Papers
Topics
Authors
Recent
Search
2000 character limit reached

Agent Design Pattern Catalogue: A Collection of Architectural Patterns for Foundation Model based Agents

Published 16 May 2024 in cs.AI and cs.SE | (2405.10467v4)

Abstract: Foundation model-enabled generative artificial intelligence facilitates the development and implementation of agents, which can leverage distinguished reasoning and language processing capabilities to takes a proactive, autonomous role to pursue users' goals. Nevertheless, there is a lack of systematic knowledge to guide practitioners in designing the agents considering challenges of goal-seeking (including generating instrumental goals and plans), such as hallucinations inherent in foundation models, explainability of reasoning process, complex accountability, etc. To address this issue, we have performed a systematic literature review to understand the state-of-the-art foundation model-based agents and the broader ecosystem. In this paper, we present a pattern catalogue consisting of 18 architectural patterns with analyses of the context, forces, and trade-offs as the outcomes from the previous literature review. We propose a decision model for selecting the patterns. The proposed catalogue can provide holistic guidance for the effective use of patterns, and support the architecture design of foundation model-based agents by facilitating goal-seeking and plan generation.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (59)
  1. R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, S. von Arx, M. S. Bernstein, J. Bohg, A. Bosselut, E. Brunskill et al., “On the opportunities and risks of foundation models,” arXiv preprint arXiv:2108.07258, 2021.
  2. U. Anwar, A. Saparov, J. Rando, D. Paleka, M. Turpin, P. Hase, E. S. Lubana, E. Jenner, S. Casper, O. Sourbut et al., “Foundational challenges in assuring alignment and safety of large language models,” arXiv preprint arXiv:2404.09932, 2024.
  3. Q. Lu, L. Zhu, X. Xu, Z. Xing, S. Harrer, and J. Whittle, “Towards responsible generative ai: A reference architecture for designing foundation model based agents,” arXiv preprint arXiv:2311.13148, 2023.
  4. J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat et al., “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774, 2024.
  5. K. Hu, “Chatgpt sets record for fastest-growing user base - analyst note,” Feb 2023. [Online]. Available: https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/
  6. Mar 2024. [Online]. Available: https://www.deeplearning.ai/the-batch/issue-242/
  7. C. Packer, V. Fang, S. G. Patil, K. Lin, S. Wooders, and J. E. Gonzalez, “Memgpt: Towards llms as operating systems,” arXiv preprint arXiv:2310.08560, 2024.
  8. S. Colabianchi, A. Tedeschi, and F. Costantino, “Human-technology integration with industrial conversational agents: A conceptual architecture and a taxonomy for manufacturing,” Journal of Industrial Information Integration, vol. 35, p. 100510, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2452414X23000833
  9. “Openai’s bet on a cognitive architecture,” Nov 2023. [Online]. Available: https://blog.langchain.dev/openais-bet-on-a-cognitive-architecture/
  10. A. E. Hassan, D. Lin, G. K. Rajbahadur, K. Gallaba, F. R. Cogo, B. Chen, H. Zhang, K. Thangarajah, G. A. Oliva, J. Lin et al., “Rethinking software engineering in the era of foundation models: A curated catalogue of challenges in the development of trustworthy fmware,” arXiv preprint arXiv:2402.15943, 2024.
