Papers
Topics
Authors
Recent
Search
2000 character limit reached

DAWN: Designing Distributed Agents in a Worldwide Network

Published 11 Oct 2024 in cs.NI, cs.AI, and cs.MA | (2410.22339v3)

Abstract: The rapid evolution of LLMs has transformed them from basic conversational tools into sophisticated entities capable of complex reasoning and decision-making. These advancements have led to the development of specialized LLM-based agents designed for diverse tasks such as coding and web browsing. As these agents become more capable, the need for a robust framework that facilitates global communication and collaboration among them towards advanced objectives has become increasingly critical. Distributed Agents in a Worldwide Network (DAWN) addresses this need by offering a versatile framework that integrates LLM-based agents with traditional software systems, enabling the creation of agentic applications suited for a wide range of use cases. DAWN enables distributed agents worldwide to register and be easily discovered through Gateway Agents. Collaborations among these agents are coordinated by a Principal Agent equipped with reasoning strategies. DAWN offers three operational modes: No-LLM Mode for deterministic tasks, Copilot for augmented decision-making, and LLM Agent for autonomous operations. Additionally, DAWN ensures the safety and security of agent collaborations globally through a dedicated safety, security, and compliance layer, protecting the network against attackers and adhering to stringent security and compliance standards. These features make DAWN a robust network for deploying agent-based applications across various industries.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (64)
  1. OpenAI, “Chatgpt,” https://www.openai.com/chatgpt, 2024.
  2. H. Touvron et al., “The llama 3 herd of models,” arXiv preprint arXiv:2407.21783, 2024. [Online]. Available: https://arxiv.org/abs/2407.21783
  3. Anthropic, “The claude 3 model family: Opus, sonnet, haiku,” https://www.anthropic.com/research, 2024.
  4. Y. Cheng, C. Zhang, Z. Zhang, X. Meng, S. Hong, W. Li, Z. Wang, Z. Wang, F. Yin, J. Zhao, and X. He, “Exploring large language model based intelligent agents: Definitions, methods, and prospects,” 2024. [Online]. Available: https://arxiv.org/abs/2401.03428
  5. O. I. Contributors, “Open interpreter: A natural language interface for code execution,” https://github.com/OpenInterpreter/open-interpreter, 2024, accessed: 2024-09-15.
  6. E. Nijkamp, S. Teyssedre, B. Zhang et al., “Codegen: An open large language model for code with a tree-structured decoder,” arXiv preprint arXiv:2303.17568, 2023. [Online]. Available: https://arxiv.org/abs/2303.17568
  7. S. Zhou, F. F. Xu, H. Zhu, X. Zhou, R. Lo, A. Sridhar, X. Cheng, Y. Bisk, D. Fried, U. Alon, and L. Zettlemoyer, “Webarena: A realistic web environment for building autonomous agents,” arXiv preprint arXiv:2307.13854, 2023. [Online]. Available: https://arxiv.org/abs/2307.13854
  8. L. Fan, Y. Zhu et al., “Voyager: An open-ended embodied agent with large language models,” arXiv preprint arXiv:2409.05152, 2024. [Online]. Available: https://arxiv.org/abs/2409.05152
  9. C. Zhang, K. Yang, S. Hu, Z. Wang, G. Li, Y. Sun, C. Zhang, Z. Zhang, A. Liu, S.-C. Zhu, X. Chang, J. Zhang, F. Yin, Y. Liang, and Y. Yang, “Proagent: Building proactive cooperative agents with large language models,” arXiv preprint arXiv:2308.11339, 2023. [Online]. Available: https://arxiv.org/abs/2308.11339
  10. Z. Hong, X. Liu, Z. Zhu, W. Wang, H. Liu, and C. Liu, “Metagpt: Meta programming for multi-agent collaboration,” arXiv preprint arXiv:2308.00352, 2023. [Online]. Available: https://arxiv.org/abs/2308.00352
  11. S. Li, X. Deng, J. Ma, and et al., “Camel: Communicative agents for “mind” exploration of large language model society,” arXiv preprint arXiv:2303.17760, 2023. [Online]. Available: https://arxiv.org/abs/2303.17760
