There is a growing effort to utilize crowdsourcing to implement “citizen science”. An early success was “Galaxy Zoo”, which recruited hundreds of thousands of ordinary people to help classify images from the Sloan Digital Sky Survey.
The idea was so successful that it has been replicated by many other projects, not only for Astronomy. The Galaxy Zoo is available as a toolkit, and many projects can be found at “Zooniverse”, and many other similar projects use other tools.
The core process of Galaxy Zoo was (and is) image classification (“does this image show a spiral or a disk galaxy?”) and other visual tasks that humans excel at. An increasing number of projects involve handwriting recognition and other linguistic tasks have picked up this methodology.
The basic idea is that some tasks are difficult for computers to do, yet very easy for humans. The crowdsourcing framework presents tasks, such as images or sounds to be processed, and asks simple questions, using an Internet portal. The resulting data can be accumulated and analyzed.
This is the same idea as crowdsourcing systems a la Amazon’s Mechanical Turk, except the incentive is to voluntarily help scientific research, rather than micropayments . In either case, the data collected can be used to cross validate other classifications, or as training data for machine learning, or possibly to answer questions that cannot be addressed otherwise (especially where natural language understanding is required).
This week the Planetary Response Network (PRN) demonstrated yet another use for this technology: disaster response.
“The purpose of the Network is to supply timely and reliable tactical information requested by established humanitarian organizations in the immediate aftermath of major disasters, through distributed analysis of satellite imagery”
In particular, this week the PRN asked the Internet to help damage assessment from the April 16 Earthquake in Ecuador, using recent satellite imagery.
“We need your help to identify damaged structures and blocked roads in real images of the aftermath of the recent earthquake in Ecuador.”
(from the Tutorial)
The web site indicates that all the images were examined in 12 hours, only 8 days after the quake. (Each image was rated by more than one person.) Even if this survey is only slightly useful, it is certainly better to have them now than to do without or wait for them.
It is important to point out that the DIY Zooniverse toolkit was a key to booting up this project so quickly and cheaply. This seems to demonstrate that the DIY aspect is pretty well designed.
I should note that there are also more than one “crowdfunding” effort raising money to support relief operations (so you don’t have to limit helping to image recognition).
- Adam Marcus and Aditya Parameswaran, Crowdsourced Data Management: Industry and Academic Perspectives. Foundations and Trends® in Databases, 6 (1-2):1-161, 2013.