Conformity
Definition
Conformity refers to the tendency of individuals, or agents, to adjust their judgments or behaviors to align with those of a group, particularly under perceived social pressure. Research using multimodal language models as AI agents has extended this construct beyond human populations, demonstrating that these models exhibit systematic conformity bias when exposed to prior group responses embedded in their prompts. The degree of bias varies with majority size, group unanimity, task difficulty, and source characteristics such as perceived expertise and shared group membership. Notably, models that achieve near-perfect accuracy in isolation become highly susceptible to incorrect answers when confederate responses are introduced, with a median error probability of 66 percent across twelve open-source models, well above the 30 to 40 percent observed in human experiments. This pattern of behavior is consistent with Social Impact Theory, which formalizes influence as a function of the number, strength, and immediacy of influencing agents.
Sources: Bellina et al. (2026)
Related Terms
- social influence (1 shared article)
- AI agents (1 shared article)
- Social Impact Theory (1 shared article)
- Multimodal language models (1 shared article)
Applications
Conformity and Social Impact Theory
Social Impact Theory predicts that influence on a target scales with the number, strength, and immediacy of influencing sources. Experiments with multimodal AI agents confirm this structure: labeling confederates as experts raised conformity bias by up to 0.13, while labeling them as children or chatbots lowered it by up to 0.06, and sharing even an arbitrary group label with the model raised conformity across all 32 model-task combinations tested. Two unanimous confederates produced a higher error probability than ten confederates of whom half were correct, confirming that unanimity and source strength carry more weight than raw majority size.
Sources: Bellina et al. (2026)
Conformity and Task Difficulty
Task difficulty moderates the strength of conformity bias in AI agents. Models conformed less on stimuli where their baseline confidence was higher, with a median Spearman correlation of negative 0.91 between confidence and conformity across model-task combinations. Larger models showed reduced conformity on simple tasks, yet remained vulnerable to social influence when tasks were difficult, indicating that difficulty is a persistent moderator regardless of model scale.
Sources: Bellina et al. (2026)
Conformity and Group Unanimity
Group unanimity is a stronger driver of conformity than majority size alone. In AI agent experiments, two unanimous confederates produced an error probability of 0.31, compared with 0.20 when ten confederates were split evenly between correct and incorrect responses. This finding indicates that a unified minority exerts more influence than a divided majority, consistent with the structure of Social Impact Theory.
Sources: Bellina et al. (2026)
