Search and Filter

Evolution In Simulation: Ai-Agent School With Dual Memory For High-Fidelity Educational Dynamics

Date

AI Overview

Why it matters

Educational AI simulations often lack systematic modeling of the teaching process, and agents struggle to simulate diverse participant behaviors and interactions accurately.

Study at a glance
  • Question: Can a dual-memory system help AI agents autonomously learn to simulate realistic teacher and student behaviors?
  • Setting and data: Simulated school environment with 50 AI agents (10 teachers, 40 students) across a 5-day schedule; evaluated by 9 Chinese university teachers.
  • Method: Multi-agent system development, automated text similarity comparisons across nine memory configurations, and blind evaluation by educational experts.
  • AI tools: AI-Agent School (AAS), QwQ-32B, Gemini-2.5-Pro, GPT-4o, Qwen3-235B-A22B, Qwen3-8B
Key findings
  • The full model combining dual knowledge and experience memory with short- and long-term structures achieved the highest text similarity scores to expert ground truth.
  • In human evaluations, expert preference for the full model's realism increased over time, reaching 44% to 46% and rivaling curated human ground truth.
  • Ablation comparisons showed that external memory, separated knowledge and experience bases, and short-term prioritization each improved agent response quality over unified or context-only setups.

Summary drafted with AI from the full paper and may contain errors. See the full paper for details.

Large language models (LLMs) based Agents are increasingly pivotal in simulating and understanding complex human systems and interactions. We propose the AI-Agent School (AAS) system, built around a self-evolving mechanism that leverages agents for simulating complex educational dynamics. Addressing the fragmented issues in teaching process modeling and the limitations of agents performance in simulating diverse educational participants, AAS constructs the Zero-Exp strategy, employs a continuous "experience-reflection-optimization" cycle, grounded in a dual memory base comprising experience and knowledge bases and incorporating short-term and long-term memory components. Through this mechanism, agents autonomously evolve via situated interactions within diverse simulated school scenarios. This evolution enables agents to more accurately model the nuanced, multi-faceted teacher-student engagements and underlying learning processes found in physical schools. Experiment confirms that AAS can effectively simulate intricate educational dynamics and is effective in fostering advanced agent cognitive abilities, providing a foundational stepping stone from the "Era of Experience" to the "Era of Simulation" by generating high-fidelity behavioral and interaction data.