Facebook’s ad system shown failing to enforce its own anti-discriminatory policy

Can Facebook be trusted to abide by even its own stated standards? In the case of Internet political advertising the social giant wants to be allowed to continue to self regulate — despite the scandal of Russian bought socially divisive ads which (we now know) were tainting democratic discussion during the 2016 US presidential election (and beyond).

‘Don’t regulate us, we can regulate ourselves — honest!‘ is shaping up to be CEO Mark Zuckerberg’s massively moonshot new year project for 2018.

But results from a new ProPublica investigation suggest the tech giant is failing at even simple self-policing — undermining any claims it can responsibly manage the bad and even out-and-out illegal outcomes that are being enabled via its platform, and bolstering the case for more formal regulation.

Case in point: A year ago Facebook said it would disable ethnic affinity ad targeting for housing, employment and credit-related ads, following a ProPublica investigation that had suggested the platform’s ad-targeting capabilities could be used for discriminatory advertising — particularly in housing and employment, where such practices are illegal.

This month ProPublica checked in again, to see how Facebook is doing — by purchasing dozens of rental housing ads and asking that Facebook’s ad platform exclude groups that are protected from discrimination under the US Federal Fair Housing Act — such as African Americans and Jews.

Its test ads promoted a fictional apartment for rent, targeted at people aged 18 to 65 who were living in New York, house hunting and likely to move — with ProPublica narrowing the audience by excluding certain “Behaviors”, listed in a section Facebook now calls “Multicultural Affinity”, including “Hispanic”, “African American” and “Asian American”.

However instead of the platform blocking the potentially discriminatory ad buys, ProPublica reports that all its ads were approved by Facebook “within minutes” — including an ad that sought to exclude potential renters “interested in Islam, Sunni Islam and Shia Islam”. It says that ad took the longest to approve of all its buys (22 minutes) — but that all the rest were approved within three minutes.

It also successfully bought ads that it judged Facebook’s system should at least flag for self-certification because they were seeking to exclude other members of protected categories. But the platform just accepted housing ads blocked from being shown to categories including ‘soccer moms’, people interested in American sign language, gay men and people interested in wheelchair ramps.

Yet, back in February, Facebook announced new “stronger” anti-discriminatory ad polices, saying it was deploying machine learning tech tools to help it identify ads in the categories of concern.

“We’ve updated our policies to make our existing prohibition against discrimination even stronger. We make it clear that advertisers may not discriminate against people based on personal attributes such as race, ethnicity, color, national origin, religion, age, sex, sexual orientation, gender identity, family status, disability, medical or genetic condition,” it wrote then.

Of the new tech tools, Facebook said: “This will allow us to more quickly provide notices and educational information to advertisers — and more quickly respond to violations of our policy.”

Explaining how the new system would work, Facebook said advertisers who attempt to show “an ad that we identify as offering a housing, employment or credit opportunity” and which “either includes or excludes our multicultural advertising segments — which consist of people interested in seeing content related to the African American, Asian American and US Hispanic communities” will find the platform disapproves the ad.

The new system would also require all advertisers that attempt to buy targeted advertising in the categories of concern to self-certify they are complying with Facebook’s anti-discrimination policies and with “applicable anti-discrimination laws”.

ProPublica says it never even encountered these self-certification screens, as well as never having any of its ad buys blocked.

“Under its own policies, Facebook should have flagged these ads, and prevented the posting of some of them. Its failure to do so revives questions about whether the company is in compliance with federal fair housing rules, as well as about its ability and commitment to police discriminatory advertising on the world’s largest social network,” it writes.

Responding to ProPublica’s findings, Facebook sent a statement attributed to Ami Vora, VP of product management, in which she concedes its system failed in this instance. “This was a failure in our enforcement and we’re disappointed that we fell short of our commitments. The rental housing ads purchased by ProPublica should have but did not trigger the extra review and certifications we put in place due to a technical failure,” said Vora.

She went on to claim Facebook’s anti-discrimination system had “successfully flagged millions of ads” in the credit, employment and housing categories — but also said Facebook will now begin requiring self-certification for ads in all categories that choose to exclude an audience segment.

“Our systems continue to improve but we can do better,” she added.

The latter phrase is now a very familiar refrain from Facebook where content review and moderation is concerned. Aside from socially divisive political disinformation, it has faced growing criticism this year for enabling the spread of content such as extremist propaganda and child exploitation, as well as for multiple incidents of its tools being used to broadcast suicides and murders.

The wider question for governments and regulators is at what point will Facebook’s attempts to ‘do better’ be deemed just not good enough?