  12. Z. Li, W. Zhang, J. Liu, and et al., “Agentverse: Facilitating multi-agent collaboration and exploring emergent behaviors in agents,” arXiv preprint arXiv:2305.15021, 2023. [Online]. Available: https://arxiv.org/abs/2305.15021
  13. LangChain, “Langgraph: A graph-based framework for building ai agents,” https://www.langchain.com/langgraph, 2023, accessed: 2024-09-15.
  14. Q. Wu, T. Li, H. Zhao et al., “Autogen: Enabling next-gen llm applications via multi-agent conversation,” arXiv preprint arXiv:2308.08155, 2023. [Online]. Available: https://arxiv.org/abs/2308.08155
  15. C. S. Xia, Y. Deng, S. Dunn, and L. Zhang, “Agentless: Demystifying llm-based software engineering agents,” 2024. [Online]. Available: https://arxiv.org/abs/2407.01489
  16. K. Greshake, S. Abdelnabi, S. Mishra, C. Endres, T. Holz, and M. Fritz, “Not what you’ve signed up for: Compromising real-world llm-integrated applications with indirect prompt injection,” 2023. [Online]. Available: https://arxiv.org/abs/2302.12173
  17. J. Yu, X. Lin, Z. Yu, and X. Xing, “Gptfuzzer: Red teaming large language models with auto-generated jailbreak prompts,” 2024. [Online]. Available: https://arxiv.org/abs/2309.10253
  18. OWASP Top 10 for LLM Applications Team, “LLM AI Cybersecurity & Governance Checklist,” April 2024.
  19. W. Zhao, V. Khazanchi, H. Xing, X. He, Q. Xu, and N. D. Lane, “Attacks on third-party apis of large language models,” 2024. [Online]. Available: https://arxiv.org/abs/2404.16891
  20. J. Kaplan, S. McCandlish, T. Henighan, T. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei, “Scaling laws for neural language models,” arXiv preprint arXiv:2001.08361, 2020. [Online]. Available: https://arxiv.org/abs/2001.08361
  21. S. Bubeck, V. Chandrasekaran, R. Eldan, J. Gehrke, E. Horvitz, E. Kamar, P. Lee, and N. Tandon, “Sparks of artificial general intelligence: Early experiments with gpt-4,” arXiv preprint arXiv:2303.12712, 2023. [Online]. Available: https://arxiv.org/abs/2303.12712
  22. Z. Xu, S. Jain, and M. Kankanhalli, “Hallucination is inevitable: An innate limitation of large language models,” arXiv preprint arXiv:2401.11817, 2024. [Online]. Available: https://arxiv.org/abs/2401.11817
  23. H. Ye, T. Liu, A. Zhang, W. Hua, and W. Jia, “Cognitive mirage: A review of hallucinations in large language models,” arXiv preprint arXiv:2309.06794, 2023. [Online]. Available: https://arxiv.org/abs/2309.06794
  24. M. Wooldridge and N. R. Jennings, “Intelligent agents: Theory and practice,” The Knowledge Engineering Review, vol. 10, no. 2, pp. 115–152, 1995.
  25. L. Wang, C. Ma, X. Feng, Z. Zhang, H. Yang, J. Zhang, Z. Chen, J. Tang, X. Chen, Y. Lin, W. X. Zhao, Z. Wei, and J.-R. Wen, “A survey on large language model based autonomous agents,” arXiv preprint arXiv:2307.11413, 2023. [Online]. Available: https://arxiv.org/abs/2307.11413
  26. J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, E. H. Chi, Q. V. Le, and D. Zhou, “Chain-of-thought prompting elicits reasoning in large language models,” arXiv preprint arXiv:2201.11903, 2022. [Online]. Available: https://arxiv.org/abs/2201.11903
  27. S. Yao, D. Yu, J. Z. Lu, I. Shafran, K. Narasimhan, Y. Cao, and W. Chen, “React: Synergizing reasoning and acting in language models,” arXiv preprint arXiv:2210.03629, 2022. [Online]. Available: https://arxiv.org/abs/2210.03629
  28. S. Yao, J. Zhao, D. Yu, Y. Cao, K. Narasimhan, and P. Ammanabrolu, “Tree of thoughts: Deliberate problem solving with large language models,” arXiv preprint arXiv:2305.10601, 2023. [Online]. Available: https://arxiv.org/abs/2305.10601
  29. E. Salas, K. C. Stagl, C. S. Burke, and G. F. Goodwin, “Teamwork: Emerging principles,” Philosophy and Technology, vol. 7, no. 3, pp. 341–350, 2000. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1111/1468-2370.00046
