AutoAgents: A Framework for Automatic Agent Generation
Guangyao Chen, Siwei Dong, Yu Shu, Ge Zhang, Jaward Sesay, B\"orje F., Karlsson, Jie Fu, Yemin Shi
TL;DR
AutoAgents is a novel framework that dynamically generates and coordinates specialized AI agents for different tasks, improving multi-agent collaboration and solution quality over existing methods.
Contribution
It introduces an adaptive system that creates task-specific agents and incorporates an observer role, enhancing multi-agent task-solving capabilities.
Findings
AutoAgents outperforms existing multi-agent methods on various benchmarks.
The framework produces more coherent and accurate solutions.
Dynamic agent generation improves adaptability to different tasks.
Abstract
Large language models (LLMs) have enabled remarkable advances in automated task-solving with multi-agent systems. However, most existing LLM-based multi-agent approaches rely on predefined agents to handle simple tasks, limiting the adaptability of multi-agent collaboration to different scenarios. Therefore, we introduce AutoAgents, an innovative framework that adaptively generates and coordinates multiple specialized agents to build an AI team according to different tasks. Specifically, AutoAgents couples the relationship between tasks and roles by dynamically generating multiple required agents based on task content and planning solutions for the current task based on the generated expert agents. Multiple specialized agents collaborate with each other to efficiently accomplish tasks. Concurrently, an observer role is incorporated into the framework to reflect on the designated plans…
Peer Reviews
Decision·Submitted to ICLR 2024
The idea of dynamically generating agents who play different roles to solve team tasks is interesting and useful. I found the idea to be novel. It is easy for the reader to get a good overview of the idea of AutoAgents. However, there was a need to look at supplementary materials to understand aspects of what the different predefined roles were supposed to do. The visuals helped me understand the idea better. The background was sufficient, in my opinion, and well-written. This discussion and Tab
Section 3: For the agent generation, the motivation for the format of the Prompt P is unclear. Additionally, when we look at the supplementary material, the specific elements of the prompt are not explained -- are these taken from existing works? Others: I also found details that needed to be included in a few other sections, such as the self-refinement process. Furthermore, I had questions about specific choices of parameters during the evaluations. I have included my questions in the next par
1. This paper presents a framework that adaptively generates and coordinates multiple specialized agents to build an AI team according to different tasks. 2. The paper is technically sound and the research question is clear. 3. The contribution of the paper is relevant for LLM-based multi-agent collaboration. The results of this paper is interesting and significant in automatic agent generation. The proposed AutoAgents framework generates more coherent and accurate solutions than the existing m
1. How the proposed AutoAgents framework expands the scope of collaborative applications and reduces the consumption of resources should be elaborated. 2. The authors do not explain how to determine the number of agents in the section of the framework for automatic agent generation. 3. The section about automatic agent generation is too tedious to introduce too much related works 4. In addition to ChatGPT, Vicuna-13B and GPT4 in Table 2, it has not enough recent models to further show the super
- Clear presentation of high-level idea: the overall framework and process is clearly presented through well-drawn figures like Fig. 1 and 2. - Strong reproducibility: the author provides source code and the temperature of LLM is set to 0, which makes it easy to reproduce the result in the paper.
- Limited novelty: according to Table 1, the main difference between the proposed framework and other existing methods like Social Simulacra, Epidemic Modeling, SSP, and AgentVerse is that this work uses self-refinement and collaborative refinement. This difference is more of a prompting technique and has already been used in many existing works like [1, 2, 3] - Unclear presentation of detailed techniques: though the high-level idea is well-presented, the details of many technique are unclear. F
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsTopic Modeling · Multi-Agent Systems and Negotiation · AI in Service Interactions