Bartolo Colon Versus Microsoft's Age-Determining Technology
Microsoft rolled out a public test of its new face-detection technology on Thursday at How-Old.net. It is not ready for Bartolo Colon. Not even a little.
On Thursday, the internet took a break from what has already been a long and shittier than usual week to enjoy a new Microsoft face detection program that, truth be told, doesn't work very well yet. The program was designed to detect faces in a photo and ascertain the age and gender of the people those faces belong to; Microsoft was nice enough to put this technology online at How-Old.net, and a grateful nation took a few minutes off from yelling at each other and worrying about things to let a robot guess, wrongly, at how old they are. We laughed, secure in the knowledge that this is surely not the sort of technology that could ever have any creepy panopticonic application if used by powerful creeps in a creepy way.
Advertisement
Here's how it works: you take a photo—say, a picture of yourself or a loved one—and put it into the tool. Here is a photo selected at random after performing a Google search for "very normal image."And then you just upload that image to How-Old.net and get a result.
Or you don't. The technology, as Microsoft admits, is still being perfected. In the interest of helping this process along, our VICE Sports technologists selected photos at random to see if we could figure out where this new application falls short.We began with a question: is it harder for the How-Old robot to determine the age of people who are in motion? We tested this by doing a random image search using the parameters "graceful athlete in motion."
While the robot was able to identify Bartolo Colon's gender correctly, it was wrong about his age—Bartolo Colon is a member of a secret race of immortal warriors and has no age, at least as humans commonly understand the concept. From there, we wondered whether the robot would do better at identifying people in a more straightforward scenario. So: another image search, this time for "good regular men walking in sunlight."
It is admittedly unusual that such a search would also return an image of Bartolo Colon, but this is the sort of "noise" that technologists deal with all the time. At any rate, we see here that the robot still struggles with age—if continues to nail the gender stuff—even in straightforward situations.
Advertisement
The next step was to test how well the robot worked with people in emotional extremis. We did an image search for "good normal happy man smiling thank you" and plugged the result into the robot.
Again, the gender is right on. And the age is closer! Where the robot struggled with Bartolo Colon, missing on his actual age by an infinite number of years, it overshot on Juan Uribe's age by a mere seven years.We knew we were getting someplace. The next step would be to test the other side of the emotional binary. We searched a comprehensive archive of images featuring sad people, using the parameters "the sports man is sad," and entered the result at How-Old.
ONE EMAIL. ONE STORY. EVERY WEEK. SIGN UP FOR THE VICE NEWSLETTER.
By signing up, you agree to the Terms of Use and Privacy Policy & to receive electronic communications from Vice Media Group, which may include marketing promotions, advertisements and sponsored content.
Or you don't. The technology, as Microsoft admits, is still being perfected. In the interest of helping this process along, our VICE Sports technologists selected photos at random to see if we could figure out where this new application falls short.We began with a question: is it harder for the How-Old robot to determine the age of people who are in motion? We tested this by doing a random image search using the parameters "graceful athlete in motion."
It is admittedly unusual that such a search would also return an image of Bartolo Colon, but this is the sort of "noise" that technologists deal with all the time. At any rate, we see here that the robot still struggles with age—if continues to nail the gender stuff—even in straightforward situations.The next step was to test how well the robot worked with people in emotional extremis. We did an image search for "good normal happy man smiling thank you" and plugged the result into the robot.
Again, the gender is right on. And the age is closer! Where the robot struggled with Bartolo Colon, missing on his actual age by an infinite number of years, it overshot on Juan Uribe's age by a mere seven years.We knew we were getting someplace. The next step would be to test the other side of the emotional binary. We searched a comprehensive archive of images featuring sad people, using the parameters "the sports man is sad," and entered the result at How-Old.
Honestly, it was close enough.