CAREER: Enhancing Diversity and Personalization in Human-AI Collaborative Writing

像大型语言模型(LLM)这样的生成性人工智能技术正在迅速改变人们创造内容的方式。通过起草、编辑和建议文本，基于LLM的写作助手具有改善写作质量和提高作者生产力的潜力。然而，由于数以百万计的用户依赖相同的基础模型来生成文本，因此存在内容创建同质化的潜在风险--导致内容相似度增加，内容多样性总体降低。该项目旨在衡量基于LLM的写作助理对内容多样性的影响，并为下一代写作助理开发增强(而不是取代)个人声音的方法。除了技术贡献外，该项目还将为社会科学家和政策制定者提供关于监管生成性人工智能技术的见解和最佳实践。这项研究的结果也将通过教学和研究活动整合到本科生和研究生的学习中。首先，研究人员旨在通过量化联合写作如何在个人属性和整体多样性方面改变产生的内容来理解使用LLMS写作的意想不到的影响。在这次调查的基础上，下一个目标是通过探索促进以人为中心的写作助理的计算方法来解决已确定的问题。主要的方法包括微调具有多样性意识的目标的LLM，以及在线调整LLM以学习并适应每个用户在写作过程中的偏好。总体而言，该项目将产生有助于更多以人为中心的写作辅助的指标、数据集和方法。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Generative AI technologies like large language models (LLMs) are rapidly changing how people create content. By drafting, editing, and suggesting text, LLM-based writing assistants have the potential to improve writing quality and increase author productivity. However, as millions of users rely on the same underlying model to produce text, there is a potential risk of homogenizing content creation - resulting in increased content similarity and an overall reduction in content diversity. This project aims to measure the impact of LLM-based writing assistants on content diversity and develop methods for the next-generation writing assistants that enhance (as opposed to replace) personal voices. Aside from the technical contributions, this project will provide insights and best practices to social scientists and policymakers on regulating generative AI technologies. The result of this research will also be integrated in undergraduate and graduate studies through both teaching and research activities.The proposed research activities consist of three directions. First, the researchers aim to understand the unintended effects of writing with LLMs by quantifying how co-writing alters the produced content in terms of personal attributes and overall diversity. Building upon the insights gained from this investigation, the next objective is to address the identified issues by exploring computational methods that promote human-centered writing assistants. The main approaches include finetuning LLMs with a diversity-aware objective and adapting LLMs online to learn and suit each user's preference during writing. Overall, this project will produce metrics, datasets, and methods that contribute to more human-centered writing assistance.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.