Proceedings of the Annual Meeting of the Cognitive Science Society

Interaction can sometimes hinder human collective performance, while aggregating independent responses ("wisdom of crowds") is expected to improve performance through statistical facilitation. Yet some coordination studies show that interacting pairs can outperform their individual members and nominal groups (Bahrami et al., 2010; Szary & Dale, 2013, 2014). We present a reservoir computing model demonstrating how interaction can enhance joint performance. Two echo state networks were trained independently to identify Japanese speakers from vowel recordings and tested jointly with and without coupled feedback. The model reproduces the dyadic advantage reported by Bahrami et al. (2010) with interaction via pooled-feedback, but not with independent self-feedback. Model dynamics revealed that interaction was beneficial when uncoupled outputs were prone to diverging from runaway feedback over time. Our results offer a formal framework for explaining when and why interaction enhances collective intelligence.