This presented fascinating possibilities for Google developers who have been struggling with the quandaries represented by the Turin test or the Chinese Room of John Searle, namely how to deal with cognitive tasks that humans find easy to deal with, but computers find difficult or cannot process at all.
The problems surrounding how to categorise pictures for the purpose of providing contextual search capabilities has been a thorn in the sides of the search engine developers since graphics first started to be used on the internet. As computers cannot yet effectively scan any image which is presented to it, and then every time come up with an accurate description of the objects shown in the picture, the process of automating the listing of graphics in a way that makes indexing and retrieval using labels which are meaningful for humans a seemingly impossible task. Allowing users to attach their own labels to their own picture is an option which quickly became unworkable due to the risks posed by spammers and simple human error.
Professor Luis von Ahn found that many people found the labelling activity of the ESP game to be fun, with users playing over 20 hours a week, and some even for as long as 15 hours at a time. This meant that over the three year period during which the game had been running, 75,000 players had successfully agreed on over 15 million labels for the images presented.
This presented an obvious possibility for potentially combining the activities of the game players and harnessing their labelling capacity, with the objective of indexing all the images online in a user search friendly manner. Von Ahn predicted that it would take 5000 simultaneous players a mere two months to successfully label all images currently contained within the Google images listings.
The upshot of this seemingly simple idea is the newly unveiled Google Labeler. While losing some of the fun factor which was inherent in the original game, such as silly sound effects, and taboo words to add to the challenge, users are still paired up with random partners and it is very easy to become caught up in the excitement of trying to work together against the clock to score points. The advantage for Google is that as the users have fun trying to quickly match up the labels which they think accurately fits for each of the pictures with those provided by their partner, Google is collecting valuable information on the terms that real people assign to the various graphics. If two randomly chosen players match up their labels, then that word is obviously relevant to the image, and could be used by someone searching for that picture. Google can therefore assign keyword search values accordingly to the graphics using these labels.
While not presenting a leap in artificial intelligence which will conclusively overcome human computer interface problem, as a means of tackling the problems presented by the differences between human cognitive processing against machine processing, especially when it comes to graphics, this represents an ingenious way to harness all the collective intelligence and processing capacities available, creating a highly useful and effective symbiosis between humans and computers.
