A relevance model of human sparse communication in cooperation

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Decision Theory

Background:

• Human communication is limited by real-time information flow, necessitating condensed data transfer.
• Existing models may not fully capture the nuanced selection of information in spontaneous communication.

Purpose of the Study:

• To introduce and evaluate a novel model of information selection in human communication based on "relevance."
• To compare the model's performance against advanced AI (GPT-4) and a heuristic model in a human-AI collaborative task.

Main Methods:

• Developed a "relevance" model integrating decision-making theory and Theory of Mind (ToM).
• Conducted simulated navigation experiments with human participants and an AI agent assisting each other to avoid traps.
• Evaluated model accuracy in predicting human information choices and its impact on collaborative task performance.

Main Results:

• The relevance model accurately predicted human choices for communicating trap information.
• The relevance model outperformed GPT-4 in the simulated navigation task.
• AI agents using the relevance model significantly improved human performance and received higher ratings compared to a heuristic model.

Conclusions:

• A relevance model grounded in decision theory and ToM effectively explains the sparse and spontaneous nature of human communication.
• This model offers a framework for developing more intuitive and effective AI communication strategies.
• The findings highlight the potential of integrating cognitive principles into AI for enhanced human-AI interaction.