MAPS: Multi-Agent Personality Shaping for Collaborative Reasoning

TL;DR

MAPS introduces a multi-agent framework where five personality-driven agents (Interpreter, Aligner, Scholar, Solver, and Critic) collaborate on multimodal reasoning, with the Critic providing Socratic feedback to enable reflective revision. Grounded in the Big Five personality theory, the approach yields a four-stage reasoning pipeline that is iteratively refined through a feedback loop, backed by a formal information-bottleneck and free-energy perspective. Empirical results on MathVista, OlympiadBench, and EMMA show state-of-the-art performance, surpassing human experts and generalizing across diverse base models such as GPT-4o, Qwen, Gemini, and others. The study demonstrates that structured personality-driven collaboration coupled with reflective critique enhances reasoning depth, accuracy, and robustness in complex multimodal tasks, offering a scalable blueprint for advanced AI problem solving.

Abstract

Collaborative reasoning with multiple agents offers the potential for more robust and diverse problem-solving. However, existing approaches often suffer from homogeneous agent behaviors and lack of reflective and rethinking capabilities. We propose Multi-Agent Personality Shaping (MAPS), a novel framework that enhances reasoning through agent diversity and internal critique. Inspired by the Big Five personality theory, MAPS assigns distinct personality traits to individual agents, shaping their reasoning styles and promoting heterogeneous collaboration. To enable deeper and more adaptive reasoning, MAPS introduces a Critic agent that reflects on intermediate outputs, revisits flawed steps, and guides iterative refinement. This integration of personality-driven agent design and structured collaboration improves both reasoning depth and flexibility. Empirical evaluations across three benchmarks demonstrate the strong performance of MAPS, with further analysis confirming its generalizability across different large language models and validating the benefits of multi-agent collaboration.

Figures (10)

• Figure 1: An example of a multimodal scientific multiple-choice problem. The correct answer is derived based on the reasoning over inputs that include context, question, and diagram.
• Figure 2: Comparison of reasoning paradigms. Single-agent and two-agent approaches offer limited adaptability. MAPS enables dynamic collaborative reasoning. Right: Built on GPT-4o, MAPS achieves the best performance across three benchmarks.
• Figure 3: The corresponding relation between the Big Five Personality theory and the five function-specific agents.
• Figure 4: The overall architecture of MAPS. The framework consists of five functional agents inspired by the Big Five personality theory. The core reasoning process is carried out by four specialized agents—Interpreter, Aligner, Scholar, and Solver—each responsible for a distinct stage in solving complex reasoning problems. Finally, the Critic agent provides reflective feedback and correction to enhance accuracy and interpretability.
• Figure 5: The schema of the Critic agent, as well as the feedback and backtracking situations of the Critic agent across different datasets.