  30. S. Qian, X. Gu, Z. Zhou, X. Wang, B. Tan, B. Chen, and et al., “Chatdev: Developing software with llm-based multi-agent chat systems,” arXiv preprint arXiv:2309.02405, 2023. [Online]. Available: https://arxiv.org/abs/2309.02405
  31. Y. Liu, H. Zhang, W. Wang, and et al., “Agentcoder: Multi-agent-based code generation with iterative testing and optimisation,” arXiv preprint arXiv:2308.09351, 2023. [Online]. Available: https://arxiv.org/abs/2308.09351
  32. H. Lin, T. Wang, and et al., “Medagents: Large language models as collaborators for zero-shot medical reasoning,” arXiv preprint arXiv:2309.12732, 2023. [Online]. Available: https://arxiv.org/abs/2309.12732
  33. T. Ju, Y. Wang, X. Ma, P. Cheng, H. Zhao, Y. Wang, L. Liu, J. Xie, Z. Zhang, and G. Liu, “Flooding spread of manipulated knowledge in llm-based multi-agent communities,” 2024. [Online]. Available: https://arxiv.org/abs/2407.07791
  34. W. Chen, Z. You, R. Li, Y. Guan, C. Qian, C. Zhao, C. Yang, R. Xie, Z. Liu, and M. Sun, “Internet of agents: Weaving a web of heterogeneous agents for collaborative intelligence,” arXiv preprint arXiv:2407.07061, 2024. [Online]. Available: https://arxiv.org/abs/2407.07061
  35. B. Xu, Z. Peng, B. Lei, S. Mukherjee, Y. Liu, and D. Xu, “Rewoo: Decoupling reasoning from observations for efficient augmented language models,” 2023. [Online]. Available: https://arxiv.org/abs/2305.18323
  36. Y. Shen, K. Song, X. Tan, D. Li, W. Lu, and Y. Zhuang, “Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face,” 2023. [Online]. Available: https://arxiv.org/abs/2303.17580
  37. HuggingFace, “Hugging face: The ai community building the future.” [Online]. Available: https://huggingface.co
  38. A. Zhou, K. Yan, M. Shlapentokh-Rothman, H. Wang, and Y.-X. Wang, “Language agent tree search unifies reasoning acting and planning in language models,” 2024. [Online]. Available: https://arxiv.org/abs/2310.04406
  39. Y. Qin, S. Liang, Y. Ye, K. Zhu, L. Yan, Y. Lu, Y. Lin, X. Cong, X. Tang, B. Qian, S. Zhao, L. Hong, R. Tian, R. Xie, J. Zhou, M. Gerstein, D. Li, Z. Liu, and M. Sun, “Toolllm: Facilitating large language models to master 16000+ real-world apis,” 2023. [Online]. Available: https://arxiv.org/abs/2307.16789
  40. M. Shridhar, X. Yuan, M.-A. Côté, Y. Bisk, A. Trischler, and M. Hausknecht, “ALFWorld: Aligning Text and Embodied Environments for Interactive Learning,” in Proceedings of the International Conference on Learning Representations (ICLR), 2021. [Online]. Available: https://arxiv.org/abs/2010.03768
  41. T. Abuelsaad, D. Akkil, P. Dey, A. Jagmohan, A. Vempaty, and R. Kokku, “Agent-e: From autonomous web navigation to foundational design principles in agentic systems,” 2024. [Online]. Available: https://arxiv.org/abs/2407.13032
  42. OpenAI. Openai guides on function calling. [Online]. Available: https://platform.openai.com/docs/guides/function-calling
  43. S. D. Sri, M. A. S, S. V. R, R. C. Raman, G. Rajagopal, and S. T. Chan, “Automating rest api postman test cases using llm,” 2024. [Online]. Available: https://arxiv.org/abs/2404.10678
  44. M. Kim, T. Stennett, D. Shah, S. Sinha, and A. Orso, “Leveraging large language models to improve rest api testing,” in Proceedings of the 2024 ACM/IEEE 44th International Conference on Software Engineering: New Ideas and Emerging Results, ser. ICSE-NIER’24.   New York, NY, USA: Association for Computing Machinery, 2024, p. 37–41. [Online]. Available: https://doi.org/10.1145/3639476.3639769
  45. S. Yuan, K. Song, J. Chen, X. Tan, Y. Shen, R. Kan, D. Li, and D. Yang, “Easytool: Enhancing llm-based agents with concise tool instruction,” 2024. [Online]. Available: https://arxiv.org/abs/2401.06201
  46. RapidAPI, “Api testing guide: Everything you need to know,” RapidAPI Blog. [Online]. Available: https://rapidapi.com/blog/api-testing/
