Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CogniPair: From LLM Chatbots to Conscious AI Agents -- GNWT-Based Multi-Agent Digital Twins for Social Pairing -- Dating & Hiring Applications

Wanghao Ye, Sihan Chen, Yiting Wang, Shwai He, Bowei Tian, Guoheng Sun, Ziyi Wang, Ziyao Wang, Yexiao He, Zheyu Shen, Meng Liu, Yuning Zhang, Meng Feng, Yang Wang, Siyuan Peng, Yilong Dai, Zhenle Duan, Lang Xiong, Joshua Liu, Hanzhang Qin, Ang Li

TL;DR

The study tackles the inadequacy of current LLM agents to emulate authentic human psychology and social dynamics by implementing Global Workspace Theory (GNWT) in a modular cognitive architecture (Emotion, Memory, Planning, SocialNorms, GoalTracking) and coordinating these via a global workspace. It introduces CogniPair, a social-influence decision system enabling large-scale GNWT-Agents to simulate dating and hiring interactions, initialized with adventure-based personality assessments to capture authentic traits. Validation on 551 GNWT-Agents and the Columbia Speed Dating dataset reports strong correlations with human attraction patterns, high match-prediction accuracy, and substantial human validation for behavioral fidelity, suggesting the approach bridges psychological and social realism gaps. The framework supports diverse applications in intelligent dating platforms and HR tech, while outlining future enhancements in cultural generalization, non-verbal cues, and efficiency to enable broader deployment.

Abstract

Current large language model (LLM) agents lack authentic human psychological processes necessary for genuine digital twins and social AI applications. To address this limitation, we present a computational implementation of Global Workspace Theory (GNWT) that integrates human cognitive architecture principles into LLM agents, creating specialized sub-agents for emotion, memory, social norms, planning, and goal-tracking coordinated through a global workspace mechanism. However, authentic digital twins require accurate personality initialization. We therefore develop a novel adventure-based personality test that evaluates true personality through behavioral choices within interactive scenarios, bypassing self-presentation bias found in traditional assessments. Building on these innovations, our CogniPair platform enables digital twins to engage in realistic simulated dating interactions and job interviews before real encounters, providing bidirectional cultural fit assessment for both romantic compatibility and workplace matching. Validation using 551 GNWT-Agents and Columbia University Speed Dating dataset demonstrates 72% correlation with human attraction patterns, 77.8% match prediction accuracy, and 74% agreement in human validation studies. This work advances psychological authenticity in LLM agents and establishes a foundation for intelligent dating platforms and HR technology solutions.

CogniPair: From LLM Chatbots to Conscious AI Agents -- GNWT-Based Multi-Agent Digital Twins for Social Pairing -- Dating & Hiring Applications

TL;DR

The study tackles the inadequacy of current LLM agents to emulate authentic human psychology and social dynamics by implementing Global Workspace Theory (GNWT) in a modular cognitive architecture (Emotion, Memory, Planning, SocialNorms, GoalTracking) and coordinating these via a global workspace. It introduces CogniPair, a social-influence decision system enabling large-scale GNWT-Agents to simulate dating and hiring interactions, initialized with adventure-based personality assessments to capture authentic traits. Validation on 551 GNWT-Agents and the Columbia Speed Dating dataset reports strong correlations with human attraction patterns, high match-prediction accuracy, and substantial human validation for behavioral fidelity, suggesting the approach bridges psychological and social realism gaps. The framework supports diverse applications in intelligent dating platforms and HR tech, while outlining future enhancements in cultural generalization, non-verbal cues, and efficiency to enable broader deployment.

Abstract

Current large language model (LLM) agents lack authentic human psychological processes necessary for genuine digital twins and social AI applications. To address this limitation, we present a computational implementation of Global Workspace Theory (GNWT) that integrates human cognitive architecture principles into LLM agents, creating specialized sub-agents for emotion, memory, social norms, planning, and goal-tracking coordinated through a global workspace mechanism. However, authentic digital twins require accurate personality initialization. We therefore develop a novel adventure-based personality test that evaluates true personality through behavioral choices within interactive scenarios, bypassing self-presentation bias found in traditional assessments. Building on these innovations, our CogniPair platform enables digital twins to engage in realistic simulated dating interactions and job interviews before real encounters, providing bidirectional cultural fit assessment for both romantic compatibility and workplace matching. Validation using 551 GNWT-Agents and Columbia University Speed Dating dataset demonstrates 72% correlation with human attraction patterns, 77.8% match prediction accuracy, and 74% agreement in human validation studies. This work advances psychological authenticity in LLM agents and establishes a foundation for intelligent dating platforms and HR technology solutions.

Paper Structure

This paper contains 71 sections, 20 equations, 9 figures, 11 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. CogniPair: Cognitive Social Pairing Agent System
  4. GNWT-Agent Cognitive Modules
  5. Single Decision-Making System Flow
  6. Simulated Social Environment Setup
  7. Generalized Environment Parameterization
  8. Multi-Agent Interaction Architecture
  9. Experiments and Results
  10. Experimental Setup
  11. Results and Key Findings
  12. Evolution of Social Dynamics
  13. Human Validation Studies
  14. Limitations and Future Work
  15. Broader Impacts & Conclusion
  16. ...and 56 more sections

Figures (9)

  • Figure 1: Comparison of Cognitive Architectures: Traditional RL/Chatbot (left), GNWT Agent Architecture (center), and GNWT Human Model (right), showing the evolution from linear processing to brain-inspired parallel processing with global workspace integration.
  • Figure 2: CogniPair Platform System Flow
  • Figure 3: Social Dynamics Evolution: Human vs. GNWT-Agent Comparison. (A) Partner preference changes from T1 to T2; (B) Self-perception adjustments across attributes; (C) Evolution of attribute-liking correlations from initial to post-interaction evaluations.
  • Figure 4: Comparison of GNWT-Agent with baseline methods across evolution dimensions. (A) Evolution accuracy showing GNWT-Agent's superior performance in all metrics; (B) Human-agent correlation demonstrating stronger alignment with human data compared to baseline approaches.
  • Figure 5: GNWT-Agent Architecture Internal Processing Flow
  • ...and 4 more figures