Twitter will reportedly be taking an even deeper look at the “unintentional harms” caused by its algorithms. The recently-announced initiative, dubbed the Responsible Machine Learning Initiative, has one primary goal. Namely, to look closely at those algorithms driving AI decision-making across the service. Specifically, as that pertains to “racial subgroups” and gender.
What is the Responsible Machine Learning Initiative and how will it stop AI bias in Twitter algorithms?
Now, Twitter is no stranger to bias. In fact, neither is effectively any other company working with AI algorithm but that’s what this Twitter initiative hopes to address. The company has previously admitted to racial bias, for example, in its image cropping algorithm. And it has tried to fix those kinds of errors where it finds them. But the problems are still there on that front, according to some reports. With the algorithms favoring lighter-skinned users.
And, in fact, that’s one of the first problems the company lists. Through its Responsible Machine Learning Initiative, Twitter hopes to take a closer look at and analyze the cropping system. That, according to Twitter, could also have some issues based on gender. And those will be looked at as well.
It will also perform a “fairness assessment” with regard to its Home timeline recommendations, specifically as that applies to racial subgroups.
Beyond race and gender, Twitter additionally hopes to analyze content recommendations based on political ideologies. Those will be performed across seven different countries.
How will you know when this initiative starts having an impact?
Of course, none of the analysis or examination would be noteworthy if Twitter didn’t plan to do anything with the information. And, to that end, Twitter wants to apply whatever it learns to improve the site and app. For example, Twitter says that the findings may result in the removal of algorithms. Effectively, giving users more control over the images they choose to post on the site. Or it could translate to new standards for building and designing the AI algorithms altogether.
But, regardless of the route it takes, it also wants to educate its userbase. In particular, not only about how algorithms work and what “informs them” but also how they impact what users actually see when they visit or use Twitter.
Moreover, the company plans to share all of its findings and to request feedback from users. That will, the company says, help it generate best practices for the industry and help hold the company accountable.