免费论文范文 人工智能驱动的智能体：当前趋势、应用及未来可能

AI-Driven Agents: Current Trends, Applications, and Future Possibilities

1. Introduction

The landscape of Artificial Intelligence (AI) is undergoing rapid transformation, with AI-driven agents emerging as pivotal components across diverse technological and societal domains. Intelligent agents, defined as autonomous entities capable of perceiving their environment, making informed decisions, and executing actions [13], are increasingly sophisticated due to advancements in Large Language Models (LLMs). This surge in capability and application underscores the growing importance of AI-driven agents in contemporary research and practical applications, necessitating a comprehensive survey to capture the current state and future directions of this dynamic field.

The rationale for this survey is grounded in the escalating impact of AI agents across numerous sectors. From revolutionizing customer service with neural agent assistants [24] and transforming data science through virtual interactive data scientists [42], to enhancing healthcare with medical dialogue systems [11] and facilitating information retrieval in assisted living [47], AI agents are demonstrating their versatility and potential. Their utility extends to complex tasks such as web navigation [19], code generation [3], and even simulating judicial rulings [49], highlighting a broad spectrum of applications that continue to expand.

The evolution of AI agents reflects a significant shift from rule-based systems to models leveraging the power of neural networks and, more recently, LLMs. Initially focused on simpler tasks, AI agents have progressively incorporated advanced natural language understanding, reasoning, and planning capabilities. The field has witnessed a move towards end-to-end systems for task-oriented dialogues [2,26], and an increasing emphasis on situated interaction [38], multi-agent collaboration [46,51], and personalized user experiences [28,41]. This progression is marked by a transition from modular approaches to more integrated, LLM-centric architectures, capable of handling complex, multi-step reasoning and interaction with external knowledge sources [5,39].

The field of AI-driven agents is characterized by its dynamism and continuous advancement. The ongoing research is not only expanding the application areas but also delving into fundamental challenges such as safety [4], evaluation methodologies [12,40], and the refinement of agent capabilities like reasoning, planning, and collaboration. This survey aims to provide a structured overview of this rapidly evolving landscape, offering insights into current trends, diverse applications, and potential future possibilities. By synthesizing the latest research, we seek to establish a comprehensive understanding of AI-driven agents and set the stage for future explorations in this significant domain.