Cognitive Anthropomorphism of AI: How Humans and Computers Classify Images

Humans wrongly assume AI "sees" like they do: People tend to anthropomorphize AI systems by expecting them to classify images using the same cognitive processes as human brains, but AI classifiers actually work through fundamentally different mechanisms—a gap that leads to confusion about why AI makes unexpected errors.

This mismatch creates real problems for human-AI interaction: When users' expectations about how AI thinks don't align with how it actually operates, it becomes an obstacle to appropriate and effective collaboration between humans and AI systems in practical applications.

Explainable AI can help bridge the gap: One key solution is making AI decision-making more transparent so users understand why a system classified something a particular way, rather than treating it as a "black box."

Users need different training, not just better AI: Rather than only improving algorithms, the paper suggests we should develop novel training methods to help people understand the actual nature of AI cognition and adjust their expectations accordingly.

AI algorithms could be redesigned to work more like human thinking: The third strategy involves developing new algorithms that more closely match human cognitive processes, potentially making AI behavior more intuitive and predictable to users.