arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2501.09434cs.SE
收藏

面向AI系统的敏捷系统开发生命周期:决策架构

Agile System Development Lifecycle for AI Systems: Decision Architecture

Asif Q. Gill

AI总结:

针对传统敏捷SDLC无法适配AI系统开发需求的问题,本文引入决策科学增强敏捷SDLC,聚焦决策自动化场景下的决策架构设计,并通过保险理赔示例验证可行性,为该新领域研究奠定初步基础。

摘要:

敏捷系统开发生命周期(SDLC)聚焦于传统软件系统开发中的典型功能与非功能系统需求。但人工智能(AI)系统本质上存在差异,具备六项独特属性:(1)自主性、(2)自适应性、(3)内容生成、(4)决策制定、(5)可预测性、(6)推荐能力。现有敏捷SDLC需加以增强,才能支撑AI系统的开发及部署后的持续适配工作,当前面临的挑战是:如何增强敏捷SDLC以适配AI系统需求。本文的研究范围限定于AI系统赋能的决策自动化场景,因此提出利用决策科学(Decision Science)增强敏捷SDLC,以支撑AI系统开发。决策科学是研究决策制定的学科,可用于识别、分析和描述待通过AI系统实现自动化的决策及其架构。具体而言,本文在面向AI系统的敏捷SDLC整体框架下,详细探讨了决策架构的相关内容。研究通过保险理赔处理的示例场景,验证了所提方法的应用可行性。这项初步工作表明,决策科学可用于增强敏捷SDLC,以设计和实现面向决策自动化的AI系统,为决策架构与AI系统敏捷SDLC这一新领域的后续研究奠定了初步基础。

英文摘要:

Agile system development life cycle (SDLC) focuses on typical functional and non-functional system requirements for developing traditional software systems. However, Artificial Intelligent (AI) systems are different in nature and have distinct attributes such as (1) autonomy, (2) adaptiveness, (3) content generation, (4) decision-making, (5) predictability and (6) recommendation. Agile SDLC needs to be enhanced to support the AI system development and ongoing post-deployment adaptation. The challenge is: how can agile SDLC be enhanced to support AI systems? The scope of this paper is limited to AI system enabled decision automation. Thus, this paper proposes the use of decision science to enhance the agile SDLC to support the AI system development. Decision science is the study of decision-making, which seems useful to identify, analyse and describe decisions and their architecture subject to automation via AI systems. Specifically, this paper discusses the decision architecture in detail within the overall context of agile SDLC for AI systems. The application of the proposed approach is demonstrated with the help of an example scenario of insurance claim processing. This initial work indicated the usability of a decision science to enhancing the agile SDLC for designing and implementing the AI systems for decision-automation. This work provides an initial foundation for further work in this new area of decision architecture and agile SDLC for AI systems.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