Assessing Consciousness-Related Behaviors in Large Language Models Using the Maze Test

2025-08-22

AI Summary

This study investigates whether large language models (LLMs) exhibit behaviors associated with consciousness—specifically spatial awareness, perspective-taking, goal-directedness, and temporal consistency. We propose the first behaviorally grounded evaluation framework, grounded in 13 theoretical accounts of consciousness, and design a first-person maze navigation task to systematically assess 12 state-of-the-art LLMs under zero-shot, one-shot, and few-shot settings. Our key contribution lies in operationalizing abstract consciousness theories into quantifiable behavioral metrics and introducing the first subjective-perspective evaluation of LLMs’ self-consistency and narrative coherence. Results show that reasoning-augmented models substantially outperform baselines: Gemini 2.0 Pro achieves 52.9% full-path accuracy, while DeepSeek-R1 attains 80.5% partial-path accuracy. However, all models fail to sustain stable self-representation across trials, revealing a fundamental limitation in unified, temporally persistent self-awareness.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

We investigate consciousness-like behaviors in Large Language Models (LLMs) using the Maze Test, challenging models to navigate mazes from a first-person perspective. This test simultaneously probes spatial awareness, perspective-taking, goal-directed behavior, and temporal sequencing-key consciousness-associated characteristics. After synthesizing consciousness theories into 13 essential characteristics, we evaluated 12 leading LLMs across zero-shot, one-shot, and few-shot learning scenarios. Results showed reasoning-capable LLMs consistently outperforming standard versions, with Gemini 2.0 Pro achieving 52.9% Complete Path Accuracy and DeepSeek-R1 reaching 80.5% Partial Path Accuracy. The gap between these metrics indicates LLMs struggle to maintain coherent self-models throughout solutions -- a fundamental consciousness aspect. While LLMs show progress in consciousness-related behaviors through reasoning mechanisms, they lack the integrated, persistent self-awareness characteristic of consciousness.

Problem

• Assessing consciousness-like behaviors in large language models
• Evaluating spatial awareness and perspective-taking via maze navigation
• Testing integrated self-awareness and coherent self-models in AI

Innovation

• Using Maze Test to evaluate consciousness-like behaviors
• Synthesizing consciousness theories into 13 measurable characteristics
• Testing LLMs across zero-shot one-shot and few-shot scenarios