Political Alignment in Recommendations
Experimental study of recommender alignment as a public-good dilemma.
This project studies political alignment in recommender systems as a public-good problem. The core idea is that steering recommendations toward a preferred content balance requires costly user input and platform effort, while many of the benefits are shared across users. That creates the central dilemma of the project: aligned recommendations may be collectively desirable but individually underprovided because users can free-ride on others’ contributions.
The project is joint work with Dietmar Jannach, Silvia Milano, Caterina Giannetti, Cecilia Vergari, Nicola Meccheri, and Marco Catola. The collaboration brings together recommender-systems research, AI ethics, behavioural and experimental economics, and applied microeconomics.
The empirical component is an incentivized online experiment implemented in oTree and SurveyJS. Participants first report political and non-political preferences, rank fictional movies to establish a ground truth, and then interact with a shared recommender through repeated movie-rating decisions. Ratings are privately costly but improve recommendation quality for both members of a matched pair, allowing the design to measure willingness to contribute, free-riding, and how contribution changes when partners are politically similar or opposed.
The treatment design crosses domain and match type: political versus non-political disagreement, and homogeneous versus heterogeneous matches. A pilot on Prolific generated usable variation in contribution behaviour and highlights the main identification challenge: separating strategic response from beliefs, framing, and instruction-induced priming.
On the implementation side, the platform combines structured preference elicitation, treatment-based matching, repeated contribution decisions, and transparent payoff rules tied to recommendation accuracy. It is built as reusable infrastructure for research on recommender governance, multi-user AI systems, and the behavioural conditions under which people support politically constrained algorithmic decisions.
Experiment interface
