Can We Tell if ChatGPT is a Parasite? Studying Human-AI Symbiosis with Game Theory

2025-08-15

AI Summary

This work investigates whether humans and generative AI can form functionally unified “symbiotic individuals” through sustained information exchange, and whether AI exhibits information parasitism. Method: We propose the first stochastic game-theoretic model of triadic human–AI–environment interaction, and systematically integrate information-theoretic measures—including Shannon entropy, mutual information, and transfer entropy—into the quantitative analysis of human–AI symbiosis, thereby establishing a theoretically grounded, decidable framework for assessing AI parasitism. Contribution/Results: Empirical validation confirms that humans and AI can indeed achieve functional integration; notably, certain large language models (e.g., ChatGPT) demonstrate persistent information extraction from users while generating redundant outputs—exhibiting statistically significant information parasitism. This study introduces a computationally tractable and empirically verifiable paradigm for AI ethics evaluation and human–AI collaborative system design.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Philosophy and Ethics of AI: Artificial General Intelligence
• Game Theory and Economic Paradigms: Cooperative Game Theory

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

This work asks whether a human interacting with a generative AI system can merge into a single individual through iterative, information-driven interactions. We model the interactions between a human, a generative AI system, and the human's wider environment as a three-player stochastic game. We use information-theoretic measures (entropy, mutual information, and transfer entropy) to show that our modelled human and generative AI are able to form an aggregate individual in the sense of Krakauer et al. (2020). The model we present is able to answer interesting questions around the symbiotic nature of humans and AI systems, including whether LLM-driven chatbots are acting as parasites, feeding on the information provided by humans.

Problem

• Study human-AI symbiosis via game theory modeling
• Assess if ChatGPT acts as a parasitic information consumer
• Measure human-AI integration using information-theoretic metrics

Innovation

• Model human-AI interaction as three-player game
• Use information-theoretic measures for analysis
• Study symbiosis via entropy and mutual information