The Google Photos appeared in late May with many attractions, including the ability to recognize the context of the pictures to categorize them. Everyone knows what kind of technologies (still) are not perfect yet, no one expected that the function would put Google in a tight spot: the service ended scoring two people like gorillas.
The failure was noted by Jacky Alcine, a programmer who lives in New York. By giving your photos in Google Photos, he realized that pictures showing him with a friend were automatically tagged by the service as “Gorillas”.
Immediately, Alcine expressed its indignation on Twitter. In subsequent tweets, the programmer made ??it clear he understood that the problem was caused by a failure (after all, it works with development), but questioned what types of images the Google system would be using as a benchmark for the mistake. “This could have been avoided with more accurate classification and full of black people, especially people with darker skin tones like me and my friend,” he added.
GOOGLE PHOTOS, Y’ALL FUCKED UP. MY FRIEND’S NOT THE GORILLA. PIC.TWITTER.COM/SMKMCSNVX4
– DIRI NOIR AVEC BANAN (@JACKYALCINE) zolpidem JUNE 29, 2015
The case, of course, won rebound quickly. As you might expect, some people tried to lift a racial debate, stating, for example, that this type of problem does not happen to white people.
About an hour and a half after the first tweet, the chief architect of Google+ Yonatan Zunger apologized for the failure and asked permission to Alcine for an analysis depth to your account in order to investigate the causes.
AND IT’S ONLY PHOTOS I HAVE WITH HER ??IT’S DOING THIS WITH (RESULTS TRUNCATED B / C PERSONAL):PIC.TWITTER.COM/H7MTXD3WGO
– DIRI NOIR AVEC BANAN (@JACKYALCINE) JUNE 29, 2016
On the issue raised, Zunger said that learning of artificial intelligence systems is difficult, explaining that following the Google Photos have made mistakes with fair-skinned people, marking them as dogs and seals, for example.
The tag “Gorillas” has been removed from the Google Photos database. But Google knows that this is not enough. “There is clearly much work to be done with automatic image labeling and we are studying how we can avoid this kind of mistake in the future,” said a company spokesman.