When AI Influence Helps or Hurts: Guarantees for Ranking-Based Human–AI Aggregation

Jonas KargeJonas Karge, Roy FergusonRoy Ferguson, Daniel GrimaldiDaniel Grimaldi, Jonas HaldimannJonas Haldimann, Ruvarashe MadzimeRuvarashe Madzime, Thomas MeyerThomas Meyer

Download

Jonas Karge, Roy Ferguson, Daniel Grimaldi, Jonas Haldimann, Ruvarashe Madzime, Thomas Meyer
When AI Influence Helps or Hurts: Guarantees for Ranking-Based Human–AI Aggregation
The 17th International Conference on Scalable Uncertainty Management, to appear

Details
Bibtex

KurzfassungAbstract
We present, to the best of our knowledge, the first formal model of ranking-based human–AI aggregation. In our model, both human and AI judgments are represented by ranking functions, and we study how AI advice affects the epistemic quality of a collective decision. To obtain probabilistic guarantees, we introduce a top-choice abstraction that translates ranking aggregation into a plurality-based voting problem. This allows us to model a specific interaction setting that we call AI-follow assistance with concurrent panel revision. Building on this connection, we derive new probabilistic guarantees tailored to our setting. These results provide a formal account of when AI influence can be epistemically harmful by inducing dependence among human agents and when it can be epistemically beneficial by improving the probability that the true world is selected collectively.
Projekt:Project: SECAI
Forschungsgruppe:Research Group: Computational LogicComputational Logic

@article{KFGHMM2026,
  author  = {Jonas Karge and Roy Ferguson and Daniel Grimaldi and Jonas
             Haldimann and Ruvarashe Madzime and Thomas Meyer},
  title   = {When {AI} Influence Helps or Hurts: Guarantees for Ranking-Based
             {Human{\textendash}AI} Aggregation},
  journal = {The 17th International Conference on Scalable Uncertainty
             Management},
  year    = {2026}
}