Pair Programming with Intelligent Social Agents

结对编程是两个程序员一起工作的一种编码实践。一个程序员（驱动程序）编写代码，而另一个程序员（观察者）在编写代码时检查每一行代码。二十年的计算机科学教育研究表明，结对编程显著提高了编程能力，增加了学生成为并继续主修计算机专业的可能性。人工智能（AI）正在改变现代生活，并有可能改变学生学习编程的方式。现代人工智能工具，如OpenAI的ChatGPT，可以在没有人为干预的情况下编写、总结、评论和解释程序。这个项目的目标是原型，测试和研究一个名为CoCode的结对编程人工智能代理，它可以作为学生的工具和导师。这个项目将帮助我们更好地理解基于人工智能的工具在计算机教育中的作用。该项目将探索各种基于理论的设计选项，并对语言模型和协作强化学习方面的最新突破进行创新，从而开发出AI配对程序员代理。这个代理将能够与学生程序员交互，并使用结对编程教学法指导他们。一系列的人体实验将有助于完善设计选择，并对人工智能算法进行基准测试和训练。该工具将在教育环境中进行评估。学生们将在一系列的试验中使用CoCode，以确定什么样的使用模式和学习成功是可能的长期互动。结果将是行为的，提供学生的任务如何变化的理解，因为他们学习和成为合格的程序员。这些结果将用于假设如何继续为教育环境开发这些技术，以及如何将它们纳入劳动力发展以重新培训现有工人。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Pair programming is a coding practice where two programmers work together. One programmer, the driver, writes code while the other programmer, the observer, reviews each line of code as it is written. Twenty years of CS education research has shown that pair programming significantly improves programming competency and increases the likelihood that students will become and remain computing majors. Artificial Intelligence (AI) is changing modern life and has the potential to change how students learn to program. Modern AI tools like OpenAI’s ChatGPT can write, summarize, comment, and explain programs without human intervention. The goal of this project is to prototype, test, and study a pair-programming AI agent named CoCode that can serve as a tool and tutor for students. This project will help us to better understand the roles AI-based tools play in computing education. This project will explore a variety of theory-informed design options and innovate on recent breakthroughs in language models and collaborative reinforcement learning resulting in the development of an AI pair-programmer agent. This agent will be able to interact with student programmers and guide them using the pair-programming pedagogy. A series of human subject experiments will serve both to refine design choices and benchmark and train the AI algorithms. The tool will be evaluated in an educational setting. Students will use CoCode in a series of trials to determine what use patterns and learning successes are possible over longer-term interaction. Outcomes will be behavioral, providing an understanding of how students’ tasks change as they learn and become competent programmers. These results will be used to postulate how such technologies will continue to be developed for educational settings and how they might be integrated into workforce development for the re-skilling of existing workers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.