Parallel development of social behavior in biological and artificial fish

Area of Science:

• Artificial intelligence
• Computational neuroscience
• Ethology

Background:

• The reverse engineering paradigm, successful in understanding vision, is extended to social behavior.
• Embodied artificial neural networks (ANNs) offer a novel approach to studying social development.

Purpose of the Study:

• To investigate if generic learning algorithms can generate complex social behaviors in artificial agents.
• To explore the development of social preferences and collective action in embodied AI models.

Main Methods:

• Artificial neural networks were embodied in artificial fish within simulated environments.
• These artificial fish were trained using deep reinforcement learning and curiosity-driven rewards.
• The models were tested in both controlled virtual tanks and naturalistic ocean worlds.

Main Results:

• Artificial fish spontaneously developed fish-like social behaviors, including collective movement and in-group/out-group preferences.
• These behaviors generalized to more complex, naturalistic environments, demonstrating model robustness.
• The findings suggest that intrinsic motivation and reinforcement learning are key drivers of social behavior development.

Conclusions:

• Animal-like social behaviors can emerge from fundamental AI learning principles.
• Embodied AI models provide a powerful tool for reverse-engineering the developmental basis of social behavior.
• This research bridges the gap between sensory processing and collective action in artificial systems.