This week at the BUILD conference, Microsoft unveiled launched a set of machine learning APIs dubbed “Project Oxford”. As part of that launch, Microsoft built a site named How-Old.net that acts as an experimental demo of the service.
The site lets you upload a photo under 3MB and then tries to figure out your age and gender. “It’s a really fun API [application programming interface] that allows you to submit pictures, detect faces in that, and tells you an estimate of their age and gender,” Microsoft corporate vice president Joseph Sirosh said onstage. Tens of thousands of visitors tried it out and before long its amusing errors became a meme.
On the backend, the service is using Project Oxford’s Face APIs to process an endless amount of photos and Microsoft notes that the service will continue to improve. We were left to wonder why How-Old.net doesn’t have a field that lets users input their true age and gender in an effort to help the service learn that much faster.
We also found that the service seems to be easily thrown off based on a number of factors such as how bright the lighting is, the expressions being used and whether you are wearing glasses or not. Here is ProgrammableWeb's 53 year old editor-in-chief David Berlind putting on his best 39 year old face.
Another example of the service falling short, is this image submitted by Facebook user Mary O'Donnell, where the Starbucks logo was credited with an age of 36 (for what it is worth, Starbucks is 44 years old):
Glitches aside, the demo raises questions of what the potential use cases are for facial recognition APIs. Kairos CEO Brian Brackeen earlier discussed with ProgramambleWeb some of the applications including accessing the medical records of an unconscious healthcare patient, verification for those taking online exams, and preventing identity theft and credit card fraud using some form of point of sale facial recognition. One can even imagine the beauty industry using technology such as this as a way of enticing consumers in a never ending quest to look younger.
While the demo site is undoubtedly limited, the promise of machine learning APIs in this context seems anything but. For now, developers that are interested can try out the services for free at the Project Oxford site.