02 April 2006
You might have heard of Amazon's Mechanical Turk. It's a web service that enables you to integrate human tasks with your software. So you might have a program that needs to know which of ten pictures is in focus, for example. You would write your program to feed that judgement task to Amazon's Mechanical Turk. Amazon then gets people to perform that task, pays them, and takes a commission on the payment from you. The data is returned to your software as if the task had been performed automatically. A neat idea, and it remains to be seen how it's used. So far, it's mainly Amazon using it for identifying photos of businesses for its own A9 search engine. It could be exploited by spammers to get around the intelligence tests used to prevent automated data entry (including tests such as working out the jumbled letters in an image). But it's not hard for Amazon to monitor at the moment, so this is unlikely to be a problem any time soon.
Why is it called a Mechanical Turk? It's a reference to an automaton invented in the late 1700s to play chess. At the time, mechanised artworks were popular inventions. They might play instruments or animated landscape scenes. Wolfgang von Kempelen was a member of Empress Maria Theresa's court in Vienna. When he was challenged to produce an automaton better than any seen before, he created a machine capable of playing chess. Dressed as a Turk, the machine went on to play against - and mostly beat - members of the public. The Turk moved its own pieces and corrected an opponent's illegal moves. Kempelen proudly exhibited the cogs inside the cabinet at the start of the game, showing how the box was empty except for clockwork devices. For a long time, nobody came up with a conclusive explanation of how mechanics alone could simulate such intelligence.
In his book 'The Mechanical Turk', Tom Standage uses many contemporary accounts of the Turk to tell its story. He pieces together the machine's complete history, and ultimately reveals how the illusion was most likely created. It's not giving too much away to tell you that human intelligence was involved and it wasn't fully automated. That's why Amazon stole the device's name for its own service, where a fairly convoluted technical arrangement creates the appearance of machine intelligence.
At times the exhaustive research gets repetitive: reading accounts of the same event from different perspectives can be tiring. But these moments are few and far between. It's accessibly written, shows proper research and tells a story that's likely to delight anyone with an appreciation of illusion and/or technology.
I often read about software that performs such sophisticated functions that I question whether human intelligence is really being used. The developers of such applications are - like Kempelen - often unwilling to discuss how their product works even in the vaguest of terms. Given how cheap educated labour is on the global market, it wouldn't surprise me to discover that some enterprise-scale software applications that simulate human judgement owe more to the Turk than they're letting on. For the so-called developers of such software, this book would make an ideal gift.
With thanks to Patroclus for the recommendation and loan of the book.