Facebook’s nudity-spotting AI mistook a photo of some onions for ‘sexually suggestive’ content



a group of oranges in a pile: Getty


© Getty
Getty

  • A Canadian garden center had its Facebook ad for onion seeds taken down by the platform on Monday.
  • Facebook said the ad was removed for breaking its rules on “products with overtly sexual positioning.”
  • Facebook’s head of comms in Canada said the post had been restored on Wednesday, and that it had been initially removed by the platform’s automated moderation systems.
  • “We use automated technology to keep nudity off our apps. But sometimes it doesn’t know a walla walla onion from a, well, you know.”
  • Visit Business Insider’s homepage for more stories.

Facebook’s AI struggles to tell the difference between sexual pictures of the human body and globular vegetables.

A garden center in Newfoundland, Canada on Monday received a notice from Facebook about an ad it had uploaded for Walla Walla onion seeds that contained a photo of some onions.

Facebook’s notice said the ad broke its rules on “products with overtly sexual positioning,” clarifying: “listings may not position products or services in a sexually suggestive manner.”



a screenshot of fruit and vegetable stand: The Seed Company NL, Facebook.


© The Seed Company NL, Facebook.
The Seed Company NL, Facebook.

Facebook on Wednesday told Canada’s CBC News the ad had been reinstated after review. The mistake had been made by its AI moderation tech, which automatically takes down content it thinks contains nudity, it said.

“We use automated technology to keep nudity off our apps. But sometimes it doesn’t know a walla walla onion from a, well, you know. We restored the ad and are sorry for the business’ trouble,” Meg Sinclair, Facebook Canada’s head of communications told CBC.

She did not clarify what she meant by a “you know.”

This is not the first time Facebook’s automated systems have over-zealously removed content later reinstated by human moderators. In 2018 its systems took down a post containing excerpts from

Facebook’s alleged indifference of Indian hate speech linked to policy chief’s political bias

Over the past several weeks, there has been an increasing clamour for Facebook to place its India public policy head, Ankhi Das, on leave as the company continues with an audit of its India operations.

The impetus for the audit was an article written by the Wall Street Journal in mid-August. In that piece, WSJ reported that Das had resisted against taking down inflammatory content that eventually sparked rioting in the capital city of Delhi as it was posted by members of the nationalist BJP party. 

The riots left over fifty dead, most of whom were Muslims. It also led to many of these Muslims’ homes being torched.

“The company’s top public-policy executive in the country, Ankhi Das, opposed applying the hate-speech rules to [T Raja] Singh and at least three other Hindu nationalist individuals and groups flagged internally for promoting or participating in violence,” WSJ reported.

These inflammatory posts were reportedly only taken down months after the riots had already occured, and only when the paper approached the company for a statement. 

One of the BJP politicians, Raja Singh, reportedly said that Rohingya Muslim refugees should be “shot”, and had labelled Indian Muslims as traitors while also threatening to destroy their mosques. Singh, who has enshrined a reputation for these kinds of comments, has since denied these allegations and claimed his account was hacked.

The audit was initiated when a group of 54 retired civil servants wrote to Facebook CEO Mark Zuckerberg about the WSJ revelation. This call for an audit was then reiterated by a jointly-written letter to Facebook by global civil rights organisations such as the Southern Law Poverty Center, Muslim Advocates, and other organisations in countries such as the UK, US, and New Zealand.

“The audit must be removed entirely from the influence of the India

U.S. House Antitrust Chairman Calls Unwinding Facebook’s Instagram Buy ‘The Right Answer’ | Technology News

WASHINGTON (Reuters) – U.S. Representative David Cicilline, the chairman of the House Judiciary Committee’s antitrust subcommittee, said on Wednesday he would be “comfortable with unwinding” Facebook Inc’s acquisition of Instagram.

The antitrust subcommittee on Tuesday released a report on Big Tech’s abuses of market power but stopped short of naming specific companies or acquisitions that must be broken up.

Cicilline, a Democrat from Rhode Island, told Reuters in an interview that Facebook should not have been allowed to buy Instagram, a deal that the Federal Trade Commission approved in 2012.

