Human based Computation as a form of Currency to alleviate the world Poverty?

This post is a summary (+ some comments & further references) of a short paper “Secure Distributed Human Computation” in financial cryptography (2005) (the article is a position paper, so no mathematical at all compared to the previous one in machine learning [6]). Here the authors describe a new paradigm that some of you already know called “human based computation” (HBC) [2] and show that its can be used as a new form of currency for the bottom of the (economic) pyramid [4], the 4 billion poorest people (note: despite the fact that they lives with less than $2 per day, they constitute a $5 trillion global consumer market according to [5]).
You might wonder why this kind of paper has been accepted to a financial cryptography conference. Once you have made the jump to think HBC as a form of currency you will understand the relation with the financial field and the security problems lying on this new paradigm (how not to be able to game this kind of system), but that’s not what I wanted to talk about

Human Based Computation

Human based computation or distributed Human Computation is a paradigm where humans assist and help to solve problems that are easy for them but still difficult for computers. Many problems fall into this category e.g. AI problems which occur in fields such as image/speech recognition, and natural language processing. Some famous examples in image/text/speech analysis problems:

  • CAPTCHA Solutions: email providers incorporated special puzzles during the account creation phase, easily solved by humans but challenging for computers, to prevent spammers to use bots; (note: this filter can be perceived as a Turing Test [2] to differentiate Humans from Computers, so it still doesn’t protect providers from spammers who use people from poor countries as a HBC resource)
  • The ESP Game:A game where two players have to tag a set of images displayed on the screen. This kind of game was used by Google to index their database of images instead of using image recognition technics.
  • Collaborative filtering/Spam prevention: humans can more easily identify junk mail than computers, some spam prevention mechanisms leverage human votes.
  • Medical Transcription Services: Today many doctors dictate their medical notes verbally via a interface that records and transmits a compressed audio file using the Internet. At this point, a human listens to the audio file, transcribes the note, and sends it back electronically.
  • Distributed proofreaders: (www.pgdp.net) is a project that aims to eliminate optical character recognition (OCR) errors in electronic books. A portion of the image and its corresponding generated text is given to several humans to correct remaining errors.

HBC as a form of currency in the digital age

Solving a problem that can’t be solved by computers has value. The traditional way is just to pay people. As user-generated content proliferates on the web, web-sites are devoting increasing resources [2] to moderation and abuse detection. For instance, Photobucket.com, a Web service that lets people store images and video online, has had to hire an increasing number of employees to check images. The firm spent about $1 million in 2007 for content monitoring (an employee can look at nearly 150,000 images per day or about 300 a minute) [7].

But HBC may also enable the Business-to-Four-Billion (B24b) model [4] which aims to provide digital services (wireless communication, internet, etc.) to the world’s four-billion poorest people. Web sites typically rely on three revenue sources: advertisements, subscription fees, and e-commerce. Instead of cash transactions, one may provide digital services in exchange for solving DHC problems. The notion of distributed human computation yields another revenue source:companies that want solve specified hard problems can outsource them.
For instance in addition of charging a fee, the New York Times could give to the user a human computation problem (e.g., transcribing an audio feed into text). In exchange the archived articles can be provided.
Another example in Internet Telephony. The mean opinion score (MOS) is a subjective numerical measure of the quality of human speech at the receiver. Here, the “hard problem” is that it is not clear how to use a computer to measure sound quality; i.e., there is no known algorithm to measure the sound quality of call. A solution is to provide for instance one free minute of long-distance cell phone service in exchange of measuring the sound quality of a set of calls.

But this vision has clear limitations:

  • There are ethical issues: ensure that the market for human computation is not exploitative.
  • Is there enough demand of AI problem for the BOP?
  • The current business model is the advertising which aims at providing already a free access to web services.

Nevertheless I thought it was interesting to see HBC as a form or currency even if there are clear limitations and remains to be seen whether such an endeavor is economically feasible

References

[1] Gentry, C et al (2005) “Secure Distributed Human Computation” ; In 9th International Conference on Financial Cryptography FC’2005 online
[2] Human Based computation Wikipedia
[3] Turing Test, wikipedia
[4] C. K. Prahalad (2004) The Fortune at the Bottom of the Pyramid. Book
[5]
The Next 4 Billion: Market Size and Business Strategy at the Base of the Pyramid
[6] Summary of the paper “Extreme Machine Learning”
[7] J. Angwin. (2006) “A Problem for Hot Web Outfits: Keeping Pages Free From Porn” , Wall Street Journal



Comments are closed.