Elamur Research⌕
☰
home / papers / 2023 / Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents

Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents

Paper · 2023 · Yashar Talebirad and Amirhossein Nadiri

Summary

Conventional GPT-style models work as standalone systems. They cannot coordinate with other models or draw on outside knowledge sources, which limits them on tasks that need teamwork and shared information. Talebirad and Nadiri propose a conceptual framework in which several LLM-based "Intelligent Generative Agents" (IGAs), each assigned its own role, cooperate inside a black-box workspace. The user interacts with this workspace only by giving a prompt and receiving a final answer. The authors formalize the environment as a graph: nodes are agents and plugins, and edges are communication channels. Agents, plugins and messages each receive a tuple-based specification. The design is dynamic, since selected agents may spawn new agents or stop existing ones. The paper reports no datasets, benchmarks or quantitative evaluation. Instead, it illustrates the framework by analyzing Auto-GPT, BabyAGI and Gorilla and by sketching hypothetical scenarios such as a courtroom simulation and a software development team. Its value lies mainly in offering an early, structured vocabulary for designing role-based LLM agent systems and in cataloguing their open problems.

Key points

  • Each agent is defined as A_i = (L_i, R_i, S_i, C_i, H_i):
    • the underlying model and its settings, such as temperature (GPT-4 for demanding reasoning, GPT-3.5-turbo where speed matters);
    • a role;
    • a state made up of knowledge and current "thoughts";
    • a boolean flag for permission to create agents;
    • the set of agents it may halt.
  • Plugins are defined as P_j = (F_j, C_j, U_j): functionalities, configuration (e.g., API keys) and usage constraints. Messages are defined as m = (S_m, A_m, D_m): content, associated action and metadata.
  • One illustrative idea, inspired by GANs, pairs a memory-equipped IGA with an Internet-enabled IGA so that they can jointly reduce hallucinations. The software scenario distributes work among a developer, a tester, a UI designer, a debugger and an Oracle agent that critiques the others.
  • The listed challenges include looping, security risks and uncontrolled growth in the number of agents. For the last, the authors propose a resource-management module and a coordination mechanism. Further challenges are scalability, the lack of suitable evaluation metrics and ethical safeguards.
  • The work is purely conceptual and contains no experimental numbers (arXiv:2306.03314, 2023). Suggested future applications include healthcare, finance, education and transportation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of AlbertaEdmonton · Canada51 works · 5,017 citations🎓York UniversityToronto · Canada15 works · 3,325 citations

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 works

Cite

Cite as Yashar Talebirad and Amirhossein Nadiri. 2023. Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents. arXiv:2306.03314.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2306.03314 · arXiv 2306.03314

@misc{talebirad2023multi,
  title = {{Multi-Agent Collaboration: Harnessing the Power of Intelligent LLM Agents}},
  author = {Yashar Talebirad and Amirhossein Nadiri},
  year = {2023},
  doi = {10.48550/arxiv.2306.03314},
  eprint = {2306.03314},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2306.03314}
}

References

14 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. David Dohan, Winnie Xu, Aitor Lewkowycz, Jacob Austin, David Bieber, Raphael Gontijo Lopes, Yuhuai Wu, Henryk Michalewski, Rif A. Saurous, Jascha Sohl-dickstein, Kevin Murphy, and Charles Sutton. Language model cascades, 2022.
  2. Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior, 2023. find
  3. Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. Camel: Communicative agents for "mind" exploration of large scale language model society, 2023. find
  4. Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, Harsha Nori, Hamid Palangi, Marco Tulio Ribeiro, and Yi Zhang. Sparks of artificial general intelligence: Early experiments with gpt-4, 2023.
  5. Chaoning Zhang, Chenshuang Zhang, Chenghao Li, Yu Qiao, Sheng Zheng, Sumit Kumar Dam, Mengchun Zhang, Jung Uk Kim, Seong Tae Kim, Jinwoo Choi, Gyeong-Moon Park, Sung-Ho Bae, Lik-Hang Lee, Pan Hui, In So Kweon, and Choong Seon Hong. One small step for generative ai, one giant leap for agi: A complete… find
  6. Yejin Bang, Samuel Cahyawijaya, Nayeon Lee, Wenliang Dai, Dan Su, Bryan Wilie, Holy Lovenia, Ziwei Ji, Tiezheng Yu, Willy Chung, Quyet V. Do, Yan Xu, and Pascale Fung. A multitask, multilingual, multimodal evaluation of chatgpt on reasoning, hallucination, and interactivity, 2023.
  7. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models, 2023.
  8. Yao Fu, Hao Peng, Tushar Khot, and Mirella Lapata. Improving language model negotiation with self-play and in-context learning from ai feedback, 2023.
  9. Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. Teaching large language models to self-debug, 2023. find
  10. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. Self-refine: Iterative refinement with self-… find
  11. OpenAI. Introducing chatgpt. https://openai.com/blog/chatgpt, 2022. Accessed: 2023-06-04. ↗
  12. Shishir G. Patil, Tianjun Zhang, Xin Wang, and Joseph E. Gonzalez. Gorilla: Large language model connected with massive apis, 2023.
  13. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023.
  14. Sil Hamilton. Blind judgement: Agent-based supreme court modelling with gpt, 2023. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/talebirad-2023-multi-agent-collaboration-harnessing-power/