Levels of AI Agents: from Rules to Large Language Models

Yu Huang Roboraction.AI Abstract: AI agents are defined as artificial entities to perceive the environment, make decisions and take actions. Inspired by the 6 levels of autonomous driving by SAE (Society of Automotive Engineers), the AI agents are also categorized based on utilities and strongness, as the following levels: L0—no AI, with tools (with perception) plus actions; L1— use rule-based AI; L2—let rule-based AI replaced by IL/RL-based AI, with additional reasoning & decision making; L3—apply LLM-based AI instead of IL/RL-based AI, additionally setting up memory & reflection; L4—based on L3, facilitating autonomous learning & generalization; L5—based on L4, appending personality (emotion + character) and collaborative behavior (multi-agents). 1 Introduction Any entity, that is able to perceive its environment and execute actions, can be regarded as an agent. Agents can be categorized into five types: Simple Reflex agents, Model-based Reflex agents, Goal-based agents, Utilitybased agents, and Learning agents [1]. As AI advanced, the term “agent” is used to depict entities exhibiting intelligent behavior and possessing capabilities like autonomy, reactivity, pro-activeness, and social interactions. In the 1950s, Alan Turing proposed the renowned Turing Test [2].

Introduction. 2 LLMs LLMs [4] are the category of Transformer-based language models that are characterized by having an enormous number of parameters, typically numbering in the hundreds of billions or even more. These models are trained on massive text datasets, enabling them to understand natural language and perform a wide range of complex tasks, primarily through text generation and comprehension. Some well-known examples of LLMs include GPT- 3/4, PaLM, OPT, and LLaMA1/2. Extensive research has shown that scaling can largely improve the model capacity of LLMs. Thus, it is useful to establish a quantitative approach to characterizing the scaling effect. There are two representative scaling laws for Transformer language models: one from OpenAI[7], another from Google DeepMind[8]. The “pre-train+fine-tune” procedure is replaced by another procedure called “pre-train+prompt+predict” [9].

Discussion / Conclusion. In this paper, levels of AI agents are categorized based on utilities and strongness, similar to automation levels of autonomous driving by SAE. For each level, the additional modules from the previous level could provide stronger AI capabilities and agent utilities. From level 0 to level 3, the AI core has evolved from no AI, to rulebased AI, IL/RL based AI to LLM-based AI. References [1] S J Russell and P Norvig. Artificial Intelligence: A Modern Approach (4th Edition). Pearson Education, 2010. [2] Turing, A. M. Computing machinery and intelligence. Springer, 2009. [3] R Bommasani et al., On the Opportunities and Risks of Foundation Models, arXiv 2108.07258, 2021 [4] W X Zhao, K Zhou, J Li, et al., A Survey of Large Language Models, arXiv 2303.18223, 2023 [5] Wooldridge, M. J., N. R. Jennings. Intelligent agents: theory and practice. Knowl. Eng. Rev., 10(2):115–152, 1995. [6] Bisk, Y., A. Holtzman, J. Thomason, et al. Experience grounds language, arXiv 2004.10151, 2020.