What is algorithmic transparency?

algorithmic transparency (noun)

As applied to digital platforms and other companies, the idea that researchers should be able to study the algorithms that make decisions and recommendations.

Social media platforms are now a primary way that news and information — and, sometimes, misinformation — are spread. Algorithms are an integral part of this information pipeline, powering recommendation engines, content rankings, and news feeds.

Some researchers and users would like to analyze how these algorithms work and the extent of their effects, but most social media companies treat their algorithms as classified secrets. They often remain inscrutable “black boxes,” even for employees with access to the algorithms and the ability to run A/B tests on them. That’s a concern, given the significant impact platforms like Facebook, Twitter/X, and YouTube have on society, according to a panel moderated by MIT Sloan associate professor of marketing Dean Eckles at the 2022 Social Media Summit at MIT.

Eckles said one way to create algorithmic transparency is to provide a way for social media companies to share internal data with researchers, journalists, and other interested parties while still protecting user privacy. New techniques like differential privacy, which is used by government agencies and some tech companies, could make this more feasible.

Simply gaining access to algorithms isn’t sufficient to address concerns, Eckles noted — his own research has shown how hard it is to quantify the impacts of algorithmic rankings, such as bias and harm.

From MIT Social Media Summit, 5 trends experts are watching