How do recommender systems learn political opinions? A semi-synthetic step-by-step experiment
Tim Faverjon, Jean-Philippe Cointet, Pedro Ramaciotti · 2026
Read How do recommender systems learn political opinions? A semi-synthetic step-by-step experiment by Tim Faverjon, Jean-Philippe Cointet, Pedro Ramaciotti (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0349341.txt.
Description
Tim Faverjon, Jean-Philippe Cointet, and Pedro Ramaciotti ran a controlled experiment on how recommendation algorithms quietly learn users' politics. Working from real sharing data for nearly forty thousand users of the platform X, they trained a recommender on synthetic versions of the data where each step could be inspected. Their 2026 study asks whether basic recommendation logic alone manufactures an ideological map of users.
It does: elementary principles trained on what people share arranged users along a left-right line inside the system, independently of age or gender. The team then edited those internal positions and watched the offered content shift in leaning, diversity, and relevance. The authors argue the political geometry is an emergent byproduct of ordinary recommendation, with consequences for segregated information diets.
Catalog record
| CREATOR | Tim Faverjon, Jean-Philippe Cointet, Pedro Ramaciotti |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65846 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |