, ongoing project).
We also think that the field of metagenomics, and the other new "-omics" fields in biology may have some interesting applications. They generate data much faster than they can analyze it, and the tools for making comparisons between this-thing-that-we-don't-know-what-it-is and that-other-new-thing-that-looks-kinda-like-the-old-thing-sorta-maybe are incomplete and don't do a great job (plus the inherent inexact comparative nature of biology) makes us think human computation could contribute to solutions.
Another possible application: a single-stream recycling sorting center using real-time (or near enough to real-time) human computation for sorting objects according to a binary tree corresponding to a binary conveyor system.
Some other things people have already done using human computation: ask humans to generate social scripts to help "train" people with autism for various social interactions; have people answer questions about pictures taken with a smartphone app for blind people ("Is this beverage can a Coke or a beer?"); tag photos from a film company's archive and use an algorithm to link entities and "connect the dots" to realize that the photos are the lost rolls of film from the set of "American Graffiti"
So there's reason to think there are lots of interesting, as-yet-undiscovered applications for this! So again...if this lights any lightbulbs, or you have suggestions for applications, please let me know!
That's definitely a possibility! Ideally we want to harness the natural creativity that fuels capitalism - so we want to allow some flexibility in how workers get recruited while still making sure the incentives are aligned to promote beneficial outcomes.
Great points. I have personal connections in some poor rural areas of Kenya already (and already in that place, people are always asking, "How can I join?", both of me and of the current workers). My colleagues also have connections in a number of other developing countries where we could plant "seeds". How to grow the "crowd" from those seeds is an interesting problem, but not insurmountable, what with ever-increasing mobile phone/3G penetration and a mobile interface to our platform (lots of people in Kenya, for example, have a low end mobile phone or internet-capable "feature phone" while still living in mud huts), a franchise model using netbooks and small solar stations, etc.
As for quality control, you're absolutely right. There are a lot of ways to approach that (none of which Mturk implements). There's some well-established precedent for methods that work, so we feel confident we can generate high quality outputs.
Yes, Mechanical Turk is another platform that enables human computation work. We looked a lot at that in our early research. It does not, however, implement any internal quality control mechanisms, and it also only allows payment in US dollars, Indian rupees, or via Amazon gift card. The interface is also clunky and difficult for people with limited computer/internet experience to understand (too many windows within windows, basically).
So there are a number of reasons to sidestep Mechanical Turk.
Hi folks - long-time reader, first-time poster here. I'd like to share my new startup, formed over the past few months. In short, we use human computation to create employment opportunities in the developing world while enabling new types of data analysis for solving complex problems.
Some background and justification: extreme poverty is bad. I won't say too much to justify this point, other than that human suffering and the lost potential for individuals to do great things are two of the big things that make it bad. And just to be clear, by "extreme poverty" I am referring to the "living on $1 per day" kind (though that's a not a very good definition, it's at least in the ballpark). One way to combat extreme poverty is by creating employment opportunities so that people can help themselves, rather than giving them free shoes, or corn, or wells, all of which are suboptimal for meeting their varied pressing needs. So our approach is to hire them to do human computation work.
Human computation is when people do things that are easy for people but hard for computers. A good example is image processing/recognition (this is why reCAPTCHA works). By combining the things people do well with the things computers (i.e. software we currently able to write) do well, we can enable new kinds of data analysis. For example, we can mine figures from the medical literature for depictions of biochemical pathways, recreate many of them together (molecules = nodes, interactions = edges), and create a more complete picture of our biology.
So that's our approach: find an interesting, complex problem (so far they have been in the academic research world), collaborate with the domain experts to design human computation processes to enable the necessary data analysis and synthesis, and using our web platform, pay people in developing countries to do the work. Interesting problems get solved, we get paid, and people in Kenya get paid. Win, win, win.
Interested? We are looking for:
People with problems to solve using human computation. We are especially looking for domain experts here. Not sure if your problem is amenable to human computation? Let's talk.
Programmers. We've got a platform now but are continually improving it. Python/django. There's also a fair amount of cobbling together datasets for input and output that presents ever-changing challenges in many different languages.
Marketing/sales. Our concept is a hard thing to explain to people. People don't seem to be used to thinking about solving problems in this manner, so it's difficult to get people to think, "Oh yes! I have problem X and this will help me solve it!". We need to figure out some way to communicate this better in order to expand the problems to work on and increase revenue.
Funding. We need to pay programmers, marketing people, and us.
Other! Think this is a cool idea but don't fit into one of the above areas? Let's talk!