  11. S. Gao, A. Fang, Y. Huang, V. Giunchiglia, A. Noori, J. R. Schwarz, Y. Ektefaie, J. Kondic, and M. Zitnik, “Empowering biomedical discovery with ai agents,” arXiv preprint arXiv:2404.02831, 2024.
  12. M. Nafreen, S. Bhattacharya, and L. Fiondella, “Architecture-based software reliability incorporating fault tolerant machine learning,” in 2020 Annual Reliability and Maintainability Symposium (RAMS), 2020, pp. 1–6.
  13. Y. Liu, S. Chen, H. Chen, M. Yu, X. Ran, A. Mo, Y. Tang, and Y. Huang, “How ai processing delays foster creativity: Exploring research question co-creation with an llm-based agent,” arXiv preprint arXiv:2310.06155, 2023.
  14. S. S. Kannan, V. L. Venkatesh, and B.-C. Min, “Smart-llm: Smart multi-agent robot task planning using large language models,” arXiv preprint arXiv:2309.10062, 2023.
  15. X. Zeng, X. Wang, T. Zhang, C. Yu, S. Zhao, and Y. Chen, “Gesturegpt: Zero-shot interactive gesture understanding and grounding with large language model agents,” arXiv preprint arXiv:2310.12821, 2023.
  16. D. Zhao, Z. Xing, X. Xia, D. Ye, X. Xu, and L. Zhu, “Seehow: Workflow extraction from programming screencasts through action-aware video analytics,” arXiv preprint arXiv:2304.14042, 2023.
  17. C. Zhang, K. Yang, S. Hu, Z. Wang, G. Li, Y. Sun, C. Zhang, Z. Zhang, A. Liu, S.-C. Zhu et al., “Proagent: Building proactive cooperative ai with large language models,” arXiv preprint arXiv:2308.11339, 2023.
  18. A. Zhao, D. Huang, Q. Xu, M. Lin, Y.-J. Liu, and G. Huang, “Expel: Llm agents are experiential learners,” arXiv preprint arXiv:2308.10144, 2023.
  19. R. Schumann, W. Zhu, W. Feng, T.-J. Fu, S. Riezler, and W. Y. Wang, “Velma: Verbalization embodiment of llm agents for vision and language navigation in street view,” arXiv preprint arXiv:2307.06082, 2023.
  20. Y. Hu and Y. Lu, “Rag and rau: A survey on retrieval-augmented language model in natural language processing,” arXiv preprint arXiv:2404.19543, 2024.
  21. S.-Q. Yan, J.-C. Gu, Y. Zhu, and Z.-H. Ling, “Corrective retrieval augmented generation,” arXiv preprint arXiv:2401.15884, 2024.
  22. Z. Levonian, C. Li, W. Zhu, A. Gade, O. Henkel, M.-E. Postle, and W. Xing, “Retrieval-augmented generation to improve math question-answering: Trade-offs between groundedness and human preference,” arXiv preprint arXiv:2310.03184, 2023.
  23. L. Wang, C. Ma, X. Feng, Z. Zhang, H. Yang, J. Zhang, Z. Chen, J. Tang, X. Chen, Y. Lin et al., “A survey on large language model based autonomous agents,” Frontiers of Computer Science, vol. 18, no. 6, pp. 1–26, 2024.
  24. 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,” in Advances in Neural Information Processing Systems, A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, Eds., vol. 36.   Curran Associates, Inc., 2023, pp. 38 154–38 180. [Online]. Available: https://proceedings.neurips.cc/paper_files/paper/2023/file/77c33e6a367922d003ff102ffb92b658-Paper-Conference.pdf
  25. J. Zhang, R. Krishna, A. H. Awadallah, and C. Wang, “Ecoassistant: Using llm assistant more affordably and accurately,” arXiv preprint arXiv:2310.03046, 2023.
  26. B. Xu, Z. Peng, B. Lei, S. Mukherjee, Y. Liu, and D. Xu, “Rewoo: Decoupling reasoning from observations for efficient augmented language models,” arXiv preprint arXiv:2305.18323, 2023.
  27. J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V. Le, D. Zhou et al., “Chain-of-thought prompting elicits reasoning in large language models,” Advances in Neural Information Processing Systems, vol. 35, pp. 24 824–24 837, 2022.
