Independence among judgments can improve aggregation, while social influence can create confident convergence without improved accuracy.

Aggregation can reduce error when judgments reflect partly independent information or reasoning paths; shared mistakes are not canceled by combining them. Interaction can improve reasoning by transmitting useful information, but it can also align estimates, evidence selection, or confidence before accuracy improves. The relevant variable is therefore effective error independence, not the number of contributors, agents, or model labels. This claim is conditional on a task whose judgments can meaningfully be compared or aggregated, and it does not imply that participants should remain isolated, that deliberation is generally harmful, or that independence alone ensures competence. Architecturally, it motivates preserving attributable first-pass judgments, tracking shared sources and models, measuring correlated errors, and comparing interactive review with independent and centralized baselines before claiming a plurality benefit.