reCAPCHA: Crowdsourcing Intelligence

CAPCHA (or Completely Automated Turing test to tell Computers and Humans Apart) is a test used in computing to distinguish between computers and humans by presenting the user with a problem that is much more difficult for computers to solve than for humans. Examples of these types of problems include pattern recognition, object segmentation, and contextual understanding.

This technology was originally deployed on websites to prevent spammers from using automated “bots” to create fraudulent accounts on websites.  Initial efforts used background images and colors, irregular spacing, and image distortion to confuse computers.

Realizing that humans were constantly solving difficult problems for computers, a group of researchers created reCAPCHA to put the crowd to work.

Digitizing Books

Initiatives to digitize books generally began with Optical Character Recognition (OCR). However books have different text fonts, sizes, spacing, and irregularities, making perfect recognition of all words difficult.  By taking words that could not be deciphered by computer and giving them to humans to solve, reCAPCHA greatly accelerated the digitization of books, all while reducing fraud on websites.

Pathways to a Just Digital Future

Watch this tech inequality series featuring scholars, practitioners, & activists

Value creation was clear for websites who could filter out malicious users and for digitization initiatives (reCAPCHA began by digitizing the NYT Archive). Adoption was fast because the reCAPCHA plugin could be provided to websites for free, end users were a captive audience, and the cost of deciphering words was much lower than paying individuals. Value Capture was initially more difficult, relying on contracting with companies to digitize archived content.

Acquisition by Google

In 2009, Google bought reCAPCHA and has since expanded its use beyond books. It now uses reCAPCHA to assist in recognition of street signs for StreetView to improve Google Maps and of images to improve its image search tool.  These use cases directly improve Google’s services allowing it to capture value through more targeted advertising.

More importantly and somewhat ironically, using crowds to outsmart computers is now allowing Google to make computers smarter.  Google can learn about how humans use context and segmentation to understand images.  They are crowdsourcing data on how humans think and can then design algorithms that mimic that process or, if done systematically, automatically integrate that mode of thinking into its architecture using machine learning.  In doing so, Google is crowdsourcing the creation of an intelligent computer.

Governance and Challenges

The system has however faced significant challenges.  First, the words need to be sufficiently difficult for bots to recognize but easy enough for humans to recognize – striking this balance as spammers continually attempt new approaches to OCR has proven difficult and reCAPCHA has periodically shown vulnerability. Second, reCAPCHA can be a roadblock for those with visual disabilities. And finally, some end users want to be compensated for the service they are providing to reCAPCHA.  To address these issues reCAPCHA has improved each stage of the process: using better algorithms to make sure the words are sufficiently difficult; creating other traps for bots on websites; creating audible and math driven forms to assist those who cannot do the visual task; and making the user experience simpler to lower to amount of time spent on the task.

The real beauty of reCAPCHA’s use of crowds is that the crowd has no choice but to participate.  Websites are the customers and as long as the tool is free and effective and blocking fraud, they will continue to use the service.  No crowd governance required!