Emergent Intelligence in Multi-Agent and LLM Systems: A Survey and Perspective Toward Autonomous, Collaborative, and Generalizable AI

This survey examines the emerging intelligent behaviors of today’s Artificial Intelligence (AI) systems, including Large Language Models (LLMs) and multi-agent systems (MASs) that operate without explicit programming. Additionally, the study explores the basic, often loosely guided, interactions among autonomous agents that can lead to unexpectedly coordinated effects and task-driven dynamics. AutoGen is one framework that highlights AI agents' ability to assume various roles in allocation, communication, and cooperative problem-solving. An organized taxonomy of emerging intelligent behaviors is presented, covering collaborative planning, tool-assisted reasoning, and iterative self-improvement over time. The study also analyzes MAS designs, including their architectural features, shared memory and information pathways, and agent modularity. Furthermore, a key focus of the survey is the broader impact of these advancements, including Artificial General Intelligence (AGI), which transforms human and AI collaboration, and raises important questions about ethics, alignment, and responsible deployment.