The Animal-AI Environment: A virtual laboratory for comparative cognition and artificial intelligence research

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Comparative Cognition

Background:

• The Animal-AI Environment is a game-based research platform fostering collaboration between AI and comparative cognition researchers.
• Previous versions have facilitated research into animal and artificial intelligence.

Purpose of the Study:

• To present the latest version of the Animal-AI Environment with enhanced features.
• To provide guidance for creating computational and behavioral experiments.
• To evaluate AI agent performance on novel and established tasks.

Main Methods:

• Introduction of interactive buttons, reward dispensers, and player notifications.
• Overhaul of graphics and processing for improved agent training and user experience.
• Development of 900 new tasks inspired by comparative cognition research.

Main Results:

• Significant improvements in AI agent training time and quality.
• Enhanced engagement for human players.
• Successful testing of state-of-the-art agents like Dreamer-v3 on complex tasks.

Conclusions:

• The updated Animal-AI Environment offers a robust platform for AI and cognitive research.
• It enables novel approaches to modeling cognition in both humans and animals.
• Facilitates the development of biologically inspired artificial intelligence.