As an organiser of the Dagstuhl seminar, I enjoyed a week of roadmap setting for Software Architecture & Machine Learning. Here is the talk I delivered.
Hand-written abstract for historical archive in the castle!
With the successful implementation of Large Language Models (LLMs) in chatbots like ChatGPT, there is growing attention on foundation models, which are anticipated to serve as core components in the development of future AI systems. Yet, systematic exploration into the design of foundation model-based systems, particularly concerning risk management, trust, and trustworthiness, remains limited. In this talk, I propose the challenges and initial approaches in both architecting LLM-based systems and how LLM systems have an impact on software engineering. I point to some initial directions such as architecting as a process of understanding (rather than designing/building), setting and trade-offing guardrails (rather than quality attributes), and radical observability.
References
Zhu, L., Xu, X., Lu, Q., Governatori, G., Whittle, J., 2022. AI and Ethics—Operationalizing Responsible AI, in: Chen, F., Zhou, J. (Eds.), Humanity Driven AI. Springer International Publishing, Cham, pp. 15–33. https://doi.org/10.1007/978-3-030-72188-6_2
Lu, Q., Zhu, L., Xu, X., Whittle, J., Xing, Z., 2022. Towards a Roadmap on Software Engineering for Responsible AI, in: 1st International Conference on AI Engineering – Software Engineering for AI (CAIN). https://dl.acm.org/doi/abs/10.1145/3522664.3528607
Lu, Q., 2023. Towards Responsible AI in the Era of ChatGPT: Pattern-Oriented Reference Architecture for Designing Foundation Model based AI Systems. https://arxiv.org/abs/2304.11090
Lu, Q., Zhu, L., Xu, X., Xing, Z., Whittle, J., 2023. A Framework for Designing Foundation Model based Systems URL https://arxiv.org/abs/2305.05352v1
Lo, S.K., Liu, Y., Lu, Q., Wang, C., Xu, X., Paik, H.-Y., Zhu, L., 2023. Toward Trustworthy AI: Blockchain-Based Architecture Design for Accountability and Fairness of Federated Learning Systems. IEEE Internet of Things Journal 10, 3276–3284. https://doi.org/10.1109/JIOT.2022.3144450
Lo, S.K., Liu, Y., Lu, Q., Wang, C., Xu, X., Paik, H.-Y., Zhu, L., 2021. Blockchain-based Trustworthy Federated Learning Architecture. https://doi.org/10.48550/arXiv.2108.06912
Xia, B., Bi, T., Xing, Z., Lu, Q., Zhu, L., 2023. An Empirical Study on Software Bill of Materials: Where We Stand and the Road Ahead. Presented at the ICSE, arXiv. https://doi.org/10.48550/arXiv.2301.05362
Xu, X., Wang, C., Wang, Z. (Jef), Lu, Q., Zhu, L., 2022. Dependency tracking for risk mitigation in machine learning (ML) systems, in: Proceedings of the 44th International Conference on Software Engineering: Software Engineering in Practice,https://doi.org/10.1145/3510457.3513058
