clock menu more-arrow no yes

Filed under:

Google ‘fixed’ its racist algorithm by removing gorillas from its image-labeling tech

New, 51 comments

Nearly three years after the company was called out, it hasn’t gone beyond a quick workaround

The AI algorithms in Google Photos sort images by a number of categories.
Photo by Vjeran Pavic / The Verge

Back in 2015, software engineer Jacky Alciné pointed out that the image recognition algorithms in Google Photos were classifying his black friends as “gorillas.” Google said it was “appalled” at the mistake, apologized to Alciné, and promised to fix the problem. But, as a new report from Wired shows, nearly three years on and Google hasn’t really fixed anything. The company has simply blocked its image recognition algorithms from identifying gorillas altogether — preferring, presumably, to limit the service rather than risk another miscategorization.

Wired says it performed a number of tests on Google Photos’ algorithm, uploading tens of thousands of pictures of various primates to the service. Baboons, gibbons, and marmosets were all correctly identified, but gorillas and chimpanzees were not. The publication also found that Google had restricted its AI recognition in other racial categories. Searching for “black man” or “black woman,” for example, only returned pictures of people in black and white, sorted by gender but not race.

A spokesperson for Google confirmed to Wired that the image categories “gorilla,” “chimp,” “chimpanzee,” and “monkey” remained blocked on Google Photos after Alciné’s tweet in 2015. “Image labeling technology is still early and unfortunately it’s nowhere near perfect,” said the rep. The categories are still available on other Google services, though, including the Cloud Vision API it sells to other companies and Google Assistant.

It may seem strange that Google, a company that’s generally seen as the forerunner in commercial AI, was not able to come up with a more complete solution to this error. But it’s a good reminder of how difficult it can be to train AI software to be consistent and robust. Especially (as one might suppose happened in the case of the Google Photos mistake) when that software is not trained and tested by a diverse group of people.

It’s not clear in this case whether the Google Photos algorithm remains restricted in this way because Google couldn’t fix the problem, didn’t want to dedicate the resources to do so, or is simply showing an overabundance of caution. But it’s clear that incidents like this, which reveal the often insular Silicon Valley culture that has tasked itself with building world-spanning algorithms, need more than quick fixes.