Google is using the amazing pattern recognition capabilities of millions of distributed human brains to perfect its optical character recognition software.
The trouble with crowdsourcing
Crowdsourcing has been getting some bad press lately, and for bad reasons.
For example, Google built a site called Google Flu Trends, which ideally can track and even predict the spread of flu across the world. In the past, Flu Trends has been accurate. But during the most recent flu season, Flu Trends " wildly overestimated" the outbreak.
Flu Trends crowdsources search queries for information about flu and thereby is supposed to indicate actual flu.
The problem with this assumption is that it doesn't measure actual flu at all -- it measures public anxiety about flu, which is subject to media manipulation. People go rushing to the search engines to find out about flu symptoms when the news is hyped about a coming outbreak.
Another supposed failure is the crowdsourcing on the social network Reddit of the manhunt for the Boston Bomber terrorist suspects. While actual events were unfolding, many on Reddit falsely identified a young man as a suspect. The man had been missing before the Boston bombing and was later found dead.
In general, crowdsourcing that tracks actual behavior is more useful and accurate than crowdsourcing that tracks opinions and attitudes.
Crowdsourcing is great for surfacing options to choose from (as I did with my crowdsourcing for this article). It's great for brainstorming and getting ideas. And it's great for gathering large numbers of data points and applying algorithms to that data.
Crowdsourcing isn't so great when you assume a weak correlation between measured attitudes and something else. It's important to remember that all you're really crowdsourcing is the attitudes, which can be manipulated, either by the media or by the crowd itself.