  47. I. Stoica, R. Morris, D. R. Karger, M. F. Kaashoek, and H. Balakrishnan, “Semantic search,” in Proceedings of the 12th international conference on World Wide Web.   ACM, 2003, pp. 700–709. [Online]. Available: https://dl.acm.org/doi/abs/10.1145/775152.775250
  48. S. G. Patil, T. Zhang, X. Wang, and J. E. Gonzalez, “Gorilla: Large language model connected with massive apis,” 2023. [Online]. Available: https://arxiv.org/abs/2305.15334
  49. J. Granjal, E. Monteiro, and J. S. S. Silva, “A secure interconnection model for ipv6 enabled wireless sensor networks,” in 2010 IFIP Wireless Days.   IEEE, 2010, pp. 1–6.
  50. S. Jebri, M. Abid, and A. Bouallegue, “An efficient scheme for anonymous communication in iot,” in 2015 11th International Conference on Information Assurance and Security (IAS).   IEEE, 2015, pp. 7–12.
  51. Q. Jing, A. V. Vasilakos, J. Wan, J. Lu, and D. Qiu, “Security of the internet of things: perspectives and challenges,” Wireless networks, vol. 20, pp. 2481–2501, 2014.
  52. L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qin et al., “A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,” arXiv preprint arXiv:2311.05232, 2023.
  53. H. Li, D. Guo, W. Fan, M. Xu, J. Huang, F. Meng, and Y. Song, “Multi-step jailbreaking privacy attacks on chatgpt,” 2023. [Online]. Available: https://arxiv.org/abs/2304.05197
  54. X. Shen, Z. Chen, M. Backes, Y. Shen, and Y. Zhang, “”do anything now”: Characterizing and evaluating in-the-wild jailbreak prompts on large language models,” 2024. [Online]. Available: https://arxiv.org/abs/2308.03825
  55. Y. Liu, G. Deng, Y. Li, K. Wang, Z. Wang, X. Wang, T. Zhang, Y. Liu, H. Wang, Y. Zheng, and Y. Liu, “Prompt injection attack against llm-integrated applications,” 2024. [Online]. Available: https://arxiv.org/abs/2306.05499
  56. N. Carlini, M. Jagielski, C. A. Choquette-Choo, D. Paleka, W. Pearce, H. Anderson, A. Terzis, K. Thomas, and F. Tramèr, “Poisoning web-scale training datasets is practical,” 2024. [Online]. Available: https://arxiv.org/abs/2302.10149
  57. K. Kurita, P. Michel, and G. Neubig, “Weight poisoning attacks on pre-trained models,” 2020. [Online]. Available: https://arxiv.org/abs/2004.06660
  58. W. Zou, R. Geng, B. Wang, and J. Jia, “Poisonedrag: Knowledge corruption attacks to retrieval-augmented generation of large language models,” 2024. [Online]. Available: https://arxiv.org/abs/2402.07867
  59. Q. Zhan, Z. Liang, Z. Ying, and D. Kang, “Injecagent: Benchmarking indirect prompt injections in tool-integrated large language model agents,” 2024. [Online]. Available: https://arxiv.org/abs/2403.02691
  60. F. He, T. Zhu, D. Ye, B. Liu, W. Zhou, and P. S. Yu, “The emerged security and privacy of llm agent: A survey with case studies,” arXiv preprint arXiv:2407.19354, 2024.
  61. Z. Zhang, M. Jia, H.-P. Lee, B. Yao, S. Das, A. Lerner, D. Wang, and T. Li, ““it’s a fair game”, or is it? examining how users navigate disclosure risks and benefits when using llm-based conversational agents,” in Proceedings of the CHI Conference on Human Factors in Computing Systems, 2024, pp. 1–26.
  62. Y. Ruan, H. Dong, A. Wang, S. Pitis, Y. Zhou, J. Ba, Y. Dubois, C. J. Maddison, and T. Hashimoto, “Identifying the risks of lm agents with an lm-emulated sandbox,” 2024. [Online]. Available: https://arxiv.org/abs/2309.15817
  63. J. Welbl, A. Glaese, J. Uesato, S. Dathathri, J. Mellor, L. A. Hendricks, K. Anderson, P. Kohli, B. Coppin, and P.-S. Huang, “Challenges in detoxifying language models,” 2021. [Online]. Available: https://arxiv.org/abs/2109.07445
  64. S. Gehman, S. Gururangan, M. Sap, Y. Choi, and N. A. Smith, “Realtoxicityprompts: Evaluating neural toxic degeneration in language models,” 2020. [Online]. Available: https://arxiv.org/abs/2009.11462