  28. Z. Wang, Z. Liu, Y. Zhang, A. Zhong, L. Fan, L. Wu, and Q. Wen, “Rcagent: Cloud root cause analysis by autonomous agents with tool-augmented large language models,” arXiv preprint arXiv:2310.16340, 2023.
  29. Z. Zhang, Y. Yao, A. Zhang, X. Tang, X. Ma, Z. He, Y. Wang, M. Gerstein, R. Wang, G. Liu et al., “Igniting language intelligence: The hitchhiker’s guide from chain-of-thought reasoning to language agents,” arXiv preprint arXiv:2311.11797, 2023.
  30. S. Yao, D. Yu, J. Zhao, I. Shafran, T. Griffiths, Y. Cao, and K. Narasimhan, “Tree of thoughts: Deliberate problem solving with large language models,” Advances in Neural Information Processing Systems, vol. 36, 2024.
  31. J. Huang, S. S. Gu, L. Hou, Y. Wu, X. Wang, H. Yu, and J. Han, “Large language models can self-improve,” arXiv preprint arXiv:2210.11610, 2022.
  32. N. Shinn, F. Cassano, A. Gopinath, K. Narasimhan, and S. Yao, “Reflexion: language agents with verbal reinforcement learning,” in Advances in Neural Information Processing Systems, A. Oh, T. Neumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, Eds., vol. 36.   Curran Associates, Inc., 2023, pp. 8634–8652. [Online]. Available: https://proceedings.neurips.cc/paper_files/paper/2023/file/1b44b878bb782e6954cd888628510e90-Paper-Conference.pdf
  33. J. Chen, S. Yuan, R. Ye, B. P. Majumder, and K. Richardson, “Put your money where your mouth is: Evaluating strategic planning and execution of llm agents in an auction arena,” arXiv preprint arXiv:2310.05746, 2023.
  34. J. S. Park, J. O’Brien, C. J. Cai, M. R. Morris, P. Liang, and M. S. Bernstein, “Generative agents: Interactive simulacra of human behavior,” in Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, ser. UIST ’23.   New York, NY, USA: Association for Computing Machinery, 2023. [Online]. Available: https://doi.org/10.1145/3586183.3606763
  35. W. Yao, S. Heinecke, J. C. Niebles, Z. Liu, Y. Feng, L. Xue, R. Murthy, Z. Chen, J. Zhang, D. Arpit et al., “Retroformer: Retrospective large language agents with policy gradient optimization,” arXiv preprint arXiv:2308.02151, 2023.
  36. C. Qian, X. Cong, C. Yang, W. Chen, Y. Su, J. Xu, Z. Liu, and M. Sun, “Communicative agents for software development,” arXiv preprint arXiv:2307.07924, 2023.
  37. Y. Talebirad and A. Nadiri, “Multi-agent collaboration: Harnessing the power of intelligent llm agents,” arXiv preprint arXiv:2306.03314, 2023.
  38. W. Huang, F. Xia, T. Xiao, H. Chan, J. Liang, P. Florence, A. Zeng, J. Tompson, I. Mordatch, Y. Chebotar et al., “Inner monologue: Embodied reasoning through planning with language models,” arXiv preprint arXiv:2207.05608, 2022.
  39. S. Ma, Q. Chen, X. Wang, C. Zheng, Z. Peng, M. Yin, and X. Ma, “Towards human-ai deliberation: Design and evaluation of llm-empowered deliberative ai for ai-assisted decision-making,” arXiv preprint arXiv:2403.16812, 2024.
  40. Y. Wang, Z. Liu, J. Zhang, W. Yao, S. Heinecke, and P. S. Yu, “Drdt: Dynamic reflection with divergent thinking for llm-based sequential recommendation,” arXiv preprint arXiv:2312.11336, 2023.
  41. S. Hamilton, “Blind judgement: Agent-based supreme court modelling with GPT,” in The AAAI-23 Workshop on Creative AI Across Modalities, 2023. [Online]. Available: https://openreview.net/forum?id=Nx9ajnqG9Rw
  42. C.-M. Chan, W. Chen, Y. Su, J. Yu, W. Xue, S. Zhang, J. Fu, and Z. Liu, “Chateval: Towards better LLM-based evaluators through multi-agent debate,” in The Twelfth International Conference on Learning Representations, 2024. [Online]. Available: https://openreview.net/forum?id=FQepisCUWu
