Skip to main content

Google Photos’ ‘racist’ error highlights facial recognition’s limits

google apologizes for misidentifying a black couple as gorillas in photos app racist error
Credit: Jacky Alcine/Twitter, BBC
As impressive as facial recognition software is, it isn’t perfect, and sometimes algorithm errors can lead to unintended consequences. Google discovered this in what’s possibly one of the worst scenarios a company could face: Its new Photos app incorrectly labeled a black couple as “gorillas.”

Google Photos has the ability to automatically tag images, using photo comparisons, big data, and other methods the intelligence software learns over time. But its accuracy is not always successful. While mistakes are unintentional, Google is facing backlash for this particular error’s racist connotations, the BBC reports.

According to Jacky Alcine, the man in the photo who brought it to Google’s attention, Photos didn’t misclassify just one photo but an entire collection that contained him and his friend. Alcine, a software developer, made the issue public by posting the results on Twitter.

Google responded immediately with an apology, telling the BBC that it’s “appalled and genuinely sorry that this happened.” While the company wasn’t being malicious, it says the error is “100 percent not OK.”

Google executive Yonatan Zunger says, “[It was] high on my list of bugs you ‘never’ want to see happen,” and that the company is “working on longer-term fixes around both linguistics – words to be careful about in photos of people – and image recognition itself – e.g., better recognition of dark-skinned faces.”

It’s not the first time – and won’t be the last – that Google Photos mistakenly tagged photos. In May, iTech Post noted that dogs were being classified as horses. In our photos, we noticed the software tagged an image of a camera and an airplane under “cars.” But part of Photos’ machine learning is to understand its mistakes, studying how users remove or reclassify tags. Over time, Google Photos’ accuracy should improve.

“There is still clearly a lot of work to do with automatic image labelling, and we’re looking at how we can prevent these types of mistakes from happening in the future,” Google tells the BBC.

Editors' Recommendations

Enid Burns
Enid Burns is a freelance writer who has covered consumer electronics, online advertising, mobile, technology electronic…
This invisible photo filter protects you from facial recognition software
facial recognition gyfcat race fbi face mugshot 970x644g

D-ID Smart Anonymization

Machines are getting increasingly impressive at recognizing things -- most of the time. Occasionally, some reminder comes along which showcases how something that's instantly recognizable to humans can, with a few tweaks, be rendered totally baffling to a computer. Capitalizing on this, an Israeli company has developed some smart software that promises to help crack down on facial recognition technology being used against people in a way they may not be happy with.

Read more
Google contractors were reportedly told to target the homeless for facial scans
Facial recognition software showing faces and the amount of times they've appeared.

Recently, Google was reportedly developing a facial recognition feature for the Pixel 4 that would rival the iPhone's. At the time, reports indicated that Google contractors were simply roaming the streets and offering $5 to people in exchange for a facial scan.

Now, new details have come to light about the practice. According to the New York Daily News, Apple contractors were directed to "target homeless people" because they would be less likely to say anything to the media. These people were reportedly not told anything about what they were doing, simply that they would get $5 gift cards in exchange for a facial scan. Some were even lied to -- being told that the scan was a "selfie game" similar to Snapchat.

Read more
This A.I.-powered app can spot skin cancer with 95 percent accuracy
skinvision app detects cancer 95 percent  38

Detect skin cancer with your phone using SkinVision

If you’re going to trust your smartphone to do something as important as detect the most common forms of skin cancer, you want to be pretty darn sure that it works as advertised. Fortunately, the claims of an app called SkinVision appear to be accurate.

Read more