State-dependent error correlations shape voting thresholds in committees of AI agents
2026-07-27 • Computers and Society
Computers and Society
AI summaryⓘ
The authors studied how groups of AI models vote on decisions, focusing on how their mistakes might be related instead of independent. They found that when AI models tend to make similar errors, it limits the benefits of majority voting. By modeling these shared errors more accurately using a statistical approach, they could better predict the group's errors and choose smarter decision thresholds, reducing overall mistakes. Their method showed a clear improvement compared to assuming each AI voted independently.
AI committeemajority votingerror dependenceGaussian copulaSah-Stiglitz screeningbinary screeningthreshold selectionbootstrap confidence intervalexchangeable modelcost-sensitive decision
Authors
Haifeng Li, Mo Hai
Abstract
The aggregation benefit of a committee of artificial intelligence (AI) agents comes from complementary information across members. Classical voting guarantees assume independent errors. Language-model errors often co-occur on the same cases. We combine Sah-Stiglitz screening with error dependence that can differ between good and bad cases. In a homogeneous exchangeable Gaussian-copula model, shared errors create a positive asymptotic error floor for majority voting and can change the approval threshold that minimizes expected loss. We estimate a heterogeneous extension from 174,384 votes cast by 28 language models on four binary-screening benchmarks. Parameters estimated from odd-indexed items predicted committee loss on even-indexed items. For the sampled committee composition, the full-matrix dependence model increased identity-line R^2 from 0.840 under independence to 0.967. In a design-balanced analysis, cost-sensitive threshold selection under independence reduced scaled loss from 60.25 for majority to 52.50. Modeling dependence reduced it further to 50.77, an incremental improvement of 1.73 units (95% bootstrap CI, 0.68-2.33). The overall reduction from majority was 15.73% (95% bootstrap CI, 13.41-16.75%).