Talking nets: a multiagent connectionist approach to communication and trust between individuals

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Social Psychology

Background:

• Individual agents process information via recurrent networks.
• Inter-agent communication relies on belief propagation and trust mechanisms.
• Existing models often lack nuanced representations of trust in information exchange.

Purpose of the Study:

• To propose a multiagent connectionist model simulating belief updating through trust-based communication.
• To explore how trust weights influence information filtering and selective propagation.
• To investigate the impact of communicative mechanisms on social phenomena like persuasion and rumor spreading.

Main Methods:

• Developed a network of recurrent networks where agents communicate and update beliefs.
• Implemented trust as a dynamic variable influencing connection weights (trust weights) between agents.
• Utilized agent-based simulations to model phenomena such as persuasive communication, lexical acquisition, and information cascades.

Main Results:

• Trust weights effectively filter less reliable information, demonstrating selective information propagation.
• The model successfully simulated phenomena including persuasive communication, polarization, and the spread of stereotypes and rumors.
• Demonstrated how communicative mechanisms, beyond individual processing, shape group decision-making and information diffusion.

Conclusions:

• The proposed multiagent connectionist model provides a framework for understanding trust-based communication in artificial and biological systems.
• Trust mechanisms are crucial for filtering information and can explain various social communication dynamics.
• The model highlights the importance of inter-agent communication and trust in shaping collective behavior and belief formation.