Innovation: Online army turns the tide on automation
by Gareth Morgan
Innovation is our regular column in which we highlight emerging technologies and predict where they may lead
Computer automation can take jobs away from people but, thanks to Amazon’s Mechanical Turk, humans are fighting back. AMT was inspired by the 18th-century inventor Wolfgang von Kempelen, who dazzled the Roman empress Maria Theresa with a chess-playing automaton. His secret: a human chess master hid inside the machine.
In 2005, online retailer Amazon developed a version that uses a human workforce “hidden” on the internet to solve problems – for a modest price. Typically, the work undertaken is for organisations that need a little human smarts applied to bulk tasks, such as identifying objects in vast collections of images.
An echo of von Kempelen’s Turk is found in the offices of robot maker Willow Garage, in Menlo Park, California. Some of the firm’s free-roaming robots rely on humans through AMT to help them get their bearings. Whenever one gets lost within the Willow Garage offices, it sends an image to AMT with a request for nearby objects to be identified, using the answers to establish its whereabouts.
Get shorter
At the User Interface Software and Technology symposium in New York City this week there are signs that AMT rivals computer automation on some tasks.
Michael Bernstein at the Massachusetts Institute of Technology, and colleagues, have developed Soylent, an add-on for Microsoft Word that uses AMT workers to check language and grammar. In tests on text from Wikipedia entries, Word’s grammar checker picked up about a third of errors; Soylent spotted two-thirds.
Solyent’s Shortn module tasks the online workers with shortening the text – to meet a word limit, for example. The Word add-on also boasts a macro-writing module, Human Macro, which lets a writer describe how they want to manipulate text – perhaps changing it into the past tense – without the complication of having to code their own set of instructions within Word.
Say what you see
Meanwhile, Jeffrey Bigham at the University of Rochester, New York, and colleagues, are using the image-analysis capabilities of AMT workers – predominantly based in the US and India – to help the visually impaired. They have created an iPhone app called VizWiz that gets AMT workers to interpret objects in the user’s environment – checking the small use-by date on a carton of milk, for example.
The app is able to analyse the iPhone camera’s focal length and lens distortion, and data from the built-in accelerometer, to pick out a target object in sufficient detail before sending it. After identification, the result is read aloud.
However, despite their lack of real brain power, there is one advantage that computers will continue to hold over their AMT rivals: computers don’t charge for their labour.
References: Bernstein’s Soylent research paper (pdf); Bigham’s VizWiz research paper (pdf)
, 05 October 2010