Enhancing Collective AI Decision Robustness through Personality Diversity

As Artificial Intelligence (AI) evolves into a critical societal infrastructure, ensuring its decision-making robustness is paramount. AI collectives, however, are susceptible to systemic failures analogous to human "groupthink," which often manifests as "group polarization." This paper introduces "Engineered Cognitive Diversity," a novel strategy to mitigate these risks by systematically imbuing AI agents with diverse personality traits. Using a multi-agent simulation for high-stakes corporate decision-making, we demonstrate that while homogeneous AI teams exhibit severe groupthink symptoms and consistently polarize, heterogeneous teams foster healthier deliberation. This diversity significantly suppresses groupthink dynamics (e.g., reduces Self-Censorship and Illusion of Unanimity) and, consequently, reduces the Group Polarization Index (GPI), driving a 95.6% reduction under the OCEAN framework and 41.2% under the MBTI framework. This establishes that engineering personality diversity, by fostering constructive conflict, is a key design principle for building resilient AI systems.