Twitter’s Transparency Theater

Since agreeing to buy Twitter a year ago, Elon Musk has said that increasing transparency on the platform is one of his highest priorities, promising to share more information about content moderation decisions and open source Twitter’s algorithm. 

The reality has been very different, however. Musk fired the teams in charge of transparency, shut down the platform’s research API, and changed Twitter’s verification process, making it harder to determine who is real and who isn’t.

In late March, Musk finally made good on his promise to open source the algorithm. But, when combined with his other actions, specifically around the API, this latest move amounts to little more than transparency theater.

What can we learn from Twitter’s algorithm? In theory, releasing the algorithm would allow the public to see the complicated calculus deciding what appears on our feeds. In reality, Twitter released an incomplete version of the code. 

“You can’t learn much from this release in and of itself,” wrote CSMaP's Sol Messing in an analysis. “You need the underlying model features, parameters, and data to really understand the algorithm."

That’s true for the core recommendation system governing the ‘For You’ feed, which figures out what tweets to show users and how to rank them, along with the systems that govern policy violations, Messing explained on Stanford Law School’s Moderated Content podcast.

“The model that drives the most important part of the algorithm has not been open-sourced,” he told Engadget — only the code that says how to fit the model was made public. “It has the flavor of transparency. But it doesn’t really give insight into what the algorithm is doing or why someone’s tweets may be down-ranked while others might be up-ranked.”

It’s a distraction from closing the Twitter API. For years, Twitter provided free access to its API, which enabled thousands of academics, journalists, and civil society actors to study important societal issues. At CSMaP, for example, we’ve used Twitter data to evaluate how bots impact conversations online and how foreign influence campaigns operate during elections.

But Twitter will soon charge for API access, with plans starting at an astounding $42,000 per month for access to 50 million tweets. (By contrast, researchers could previously collect more than 3 million tweets per day for free.)

“We cannot pay that amount of money,” Senior Research Engineer Megan A. Brown explained to Politico, noting that half of CSMaP’s dozens of projects rely on Twitter data. “I don’t know of a research center or university that can or would pay that amount of money.”

The change will jeopardize hundreds of public interest research projects, according to the Coalition for Independent Technology Research (CSMaP is a member of the Coalition), including research into the spread of harmful content, (dis)information flows, public health, political behavior, and more.

It will also make it harder to examine the algorithm Twitter just released under the guise of transparency. “At the same time Twitter is releasing this code, it’s made it incredibly difficult for researchers to audit this code,” Messing explained.