Improving Productivity in Software Development: How Humans and AI Need to Join Forces

今天的世界运行并依赖于软件。支持和支持这些软件的创造者是我工作背后的动力。我提出的研究旨在理解和指导如何使用人工智能（AI）来提高开发人员的生产力。这一提议是及时的，因为人工智能对几乎所有科学和知识领域都产生了破坏性影响，软件工程也不例外。人工智能的好处很容易想象，因为软件开发过程的许多方面都会生成和捕获人工智能所依赖的有用数据。例如，我们预测软件错误并使用开发人员跟踪数据和用户生成数据自动修复它们的能力是许多公司的游戏规则改变者。但也有需要注意的风险，包括隐私和数据权利。滥用自动化也可能会影响公平性（不生成跟踪数据的开发人员任务可能不会被识别），降低可能对AI缺乏信任的开发人员的工作满意度，并且AI可能会对社区多样性产生负面影响，因为AI可能会为某些人口统计数据的参与引入偏见和障碍。女性在软件开发中的参与度低已经成为一个问题。这些问题可能会对软件质量产生下游的负面影响。一些人担心人工智能可能会取代开发人员的工作，但另一些人则认为，最令人担忧的是，人类可能会高估和信任机器，给它们太多的自主权。开发人员的生产力是不容易定义的，而且无论是通过工具还是过程，要提高它就更难了。开发人员的生产力不应该天真地通过编写的代码行数、交付的特性数量或其他简单的定量度量来衡量。这些措施未能抓住发展的创造性和协作性方面。因此，改进软件工程需要对行为方面有丰富而细致的理解，以及跨越无数开发环境的技术可能性。改进工程工具和工程管道是很重要的，但是开发者技术是如何被塑造的，反过来又是如何塑造行为和社会方面的，这一点也不能被忽视。开发人员的生产力受到技术和社会因素的影响，两者应该在持续的基础上共同发展，从而共同优化。例如，在考虑到人的方面之前，人工智能在自动错误修复方面的成功和采用并没有发生。我的研究计划有三个目标：1)我将提出并推广适合于研究人类和社会行为与为软件工程设计的人工智能技术共同进化的研究方法；2)我将发展社会技术理论，以捕捉人工智能和数据驱动的自动化如何影响和可能受到软件工程中人类和社会行为的影响；3)我将设计工具和新流程，以提高个人、团队和开发人员社区的生产力和满意度。

Today's world runs and depends on software. Supporting and enabling the creators of that software is the driving passion behind my work. My proposed research aims to understand and guide how artificial intelligence (AI) can be used to improve developer productivity. This proposal is timely as AI is having a disruptive effect on almost every scientific and knowledge domain, and software engineering is no exception. The benefits of AI are easy to imagine as so many aspects of the software development process generate and capture useful data, which AI relies on. For example, our ability to predict software bugs and automatically repair them using developer trace data and user generated data is a game changer for many companies. But there are risks to be aware of, including privacy and data rights. Misuse of automation may also impact fairness (developer tasks that do not generate trace data may not be recognized), reduced work satisfaction for developers who may lack trust with the AI, and AI that may negatively impact community diversity as AI may introduce biases and barriers to participation for some demographics. Low female participation in software development is already a concern. These issues may then have downstream negative effects on software quality. Some worry that AI may take over developer jobs, but others feel the biggest concern is that humans may overestimate and trust machines and give them too much autonomy. Developer productivity is not easy to define, and it is even harder to improve, whether through tools or processes. Developer productivity should not be naively measured by lines of code authored, number of features delivered or other such simple quantitative measures. These measures fail to capture the creative and collaborative aspects of development. Thus, improving software engineering calls for a rich and nuanced understanding of behavioural aspects, as well as technical possibilities across a myriad of development contexts. Improving engineering tools and the engineering pipeline is important, but how developer technologies are shaped by and in turn how they shape behavioural and social aspects must not be ignored. Developer productivity is impacted by both technical and social factors, and both should be jointly optimized as they co-evolve on a continual basis. For example, the success and adoption of AI for automated bug repair did not happen until human aspects were considered. The goals of my research program are threefold: 1) I will propose and promote research methods that are suitable for studying the co-evolution of human and social behaviours with AI technologies designed for software engineering; 2) I will develop socio-technical theories to capture how AI and data driven automation impacts and may be impacted by human and social behaviours in software engineering; and 3) I will design tools and new processes to improve productivity and satisfaction for individuals, teams and communities of developers.