Dual-System Cognitive Swarm: A Rust-Based Multi-Agent System Mimicking Human Thinking

Project Overview

The dual-system-cognitive-swarm project, developed by CIAqingchen and hosted on GitHub, aims to build a multi-agent system that mimics human cognitive patterns using psychology's dual system theory. It leverages Rust to create agents with both System1 (fast, intuitive thinking) and System2 (slow, logical reasoning) capabilities, enabling adaptive and robust behavior in complex environments.

Key Information

• Author/Maintainer: CIAqingchen
• Source: GitHub (https://github.com/CIAqingchen/dual-system-cognitive-swarm)
• Release Date: 2026-06-08

Human thinking involves two distinct modes: System1 and System2, as proposed by psychologist Daniel Kahneman in Thinking, Fast and Slow.

• System1: Fast, automatic, low cognitive load (e.g., face recognition, simple decision-making). Prone to cognitive biases.
• System2: Slow, effortful, logical (e.g., complex math, strategic planning). More accurate but resource-intensive.

This project applies this theory to machine intelligence, aiming to replicate the collaboration of these two systems in multi-agent swarms.

The project is a Rust-based cognitive simulation framework. Rust was chosen for its memory safety, zero-cost abstraction, and async support.

System1 Agents: Fast heuristic decision-making using pre-trained neural networks and pattern recognition. They respond in milliseconds to environmental stimuli (e.g., obstacle avoidance, emergency risk avoidance) based on experience and intuition.

System2 Agents: Deep logical reasoning using symbolic engines and planning algorithms. They handle complex tasks like resource allocation and path planning, with higher decision quality but slower response times.

Subconscious Bias Modeling: The project explicitly models cognitive biases (e.g., confirmation bias, availability heuristic, anchor effect) based on psychology research. This helps study their impact on group behavior and explore correction strategies.

Dual-System Collaboration: Agents use a 'cognitive load' metric to switch dynamically: System2 for stable environments/long-term planning, System1 for sudden changes. System2 also reviews System1 decisions to correct errors, balancing speed and quality.

The swarm consists of many collaborative agents with dual-system capabilities. Coordination mechanisms include pheromone marking, local communication, and role division.

Emergent behaviors arise at the swarm level: some agents focus on fast exploration (System1), others on deep analysis (System2). This allows the swarm to handle both immediate responses and long-term strategies simultaneously.

Rust provides critical technical advantages:

• Memory Safety: Ownership system eliminates data races and memory leaks, essential for long-running simulations.
• Performance: Zero-cost abstraction allows high-level cognitive models without performance loss.
• Concurrency: Tokio runtime manages thousands of agents concurrently, enabling linear scalability.

Applications: Potential uses include robot teams (autonomous exploration), game NPCs (realistic behavior), social science simulations, and autonomous driving (balance between fast response and strategic planning).

Limitations: Challenges include optimizing dual-system coordination, refining bias modeling, and reducing computational complexity for large swarms.

Conclusion: This project represents a shift from single-function AI to human-like cognitive architectures, paving the way for more adaptive and flexible intelligent systems.