Researchers Question Transparency of X Algorithm Code Release

X, the social media platform, recently released portions of the source code behind its 'For You' recommendation algorithm, claiming to promote unprecedented transparency among major platforms. Elon Musk, the owner of X, framed the move as a significant step forward for openness in social media: “At least you can see us struggle to make it better in real-time and with transparency.”

However, leading AI researchers and computer scientists have raised doubts about the usefulness of the disclosure. John Thickstun, assistant professor of computer science at Cornell University, characterized the publication as a “redacted” version that does not enable meaningful independent analysis. “What troubles me about these releases is that they give you a pretense that they’re being transparent... and the fact is that that’s not really possible at all,” Thickstun told Engadget.

After the code’s publication, users on X began sharing speculative guides on how to optimize their posts, suggesting strategies such as prioritizing conversation or video content. Yet researchers caution against drawing conclusions from the release, as the code omits substantial operational detail. Visible rules, such as filtering out posts older than one day, offer only a limited glimpse into the system’s workings.

A major structural change from the 2023 algorithm is that post ranking now hinges on a large language model, described as "Grok-like," instead of explicit, hard-coded interaction metrics. Previously, likes, shares, and replies had fixed values influencing rankings; now, the model predicts user engagement based on neural network outputs. Neural networks, a type of AI model loosely inspired by the brain, learn decision patterns from data but often lack transparency because of their complex internal structures.

This shift has increased the opacity of the system, even for the platform’s engineers. “More and more of the decision-making power... is shifting not just out of public view, but actually really out of view or understanding of even the internal engineers,” Thickstun commented. Details that were previously explicit—such as the exact weight of a reply compared to a retweet—are now absent, redacted for "security reasons."

Another significant omission is any information about the training data used for the model, a concern raised by Mohsen Foroughifar, assistant professor at Carnegie Mellon University. “If the data that is used for training this model is inherently biased, then the model might... end up still being biased, regardless of what kind of things that you consider within the model.” Without insight into training data, meaningful auditing for fairness or bias is severely limited.

For European researchers like Ruggero Lazzaroni at the University of Graz, who works on EU-funded simulations of social media recommendation systems, the release is insufficient. Lazzaroni notes that while the code to run the algorithm is available, the underlying trained model is not, making independent reproduction impossible.

The move by X is emblematic of larger challenges in AI and generative systems. Thickstun draws parallels between social recommendation algorithms and next-generation generative AI platforms, highlighting recurring transparency concerns.

Some researchers, including Lazzaroni, also warn that current platform priorities optimize user engagement, not user well-being or truthfulness, which could exacerbate negative social impacts.

User reactions on X show a strong appetite for decoding algorithmic systems, but experts emphasize that the current redacted release neither advances genuine transparency nor enables productive oversight. As the shift toward AI-based recommendation systems accelerates, the debate over transparency, auditability, and accountability is likely to grow.

Reference: dataconomy.com