This research examines how AI visual reasoning compares to humans' dual thinking processes (fast intuition vs. deliberate logic).

Key Insights:

• Introduces a novel adversarial dataset to test logical processing in both humans and AI
• Demonstrates the presence of dual thinking mechanisms in human visual reasoning
• Evaluates how multi-modal LLMs perform with intuitive versus logical visual tasks
• Highlights important security implications for AI deployed in safety-critical domains where logical reasoning is essential

This research is particularly relevant for security applications as it examines AI reliability in environments where logical processing is crucial for safe operation, highlighting current gaps between human and machine reasoning capabilities.

Dual Thinking and Logical Processing -- Are Multi-modal Large Language Models Closing the Gap with Human Vision ?