Summary

  • The paper introduces DAWN, a sophisticated framework for designing and managing collaborative LLM-based agents across a global, distributed network.
  • DAWN features distributed agent discovery, flexible operational modes (No-LLM, Copilot, LLM Agent), and a robust security layer against LLM-specific threats.
  • DAWN offers practical implications including enhanced productivity, scalability, and adaptability for various domains, setting a foundation for future agentic AI.

Analysis of DAWN: Designing Distributed Agents in a Worldwide Network

The paper entitled "DAWN: Designing Distributed Agents in a Worldwide Network" outlines a sophisticated framework designed to integrate LLM-based agents with traditional software systems, known as DAWN. This platform provides an architecture enabling diverse collaborations among agents globally to achieve complex goals through a flexible, secure, and scalable system.

Overview of DAWN

The core contribution of DAWN lies in its structured yet versatile approach to managing agent collaboration across a vast network. DAWN introduces a hierarchy of agents, primarily composed of Principal Agents and Gateway Agents. The Principal Agent serves as an orchestrator, planning and executing user tasks by dynamically discovering and invoking resources through various Gateway Agents.

Key Features of DAWN

  1. Agent Registration and Discovery: DAWN enables distributed agents worldwide to register and be discovered through Gateway Agents. These Gateway Agents keep an updated registry of resources, including tools and other agents, thus enabling efficient discovery and management of registered applications.
  2. Operational Modes:
    • No-LLM Mode: Geared towards deterministic tasks using traditional algorithms.
    • Copilot Mode: Combines deterministic workflows with LLM capabilities to enhance decision-making.
    • LLM Agent Mode: Facilitates autonomous operations allowing agents to make real-time decisions utilizing complex reasoning.
  3. Safety and Security: DAWN incorporates a dedicated security layer, addressing threats peculiar to LLM environments, such as prompt injection and misuse of API capabilities. This layer ensures that all agents comply with stringent security protocols to safeguard against unauthorized access and data breaches.
  4. Interoperability and Flexibility: DAWN accommodates a wide range of task requirements, ensuring the most appropriate blend of LLM and conventional methodologies is deployed for each task. This adaptability is achieved through a unified framework allowing seamless integration.
  5. Global Communication and Compliance: Using open protocols and encrypted communication channels, DAWN allows diverse agents to interact securely. Compliance mechanisms adhere to industry and organizational standards, addressing regulatory concerns.

Implications and Future Directions

The development of DAWN represents a significant stride towards creating a globally distributed network of intelligent agents capable of solving complex tasks autonomously. The integration flexibility allows DAWN to be applied across diverse domains, including cybersecurity, healthcare, and logistics.

Practical Implications:

  • Enhanced Productivity: By automating repetitive and complex tasks, DAWN could significantly increase operational efficiency across various industries.
  • Scalability and Dynamic Adaptation: DAWN's architecture ensures it can handle an increasing number of agents, adapting to their evolving capabilities and the tasks at hand.

Theoretical Implications:

  • Inter-agent Collaboration: DAWN lays the groundwork for future research in optimizing inter-agent communication and collaborative task execution.
  • Security Models: The framework's approach to integrating comprehensive security and compliance layers could inform future AI system designs, emphasizing similar concerns.

Speculation on Future Developments:

  • As LLMs continue to evolve, future iterations of frameworks like DAWN could see even more enhanced decision-making capabilities through more sophisticated reasoning strategies.
  • Advances in distributed systems may lead to more efficient global agent networks with reduced latency and increased reliability.

In conclusion, DAWN offers a robust and promising approach for deploying multi-agent systems globally, harnessing the power of LLMs within a secure, adaptable, and flexible network. This framework not only addresses current challenges in LLM-based systems but also sets a foundation for future advancements in agentic AI applications.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Sign Up to Generate

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Generate Now

Continue Learning

We haven't generated follow-up questions for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Generate Now

Collections

Sign up for free to add this paper to one or more collections.

User Circle Single Streamline Icon: https://streamlinehq.com Sign Up

Tweets

Sign up for free to view the 2 tweets with 1 like about this paper.

User Circle Single Streamline Icon: https://streamlinehq.com Sign Up for Free

HackerNews

  1. DAWN: Designing Distributed Agents in a Worldwide Network (10 points, 3 comments)