SoD-TEAM: Teleological Reasoning in Adaptive Software Design

对物理设备概念设计的人工智能研究揭示了目的论知识和推理在自适应设计中的核心作用。物理设备的声明性目的论模型明确地捕捉其设计的结构、行为和功能之间的目的论关系，使结构的修改本地化以实现新功能。目的论模型的本体提供了一个词汇表，用于分类、表示、索引和访问特定的设计案例、通用的适应计划、原始设计组件和抽象设计模式。与物理设备类似，PI将软件工件视为抽象设备，也就是说，作为具有结构和行为的抽象目的论工件，这些结构和行为导致了所需功能的完成。对适应性物理设计的进一步类比导致目的论是适应性软件设计的基本组织原则的假设。当然，软件工件也与物理设备非常不同，因为这两个领域中的结构、行为、功能和目的论的概念是不同的。因此，这个项目将解决以下六个问题：在软件设计中，结构、行为、功能和目的论的生产性概念是什么？软件工件的声明性目的论模型如何指定它的功能体系结构？目的分析如何使软件工件的体系结构所需的修改本地化，以实现新的功能？软件工件的目的论模型的本体论如何支持分类、表示、索引和检索通用适应计划、原始软件组件和抽象设计模式？什么是目的论驱动的自适应软件设计过程？如何使用这些构建块来比现在更容易、更高效、更有效地支持交互式软件设计？为了回答这些问题，PI将以知识表示、分类法、库、方法和过程的形式开发自适应软件设计中的目的论推理的构建块。为此，PI采取了三管齐下的方法。他将对交互式策略游戏（如Freeciv和C-evo）的游戏软件代理设计的进化史进行详细的实证案例研究。他将开发一个交互式软件设计环境，并针对游戏代理的功能需求中的一小部分变化，自动调整设计，这将有助于确保所有构建块都能很好地协同工作。他将利用本科软件工程课程中的交互环境来支持设计适应游戏代理的功能需求的大量变化，并使用对照组进行正式实验来评估目的论推理在支持适应性软件设计中的功效。更广泛的影响：计算工件，包括软件工件，越来越普遍和无处不在。软件开发成本中的一个主要因素是随着设计需求的发展维护软件工件；即使这个成本的很小的减少也可能对软件实践产生重大的影响。由于本研究将解决软件设计适应功能设计需求变化的问题，因此其潜在应用范围广泛且开放，游戏软件代理的设计和开发只是一个例子。因为在这个项目中开发的交互式软件设计新工具将被引入课堂，它们将对计算机科学中使用的教育技术产生直接影响。从项目中获得的知识将在软件工程、设计认知与计算、人工智能和游戏等领域的会议和期刊上传播。

AI research on conceptual design of physical devices has revealed the central role that teleological knowledge and reasoning play in adaptive design. A declarative teleological model of a physical device that explicitly captures the teleological relationships among the structure, behaviors and functions of its design enables localization of the modifications to the structure to achieve new functions. The ontology of teleological models provides a vocabulary for classifying, representing, indexing and accessing specific design cases, generic adaptation plans, primitive design components, and abstract design patterns. In analogy to physical devices, the PI views software artifacts as abstract devices, i.e., as abstract teleological artifacts with structures and behaviors that result in the accomplishment of desired functions. A further analogy to adaptive physical design leads to the hypothesis that teleology is a fundamental organizational principle of adaptive software design. Of course, software artifacts are also quite different from physical devices because the notions of structure, behavior, function, and teleology in the two domains are different. Thus, this project will address the following six questions: What are productive notions of structure, behavior, function, and teleology in software design? How might a declarative teleological model of a software artifact specify its functional architecture? How might teleological analysis enable localization of the modifications needed to the architecture of a software artifact to achieve new functions? How might the ontology of the teleological models of software artifacts support classification, representation, indexing and retrieval of generic adaptation plans, primitive software components, and abstract design patterns? What might be a teleology-driven process of adaptive software design? And how might these building blocks be used to support interactive software design more easily, efficiently and effectively than at present? To answer these questions, the PI will develop building blocks of teleological reasoning in adaptive software design in the form of knowledge representations, taxonomies, libraries, methods, and processes. To these ends, the PI has adopted a three-pronged approach. He will conduct detailed empirical case studies of the evolutionary histories of designs of game-playing software agents for interactive strategy games such as Freeciv and C-evo. He will develop an interactive software design environment and, for a small class of changes in the functional requirements of game-playing agents, automate the design adaptations, which will help insure that all the building blocks work well together. And he will use the interactive environment in undergraduate software engineering courses to support design adaptations for a large class of changes in the functional requirements of game-playing agents, and conduct formal experiments using control groups to evaluate the efficacy of teleological reasoning in support of adaptive software design.Broader Impacts: Computational artifacts, including software artifacts, are increasingly pervasive and ubiquitous. A major factor in the cost of software development is maintaining the software artifact as its design requirements evolve; even a small reduction in this cost may have a significant impact on software practice. Since this research will address the issue of software design adaptations in response to changes in functional design requirements, its potential applications are wide-ranging and open-ended, the design and development of game-playing software agents being just one example. Because the new tools developed in this project for interactive software design weill be introduced into the classroom, they will have a direct impact on the educational technology used in the computing sciences. Knowledge acquired from the project will be disseminated in conferences and journals in software engineering, design cognition and computing, AI, and game playing.