Facebook CTO says hiring matters for mitigating AI bias, but the company lacks AI research diversity stats

Facebook CTO Mike Schroepfer endorses the idea that hiring is an important part of diversity in AI and preventing bias for teams building products for users, but he can’t tell you the number of Black people who work at Facebook AI Research. Created with Yann LeCun in 2013, Facebook AI Research has locations in Silicon Valley, New York, and Paris. With more than 100 employees, FAIR has become one of the largest and most influential AI research organizations in the world.

A Facebook AI spokesperson subsequently said Facebook has reported employee diversity numbers for six years but does not tally diversity statistics by individual teams. VentureBeat asked Facebook in July 2019 about the number of Black employees at Facebook AI Research and received no response. In November 2019, VentureBeat asked Google and Facebook (again) about diversity stats and was told by both companies that they do not supply AI research division diversity numbers.

On a separate but potentially related note, Snap CEO Evan Spiegel last Friday said in an internal meeting that he does not intend to release diversity statistics because they will only reinforce the perception that Silicon Valley is a less than diverse place. After this story was published, a Snap spokesperson contacted VentureBeat to say the company plans to release its own version of a diversity report in the “near term” but declined to share a specific time frame.

Schroepfer spoke publicly with AI journalists last week for the first time since protests against White supremacy set off by the police killing of George Floyd erupted in more than 2,000 cities across the U.S. and in major cities around the world. The subject of bias models came up after a reporter asked whether Facebook assessed winning models from the Deepfake Detection challenge for algorithmic bias based on skin tone.

“Look, I think that representation is really important … I’d say we care a lot about those [issues], which is why we’ve focused a lot on improving diversity across the board in the company,” Schroepfer said. “I also think that the real solution to these problems for things like making sure you have a diverse data set is actually the process, understanding of formalizing this across the company, so there are statistical methods to determine whether this data set is representative in the ways you care about.”

A 2019 analysis by Algorithmic Accountability Act coauthor Mutale Nkonde found that Facebook AI Research had no Black employees. Proposed in April 2019, the Algorithmic Accountability Act would require corporations to assess AI for safety, security, and bias.

Above: Facebook 2019 Diversity Report statistics on technical staff by ethnicity

In response, Schroepfer said bias is generally found in AI because training data fails to be representative of users. However, policymakers in Washington, AI ethics researchers, and many others stress that hiring diverse, pluralistic teams is essential for making AI with more people in mind.

Prior to the killing of George Floyd, diversity efforts at companies like Facebook and Google had made only incremental progress. In the wake of Floyd’s death, government and business leaders are being challenged to fight White supremacy and institutional racism. But many critics and members of the Black tech community are frustrated by the tech industry’s lack of progress despite years of public diversity reports at companies like Facebook and Google, and historic underfunding of startups with Black founders. Human Utility founder Tiffani Ashley Bell coined the phrase “Make the hire. Send the wire.” to succinctly answer the question of how venture capitalists and tech giants can make a real difference.

In recent weeks, Facebook CEO Mark Zuckerberg defended President Trump’s right to post the phrase “when the looting starts, the shooting starts,” which is associated with bigotry and suppression of civil rights protests dating back to the 1960s. Twitter censured a tweet with the same message for violating company policy and labeled it as “glorifying violence.” Several senior Facebook staff members have threatened to resign over Zuckerberg’s stance, and employees staged a virtual walkout on June 1. In recent days, multiple news outlets reported that Facebook fired an engineer who participated in the protest.

Current and former Facebook employees had negative things to say about the company before the death of George Floyd. In November 2018, former Facebook employee Mark Luckie asserted that “Facebook is failing its Black employees and its Black users.” A year later, a Medium post by an anonymous group of a dozen current and former Black employees of Facebook went viral with accounts of repeated failures by company executives to address racial issues in the workplace.

But Schroepfer defended Facebook’s record on diversity, emphasizing the company’s efforts to build the Society in AI and Responsible Innovation labs over the past two years. In a 2019 F8 keynote address, he highlighted Facebook’s increased use of AI for content moderation and the work of Responsible Innovation teams working with product teams on subjects like election integrity, security, and algorithmic fairness. Among highlights of the algorithmic fairness team’s work, Schroepfer said last year one Black employee “took it upon herself” to ensure pose estimation AI used for Portal’s camera works as well for people with light skin as it does for people with dark skin.

“I think that the important thing is to build tools that work, that people use by default, so it’s representative by default. But obviously, you know, the more representative the team is, the better we can make sure that all the perspectives of our users are incorporated in the products we build,” he said.

VentureBeat asked a Facebook spokesperson for more details about the activities of the algorithmic fairness team and whether there were specific initiatives within the company to audit algorithms Facebook uses in production, but did not receive a response at the time this story was published.

An AI Now Institute report on diversity at major tech companies like Facebook and Google released last year found that more than 80% of computer science academics teaching AI are men, with women making up only 15% of AI researchers at Facebook and 10% of AI researchers at Google. The report insists that companies that do not make concerted efforts to create diverse teams can perpetuate structural inequality that exists today.

Updated 1:12 p.m. June 16 to include comment on diversity statistics from a Snap spokesperson.