  43. J. C. Yang, M. Korecki, D. Dailisan, C. I. Hausladen, and D. Helbing, “Llm voting: Human choices and ai collective decision making,” arXiv preprint arXiv:2402.01766, 2024.
  44. J. Li, Q. Zhang, Y. Yu, Q. Fu, and D. Ye, “More agents is all you need,” arXiv preprint arXiv:2402.05120, 2024.
  45. S. Hong, X. Zheng, J. Chen, Y. Cheng, J. Wang, C. Zhang, Z. Wang, S. K. S. Yau, Z. Lin, L. Zhou et al., “Metagpt: Meta programming for multi-agent collaborative framework,” arXiv preprint arXiv:2308.00352, 2023.
  46. X. Tang, A. Zou, Z. Zhang, Y. Zhao, X. Zhang, A. Cohan, and M. Gerstein, “Medagents: Large language models as collaborators for zero-shot medical reasoning,” arXiv preprint arXiv:2311.10537, 2023.
  47. M. Li, L. Chen, J. Chen, S. He, J. Gu, and T. Zhou, “Selective reflection-tuning: Student-selected data recycling for llm instruction-tuning,” arXiv preprint arXiv:2402.10110, 2024.
  48. T. Liang, Z. He, W. Jiao, X. Wang, Y. Wang, R. Wang, Y. Yang, Z. Tu, and S. Shi, “Encouraging divergent thinking in large language models through multi-agent debate,” arXiv preprint arXiv:2305.19118, 2023.
  49. Y. Du, S. Li, A. Torralba, J. B. Tenenbaum, and I. Mordatch, “Improving factuality and reasoning in language models through multiagent debate,” arXiv preprint arXiv:2305.14325, 2023.
  50. H. Chen, W. Ji, L. Xu, and S. Zhao, “Multi-agent consensus seeking via large language models,” arXiv preprint arXiv:2310.20151, 2023.
  51. T. Rebedea, R. Dinu, M. Sreedhar, C. Parisien, and J. Cohen, “Nemo guardrails: A toolkit for controllable and safe llm applications with programmable rails,” arXiv preprint arXiv:2310.10501, 2023.
  52. H. Inan, K. Upasani, J. Chi, R. Rungta, K. Iyer, Y. Mao, M. Tontchev, Q. Hu, B. Fuller, D. Testuggine et al., “Llama guard: Llm-based input-output safeguard for human-ai conversations,” arXiv preprint arXiv:2312.06674, 2023.
  53. J. Ruan, Y. Chen, B. Zhang, Z. Xu, T. Bao, du qing, shi shiwei, H. Mao, X. Zeng, and R. Zhao, “TPTU: Task planning and tool usage of large language model-based AI agents,” in NeurIPS 2023 Foundation Models for Decision Making Workshop, 2023. [Online]. Available: https://openreview.net/forum?id=GrkgKtOjaH
  54. G. Wang, Y. Xie, Y. Jiang, A. Mandlekar, C. Xiao, Y. Zhu, L. Fan, and A. Anandkumar, “Voyager: An open-ended embodied agent with large language models,” in Intrinsically-Motivated and Open-Ended Learning Workshop @NeurIPS2023, 2023. [Online]. Available: https://openreview.net/forum?id=nfx5IutEed
  55. T. Xie, F. Zhou, Z. Cheng, P. Shi, L. Weng, Y. Liu, T. J. Hua, J. Zhao, Q. Liu, C. Liu et al., “Openagents: An open platform for language agents in the wild,” arXiv preprint arXiv:2310.10634, 2023.
  56. Y. Liu, Q. Lu, G. Yu, H.-Y. Paik, H. Perera, and L. Zhu, “A pattern language for blockchain governance,” in Proceedings of the 27th European Conference on Pattern Languages of Programs, ser. EuroPLop ’22.   New York, NY, USA: Association for Computing Machinery, 2023. [Online]. Available: https://doi.org/10.1145/3551902.3564802
  57. “Information Technology - Artificial Intelligence - Management System,” International Organization for Standardization, Standard ISO/IEC 42001:2023, 2023. [Online]. Available: https://www.iso.org/standard/81230.html
  58. B. Xia, Q. Lu, L. Zhu, S. U. Lee, Y. Liu, and Z. Xing, “From principles to practice: An accountability metrics catalogue for managing ai risks,” arXiv preprint arXiv:2311.13158, 2023.
  59. B. Xia, Q. Lu, L. Zhu, and Z. Xing, “Towards ai safety: A taxonomy for ai system evaluation,” arXiv preprint arXiv:2404.05388, 2024.
Citations (7)
View on Semantic Scholar