“I would be comfortable with unwinding that. I think that’s the right answer,” he said.

Facebook did not immediately respond to a request for comment. It has said previously that Instagram was insignificant at the time it was purchased and that Facebook built it into the success it has become.

Any effort to unwind the deal would entail the government filing a lawsuit and asking a judge to order the divestiture.

The congressional report released on Tuesday said that Instagram was small at the time it was purchased, but that Facebook CEO Mark Zuckerberg saw its potential and noted it was “building networks that are competitive with our own” and “could be very disruptive to us.”

According to the House panel’s report on Tuesday, the committee received an email from an unnamed former Instagram employee on Sunday that disputed Facebook’s contention that the two apps could not easily be separated.

“They can just roll back the changes they’ve been making over the past year and you’d have two different apps again,” the person wrote. “It’s turning something on and off.”

(Reporting by Nandita Bose and Diane Bartz in Washington; Editing by Chris Reese and Matthew Lewis)

Copyright 2020 Thomson Reuters.

Source Article

Facebook’s Dilemma, Subpoena Time, and Twitter Fails

Illustration for article titled Facebooks Dilemma, Subpoena Time, and Twitter: There Was an Attempt

Image: Exposure Labs/Netflix

HellfeedHellfeedHellfeed is your bimonthly resource for news on the current heading of the social media garbage barge.

It’s time for Hellfeed, your biweekly summary of who’s kicking and screaming online, and we really couldn’t have picked a better day: The president tweeted that he is ill with the novel coronavirus this morning and then totally dropped off the goddamn radar.

While the nation waits for updates on the president’s condition—the White House has quietly upgraded the severity of his coronavirus infection from “mild” to “very moderate”—it is simultaneously hooting, claiming it’s a hoax, preparing for chaos, demanding civility, and generally not having any idea what the hell is going to happen next. This really caps off two weeks of everything breaking, which we’ve summarized below.

Facebook hits back at The Social Dilemma

The Social Dilemma, Netflix’s documentary on the deliberate design choices behind the online attention economy, is sort of a hot mess. It certainly raises good points about the business model and manipulative nature of companies like Facebook, but it also posits social media as the cause of rather than an amplifying factor in the state of society and doesn’t meaningfully interrogate how all of this is shaped by factors like capital, class, and politics. (It’s perhaps best summed up by a ridiculous Reefer Madness-style frame narrative, featuring a teen who is radicalized into some kind of nebulous “Extreme Center” extremist organization after staring at his phone for a few days.)

That aside, The Social Dilemma has earned largely glowing reviews and a flurry of attention because of… well, everything that’s going on right now. Facebook itself has now responded with a point-by-point list of what it says the film got wrong, complete with helpful pointers like

Facebook’s AI matches people in need with those willing to assist

Facebook says it has deployed a feature in its Community Help hub to make it easier for users to assist each other during the pandemic. As of this week, AI will detect when a public post on News Feed is about needing or offering help and will surface a suggestion to share it on Community Help. Once a post is moved or published directly to the hub, an algorithm will recommend matches between people.

For example, if someone posts an offer to deliver groceries, they’ll see recommendations within Community Help to connect with people who recently posted about needing this type of assistance. Similarly, if someone requests masks, AI will surface suggested neighbors who recently posted an offer to make face coverings.

Building this Community Help feature, which Facebook says is available in all countries in English and 17 other languages, involved a difficult engineering challenge because the system needs to make recommendations even when semantic structures in posts are very different. (For example, consider “Does anyone have masks for kids?” and “We can donate face coverings of any size.”) The feature also needs to go beyond existing candidate-matching logic to incorporate general statements like “I can lend a hand to anyone!”

Facebook Community Help AI

Facebook says it built and deployed the matching algorithm using XLM-R, its natural language understanding model that produces a score ranking how closely a request for help matches offers in a community. XLM-R, which has 550 million parameters (variables internal to the model that fine-tune its predictions), was trained on 2.5 terabytes of webpages and can perform translations among roughly a hundred different human languages.

The system integrates posts’ score into a set of models trained on PyText, an open source framework for natural language processing. People needing or offering help receive matches through an overlay that suggests