Minority-size tipping point in LLM consensus formation

Determine whether increasing the size of a well-connected minority group in a Barabási–Albert network causes consensus to converge on the minority’s preferred colour over the majority’s, or whether LLM minorities remain unable to influence collective outcomes regardless of their size.

Background

The study examines whether a well-connected minority can influence consensus formation among LLM agents in the Network Consensus game. Unlike prior human experiments, the simulated LLM agents reached consensus only for the less-connected majority rather than for the well-connected minority. The authors note that their experiment used only five minority agents among thirty-six total agents, leaving unresolved whether minority size changes this outcome.

The open problem concerns a possible tipping point: a sufficiently large, well-connected minority might cause the collective consensus to shift toward its preferred colour, or LLM minorities might remain ineffective regardless of their numerical size. Resolving this would clarify whether minority influence in LLM collectives follows the patterns observed in human networks.

References

Therefore, future studies could investigate if there is a tipping point at which increasing the size of a minority group causes consensus to converge on their preferred colour over the majority, or whether LLM minorities are unable to influence the collective outcomes, regardless of size.

— Drivers of LLM Consensus Formation Under Competing Incentives: A Simulation Study  (2610.20200 - Jones et al., 14 Oct 2026) in Section 4, subsection “Comparison of Minority Influence vs. Balanced Preferences”