Summary

  • The paper details a catalogue of 16 architectural patterns addressing goal creation, plan generation, and cooperativity in FM-based agents.
  • It employs a systematic literature review and real-world case studies to validate design approaches and evaluate trade-offs.
  • The work offers actionable design principles to enhance robustness, trustworthiness, and ethical compliance in agent systems.

Agent Design Pattern Catalogue: Architectural Patterns for Foundation Model Based Agents

Introduction

The paper "Agent Design Pattern Catalogue: A Collection of Architectural Patterns for Foundation Model based Agents" (2405.10467) introduces a pattern catalogue intended to guide the design of agents based on foundation models. This catalogue addresses prevalent challenges such as goal-seeking and plan generation while considering the inherent complexities and limitations of foundation models. The aim is to offer systematic and holistic guidance to practitioners in effectively implementing and utilizing FM-based agents.

The surge in Generative AI, bolstered by the capabilities of foundation models and LLMs, has stimulated the development of autonomous agents capable of proactive goal-seeking and task orchestration. However, there exists a steep learning curve in designing these agents due to the intricacies of architectural choices and the need for integration with ecosystems involving other agents, AI models, and non-AI systems. The presented catalogue categorizes 16 architectural patterns, offering insights into their contexts, benefits, and trade-offs, with the ultimate aim of fostering robust, trustworthy, and efficient agent implementations. Figure 1

Figure 1: Methodology.

Overview of Architectural Patterns

The catalogue is composed of 16 architectural patterns, each tailored to address specific challenges in the design of FM-based agents. These patterns encompass various aspects of agent design, including goal creation, plan generation, reasoning reflection, cooperation schemes, multimodal guardrails, and agent discovery. A summary of selected patterns includes:

  • Passive Goal Creator: Facilitates interactivity and clear goal identification through user prompts, enhancing goal-seeking and intuitiveness.
  • Proactive Goal Creator: Captures context through multimodal input, anticipating user goals and improving accessibility and interaction.
  • Prompt/Response Optimiser: Ensures standardisation and accuracy of outputs by refining prompts and responses according to user-determined templates.
  • Single-Path Plan Generator: Generates coherent and efficient single-path plans for task execution, reducing reasoning uncertainty.
  • Multi-Path Plan Generator: Allows flexibility by generating multiple pathways to achieve user goals, accommodating diverse preferences.
  • Voting-based Cooperation: Enables consensus among agents through voting, enhancing diversity and accountability.
  • Multimodal Guardrails: Ensures robustness and safety of foundation models by managing inputs and outputs to comply with standards and laws.

Table 1 provides a concise summary of these patterns, highlighting their benefits and potential drawbacks. Figure 2

Figure 2: Ecosystem of FM-based agent systems annotated with architectural patterns in gray boxes.

Methodological Approach

The research methodology encompasses a systematic literature review (SLR) that identified key challenges and existing solutions for FM-based agents. The process involved collecting and synthesizing data from a curated pool of studies, leading to the identification of reusable architectural patterns.

The methodology (Figure 1) involved assessing design challenges in FM-based agent ecosystems, analyzing current practices, and synthesizing knowledge into a pattern catalogue. This included extensive reviews and real-world case applications to validate the patterns and their applicability across different scenarios.

Design Principles and Practical Implications

The provided catalogue serves as a design tool not only for developing FM-based agents with enhanced goal-seeking capabilities but also for fostering cooperation among multiple agents and tools. The patterns offer pragmatic solutions to typical architectural challenges, such as reasoning uncertainty and complex accountability systems.

Each pattern addresses specific forces that complicate agent design, offering solutions with distinct benefits and trade-offs. The integration of multiple patterns can pave the way for robust, scalable, and responsible AI agent systems. Compatibility with responsible AI practices further ensures ethical considerations are embedded in design processes. Figure 3

Figure 3: Role-based cooperation.

Conclusion

The "Agent Design Pattern Catalogue" provides essential guidance for practitioners and architects aiming to develop foundation model-based agents. The inclusion of diverse patterns enables designers to choose appropriate architectural decisions to suit various software quality requirements, constraints, and application contexts. The catalogue not only supports the reliable deployment of FM-based agents but also aligns with broader objectives of AI safety and accountability.

The proposed catalogue, backed by extensive literature and real-world cases, highlights the potential for integrating existing architectural frameworks and responsible AI patterns to achieve trustworthy and effective agent systems. Future work may explore deeper interactions with extant patterns, detailed evaluations, and potential regulatory impacts on FM-based agent design.

Paper to Video (Beta)

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

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 10 tweets with 6 likes about this paper.

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