Reproduce “Reaching a Consensus in Predictive Loops” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiduan Wu, Rediet Abebe, Celestine Mendler-Dünner
Reaching a Consensus in Predictive Loops models online prediction systems as active participants in social opinion dynamics rather than passive observers. Combining Friedkin–Johnsen peer influence with performative retraining, the analysis shows that repeated prediction can reduce opinion variance and approach consensus, create indirect influence across disconnected parts of a network, and amplify targeted interventions; simulations on Pokec and Yelp networks support these patterns.
Key insights
- Increasing platform susceptibility generally lowers equilibrium opinion variance while preserving the population mean in regular graphs; as susceptibility approaches one, the system can reach consensus even when ordinary FJ dynamics would preserve disagreement.
- Model-based aggregation acts as a virtual social connection: observed users can influence unobserved or peer-stubborn users through the platform’s predictions, without direct network paths.
- Repeated interventions can propagate beyond their initial targets and become stronger through the interaction between platform feedback and peer influence; consensus therefore does not imply accuracy or social benefit.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.