The Illusion of Intelligence: AI, Interaction, and the Clever Hans Effect
This paper explores the nature of intelligence in AI, particularly large language models (LLMs). It argues that the apparent intelligence of LLMs isn't due to independent reasoning but rather emerges from interaction with users. Drawing parallels between Socratic questioning, the Clever Hans effect, and iterative prompting of LLMs, the author demonstrates that intelligence is a relational phenomenon arising from collaboration, not isolated cognition. LLMs generate responses based on probabilistic relationships within their training data, responding to user prompts like Clever Hans responded to his handler's cues. The value of AI, therefore, lies not in its inherent 'knowledge' but in its ability to facilitate insightful questions and collaborative exploration, ultimately augmenting human creativity and problem-solving.
Read more