The present study investigates the presence of Differential Item Functioning (DIF) within the extraversion dimension of the Big Five personality traits, with particular attention to the potential influence of demographic variables. Rather than focusing on manifest DIF, this study explores latent DIF arising from population heterogeneity by employing a Mixture Item Response Theory (MixIRT) model, which identifies homogeneous subgroups that may exhibit DIF more clearly. To facilitate subgroup identification and improve the detection of latent DIF, we introduced a novel artificial intelligence algorithm, the Hybrid Ant Colony Optimization (hACO) algorithm. This algorithm was employed to select indicators and the number of subgroups automatically within the MixIRT model, enabling us to form subgroups that most clearly reflect DIF in the extraversion-related survey items. The use of the hACO algorithm improves subgroup formation precision and highlights differential responses that would otherwise remain undetected with traditional methods. To qualify the subgroup heterogeneity, we applied the Generalized Many‑Facet Rasch Model (GMFRM), a comprehensive tool for assessing item responses across multiple facets. The GMFRM accounts for the multifaceted nature of the population heterogeneity and provides a rigorous framework for examining DIF in personality measurement. We utilized the open-source dataset titled "Answers to Duckworth's Grit Scale (Duckworth et al., 2007) appended to the IPIP Big Five Scales" to evaluate the DIF of extraversion. This dataset includes responses to 62 items, comprising a 13-item vocabulary checklist and 20 demographic variables. The results of this study offer valuable insights into the fairness and bias of the psychological scale when applied to diverse populations. The integration of the MixIRT and GMFRM models, combined with the innovative use of hACO for variable selection, demonstrates a robust methodological approach for addressing latent DIF in psychological assessments.