Experts say open sourcing Twitter's algorithm will not increase transparency because recommendation engines are complex and use an immense fire hose of data
Conservatives have long claimed discrimination by the social network, but there's no one algorithm that could prove it, researchers say Tweets: @chrismessina , @kevinmarks , @peterwsinger , @nbashaw , and @hypervisible Tweets: @chrismessina : So @elonmusk wants to increase trust + transparency in Twitter by open-sourcing its algorithm; is it logically equivalent that open sourcing the full self-driving mode AI + algorithms would increase pubic trust in Tesla? https://www.washingtonpost.com/ ... /by @ReedAlbergotti Kevin Marks / @kevinmarks : @Techmeme @ReedAlbergotti When I was at Google, I would explain that open sourcing the algorithm would only help if you have a copy of the web in RAM and several thousand servers. Instead we built an api, OpenSocial, that let you run an app on the users feed and friends in context. FB hated it. Peter W. Singer / @peterwsinger : “But Musk's proposal probably represents a gross oversimplification of how it would work” https://www.washingtonpost.com/ ... A Pattern Nathan Baschez / @nbashaw : Cool to see the idea of ‘algorithm choice’ mentioned in WaPo! I agree with the basic premise of the article, which is that inspecting ML algorithms to understand how they work is devilishly hard. Transparency would have less impact than choice, imo. https://www.washingtonpost.com/ ... https://twitter.com/... @hypervisible : “The algorithm is not one thing” https://www.washingtonpost.com/ ... Expand More For Next Unexpand More For Next
Context & Ripple Effects
Elon Musk had pitched open-sourcing Twitter's algorithms as a trust-building measure aimed partly at conservatives who claim the network discriminates against them. Researchers quoted by the Washington Post push back: recommendation engines are complex systems fed by an immense fire hose of data, so there is no single artifact whose release could prove or disprove bias.
The argument aged well. When Twitter later published its For You ranking code on GitHub, researchers found the released code was missing the context needed to explain recommendations — and notably excluded the ad recommendation engine entirely. A parallel proposal from Every argued disclosure alone is insufficient and called for a marketplace where anyone can build and swap in their own algorithms.
First-order effects
- Musk's transparency pledge is undercut at the moment it is made: if experts are right, releasing the code cannot settle conservatives' discrimination claims because the behavior lives in the data pipeline, not the source.
Second-order effects
- Disclosure advocates pivot from 'show the code' to structural alternatives — Every's algorithm-marketplace idea lets users choose ranking logic rather than audit one black box, shifting the debate from inspection to user control.
Third-order effects
- If code releases keep failing to explain real-world feeds, platform-transparency pressure migrates toward data access and independent researcher access regimes — and regulators writing disclosure rules learn that publishing source is a weaker accountability tool than it appears.
The trend: Platform transparency is moving from code disclosure toward data and model-context access, as released source repeatedly proves unable to explain live recommendation behavior.