2015

When Google launched its AI-powered image recognition tool, expectations were sky-high. The system was a remarkable demonstration of machine learning that could be used by everyone in everyday life. It automatically identified objects, scenes, and even people in the uploaded photos. Ask it to find “bridges,” and it would quickly find photos of bridges from your library. Ask it to find “dogs,” and it would proudly show you every pup you’d ever snapped. Great - right?

But then came one of AI’s most infamous failures. The system confidently labeled photos of a black man and his friend as “gorillas.” This was unsurprisingly, very veeery offensive and Google’s awesome feature was relentlessly criticized.

Google reacted swiftly, issuing public apologies and promising to fix the issue. But instead of confronting the root cause, the biased and incomplete training data that taught the system to see skin tone as an indicator of species, they took a simpler, more expedient route. They deleted the problem. Literally.

Rather than retraining the algorithm or rethinking its data, Google simply blocked the words “gorilla,” “chimpanzee,” “monkey,” and all related tags from the system entirely. The world’s most advanced image recognition AI became, in one fell swoop, blind to an entire species. You could search for “bridge,” “dog,” or “sunset,” but if you wanted to find photos of actual gorillas, not possible.

Nearly a decade later, that temporary fix remains in place. Even with enormous leaps in AI technology, Google Photos still refuses to label images of gorillas. The problem wasn’t solved it was silenced. Apple learned from Google’s mistake and simply copied their fix.

Here is a totally non-AI image of a gorilla.