Thursday 25 April 2019

completely automated public turing test to tell computers and humans apart

NPR’s Planet Money recently presented a really engrossing episode on CAPTCHA technologies—previously and which is the title initialism—that shows how the counter-measures against spam and subterfuge not only evolved with but also to a big degree informed the developing internet to make it a richer and more connected experience.
It reminds me of the possibly apocryphal story that the suite of games pre-installed in Windows (despite taxing precious system resources) was distributed in order to teach users the manual dexterity needed to operate a computer mouse and while I do remember some times being asked to provide my human credentials by identifying a blurry building number, what I failed to realise was that human we’re just being shown skewed text and images to thwart automated bots but were also being enlisted by comprehensive mapping services and earlier by newspaper archives to perform the optical character recognition that machines were not yet able to decipher when the copy was suboptimal. Later iterations of the test tasked people with identifying cars, boats and buildings and aided in machine-learning by tutoring neural networks in a funny sort of feedback loop that’s enabled computers to beat the original methods of minding the gates.