More than a yr after its first civil rights audit, Meta says it’s nonetheless engaged on numerous adjustments advisable by auditors. The firm launched detailing its progress on addressing the auditors’ many suggestions.

According to the corporate, it has already carried out 65 of the 117 suggestions, with one other 42 listed as ”in progress or ongoing.” However, there are six areas the place the corporate says it’s nonetheless figuring out the “feasibility” of constructing adjustments and two suggestions the place the corporate has “declined” to take additional motion. And, notably, a few of these take care of probably the most contentious points referred to as out within the authentic 2020 audit.

That authentic report, launched in July of 2020, discovered the corporate wanted to do extra to cease “pushing users toward extremist echo chambers.” It additionally mentioned the corporate wanted to handle points associated to algorithmic bias, and criticized the corporate’s dealing with of Donald Trump’s posts. In its , Meta says it nonetheless hasn’t dedicated to all of the adjustments the auditors referred to as for associated to algorithmic bias. The firm has carried out some adjustments, like partaking with exterior consultants and growing the range of its AI group, however says different adjustments are nonetheless “under evaluation.”

Specifically, the auditors referred to as for a compulsory, company-wide course of for “to avoid, identify, and address potential sources of bias and discriminatory outcomes when developing or deploying AI and machine learning models” and that it “regularly test existing algorithms and machine-learning models.” Meta mentioned the advice is “under evaluation.” Likewise, the audit additionally advisable “mandatory training on understanding and mitigating sources of bias and discrimination in AI for all teams building algorithms and machine-learning models.” That suggestion can be listed as “under evaluation,” based on Meta.

The firm additionally says some updates associated to content material moderation are additionally “under evaluation.” These embody a advice to enhance the “transparency and consistency” of selections associated to moderation appeals, and a advice that the corporate research extra features of how hate speech spreads, and the way it can use that information to handle focused hate extra rapidly. The auditors additionally advisable that Meta “disclose additional data” about which customers are being focused with voter suppression on its platform. That advice can be “under evaluation.”

The solely two suggestions that Meta outright declined had been additionally associated to elections and census insurance policies. “The Auditors recommended that all user-generated reports of voter interference be routed to content reviewers to make a determination on whether the content violates our policies, and that an appeals option be added for reported voter interference content,” Meta wrote. But the corporate mentioned it opted to not make these adjustments as a result of it might decelerate the evaluation course of, and since “the vast majority of content reported as voter interference does not violate the company’s policies.”

Separately, Meta additionally mentioned it’s a “a framework for studying our platforms and identifying opportunities to increase fairness when it comes to race in the United States.” To accomplish this, the corporate will conduct “off-platform surveys” and analyze its personal information utilizing surnames and zip codes.