Microsoft age-estimate tool unleashed real-time virality

Microsoft has launched machine-learning APIs in beta under the "Project Oxford" moniker. Its How-Old.net demo for this service went viral, reported Frederic Lardinois in TechCrunch on Thursday. How-Old uses some—but not all—of the new developer services that are part of "Project Oxford," he said. Meanwhile, Microsoft engineers Corom Thompson and Santosh Balasubramanian on Microsoft's Machine Learning blog told the related story about running a test and expecting, with luck, to get as many as 50 users. Unraveling before their eyes came the nature of the draw; the test drew over 35,000 users.

What was the hot topic? "We were playing with Microsoft's newly released Face detection API's through a webpage called how-old.net ."

This is where users upload a photo of a face and the API has to figure out how old is the person in that photo. The website will let people upload their own pictures which is then scanned followed by the API making a guess about the age; users can also decide to pick photos from a web search.

The two engineers in the blog said, "we assumed that folks would not want to upload their own pictures but would prefer to select from pre-canned images such as what they found online. But we what we found out was that over half the pictures analyzed were of people who had uploaded their own images. We used this insight to improve the user experience and did some additional testing around image uploads from mobile devices."

They had e-mailed several hundred people asking them to try the page for a few minutes and submit feedback. At least 50, they guessed, would want to give it a shot. "We monitored our real time analytics dashboard to track usage and, within a few minutes, the number of people using the site vastly exceeded the number of people we had sent our email to. We watched the usage quickly spread across continents," they wrote. Within a few hours, they said, "over 35,000 users had hit the page from all over the world (about 29k of them from Turkey, as it turned out – apparently there were a bunch of tweets from Turkey mentioning this page)."

According to VentureBeat, there was another story to be told as well. The "My Twitter timeline blew up with selfies with boxes hovering above faces guessing how old people are, thanks to a demo app Microsoft showed off today to prove its face-recognition savvy—or lack thereof." VentureBeat's Jordan Novet said Microsoft corporate vice president Joseph Sirosh taking the stage at Microsoft's Build conference in San Francisco on Thursday, told people in the crowd and online to check out the app, at how-old.net, which uses Microsoft's Face API. There were remarks about how some of the How Old Do I look? app estimates were off the mark. As Lardinois suggested, "It's best to take its results with a grain of salt."

Novet also reported that Ryan Galgon, a senior program manager at Microsoft Technology and Research, said age and gender-recognition features were ones "that we definitely want to improve over time."

Novet commented that Galgon was "right to talk about how the Face API could improve. Any usage of the app, even if it results in inaccurate renderings, is good usage, because Microsoft gets more data to improve its systems."

Lardinois, meanwhile, said that "Much of what's available through the service today is based on modern deep learning techniques the company worked on over the last few years."