‘Machine Learning Is Hard': Google Photos Has Egregious Facial Recognition Error

Google Photos director Anil Sabharwal announces Google Photos during the 2015 Google I/O conference on May 28, 2015, in San Francisco.
Getty Images

In May, Google unfurled its new Photos app as the crowning achievement of machine learning capabilities — a service that stores and catalogues your images with computing smarts that can pick out buildings, landscapes, animals, even abstract events like birthdays, on its own.

As users get their hands on the app, though, it's evident Photos is far from perfect. Two days ago, Jacky Alciné, an African-American programmer based in New York, flagged a flagrant error on Photos: It had tagged him and his friend as "gorillas."

Read more from re/code:
Xiaomi continues international push
Facebook changes the way it counts video views
The economic, moral toll of our immigration bureaucracy

In a tweet containing profanity, Alciné posted a series of photos with Google Photos' descriptions. One of them described Alciné and his friend as "gorillas."

To Google's credit, its human ambassadors responded swiftly. Yonatan Zunger, an engineer and "chief architect" of Google+, the social service from which Photos was stripped, replied to Alciné on Twitter within roughly 90 minutes, noting that he had alerted the Photos team.

In another tweet containing profanity, Zunger said "This is 100% Not OK."

Zunger checked in with Twitter missives the following day; Alciné thanked him, and noted that the erroneous label was removed.

Google has advanced more on artificial intelligence than others, infusing the advanced tech into speech and photo recognition as well as natural language processing. (Its trippy neural network construction has yet to be baked into consumer products.) The company is frank that its machine learning abilities still have a way to go. In his exchange on Twitter, Zunger noted how the company was still working on "long-term fixes" for linguistics and image recognition for "dark-skinned faces."

Read More Why this fund management boss is eyeing Google

Google put out this conciliatory statement: "We're appalled and genuinely sorry that this happened. We are taking immediate action to prevent this type of result from appearing. There is still clearly a lot of work to do with automatic image labeling, and we're looking at how we can prevent these types of mistakes from happening in the future."

Zunger, for his part, put it more bluntly.


By Mark Bergen, Re/

CNBC's parent NBCUniversal is an investor in Re/code's parent Revere Digital, and the companies have a content-sharing arrangement.