Instagram has provided an update on the progress of its new Equity Team, which was established in the aftermath of last year’s #BlackLivesMatter protests in the United States with the stated goal of addressing systemic bias within Instagram’s internal and external processes.
Following the police shooting death of George Floyd, Instagram CEO Adam Mosseri promised to do more to address inequity faced by people from marginalized communities. This work would include a review of all of Instagram’s practices, products, and policies, according to Mosseri, in order to identify problems and improve its systems.
Since then, the Equity Team has been concentrating on a few key aspects of the Instagram experience.
As explained by Instagram:
“Early work here includes extensive research with different subsets and intersections of the Black community to make sure we understand and serve its diversity. We’ve spoken with creators, activists, policy minds and everyday people to unpack the diversity of experiences people have when using the platform. We are also in the process of auditing the technology that powers our automated enforcement, recommendations and ranking to better understand the changes necessary to help ensure people do not feel marginalized on our platform.”
Any algorithm that is based on user activity is likely to reflect some level of bias relative to that input, and gorithmic bias is a key component. As a result, Instagram has been focusing on educating its staff who work on its systems about how this could affect their processes.
“Over the last few months, the Equity team launched an internal program to help employees responsible for building new products and technologies factor in equity at every step of their work. The program, called the Equitable Product Program, was created to help teams consider what changes, big and small, they can make to have a positive impact on marginalized communities.”
Instagram has also implemented new Machine Learning Model Cards as part of this effort, which provide checklists to ensure that new machine learning systems are designed with equity in mind.
“Model cards work similar to a questionnaire, and make sure teams stop to consider any ramifications their new models may have before they’re implemented, to reduce the potential for algorithmic bias. Model cards pose a series of equity-oriented questions and considerations to help reduce the potential for unintended impacts on specific communities, and they allow us to remedy any impact before we launch new technology. As an example, ahead of the US election, we put temporary measures in place to make it harder for people to come across misinformation or violent content, and our teams used model cards to ensure appropriate ML models were used to help protect the election, while also ensuring our enforcement was fair and did not have disproportionate impact on any one community.”
Again, this is a critical component of any platform’s overall equity efforts; if your algorithm’s inputs are inherently flawed, the outcomes will be as well. That means social media platforms can help eliminate bias by removing it from algorithmic recommendations and exposing users to a wider range of content.
Concerns about “shadowbanning” and users believing their content has been restricted within the app have also been addressed by the Equity Team.
Any algorithm that is based on user activity is likely to reflect some level of bias in relation to that input, and gorithmic bias is a key component. As a result, Instagram has been concentrating on informing its employees who work on its systems about how this may affect their processes.
Instagram has also implemented new Machine Learning Model Cards as part of this effort, which serve as checklists for ensuring that new machine learning systems are built with equity in mind.
“This includes tools to provide more transparency around any restrictions on a person’s account or if their reach is being limited, as well as actions they can take to remediate. We also plan to build direct in-app communication to inform people when bugs and technical issues may be impacting their content. In the coming months, we’ll share more details on these new features.”
Increased transparency, which makes it completely clear why certain posts are getting less reach and whether any limitations have been imposed, could solve a variety of problems, not just those affecting marginalized communities.
This is a key area of development for Instagram and Facebook in general, particularly in relation to machine learning and algorithmic models that are based on current user behavior, as previously mentioned.
If social platforms can identify key areas of bias within these systems, it could be a significant step toward addressing long-standing concerns, and it could also play a key role in reducing systemic bias more broadly.
In the future, Instagram says it will launch new initiatives to help amplify Black-owned businesses.
Source: Social Media Today