Tech’s top ten thinkers
With the number of internet users still growing exponentially, Hölzle has also been improving the way Google handles the millions of cores in its datacenters.
Complex equations predict which processors give the best results for the millions of parallel requests coming in from users every minute.
“In warehouse-scale systems throughput is more important than single-threaded peak performance, because no single processor can handle the full workload, and maximising single-threaded performance costs power through larger die area,” he said in a recent IEEE presentation.
THE CROWD SOURCERER: Luis von Ahn
“I build systems that combine humans and computers to solve large-scale problems that neither can solve alone,” Luis von Ahn writes on the Carnegie Mellon University website. “I call this Human Computation, but others sometimes call it crowdsourcing.”
Whatever it’s called, von Ahn’s work can be frustrating: he was the brains behind the CAPTCHA system that prevents spambots from filling in web forms, although frequently stops humans from doing so too by asking them to identify illegible phrases. Despite its flaws, it was bought by Google. Its successor, reCAPTCHA, takes advantage of the fact that 200 million people were, in von Ahn’s words, “wasting 500,000 hours a day typing annoying CAPTCHAs”.
Von Ahn thought those wasted hours could be put to better use, and now reCAPTCHA asks people to fill in two words: one regular puzzle and another that Google’s OCR has failed to identify during its attempt to digitise millions of books. Why? Because humans are better at converting images to text than the fallible OCR software. As a result, reCAPTCHAs placed on sites such as Google, Facebook and Twitter are used to digitise more than two million books a year.
His latest idea, Duolingo, uses crowd sourcing to translate the web, with von Ahn claiming that computer translators don’t work, while paying professional translators is prohibitively expensive.
I build systems that combine humans and computers to solve large-scale problems that neither can solve alone
Von Ahn overcame concerns that there wouldn’t be enough bilingual people to achieve his aim, and that it would be impossible to motivate them. His solution? Turning the scheme into an educational program where participants learn a language while they translate web content.
Von Ahn claims the results rival those of professional translators, while “students” are shown increasingly complicated sentences to improve their skills. With a million users, von Ahn says Duolingo could translate Wikipedia from English to Spanish in 80 hours, for free – $15 million less than using human linguists.
THE UI GURU: Indrani Medhi
With billions of people still without computers and unable to read, Medhi has been working on user interfaces that are more inclusive than a South London council.
A design graduate now employed by Microsoft Research in India, she is working on non-text interfaces for the world’s illiterate would-be computer users.
Medhi believes such interfaces could open up a huge market catering to poor and poorly educated first-time technology users, who could interact with smartphones and PCs without any assistance.
There are inevitable limitations to what non-literate computer users could do online, but Medhi says her research has shown high completion rates for simple tasks.