Generative AI enhances individual creativity but reduces the collective diversity of novel content

Creativity is core to being human. Generative AI—made readily available by powerful large language models (LLMs)—holds promise for humans to be more creative by offering new ideas, or less creative by anchoring on generative AI ideas. We study the causal impact of generative AI ideas on the production of short stories in an online experiment where some writers obtained story ideas from an LLM. We find that access to generative AI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers. However, generative AI-enabled stories are more similar to each other than stories by humans alone. These results point to an increase in individual creativity at the risk of losing collective novelty. This dynamic resembles a social dilemma: with generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced. Our results have implications for researchers, policy-makers, and practi..., This dataset is based on a pre-registered, two-phase experimental online study. In the first phase of our study, we recruited a group of N=293 participants (“writers”) who are asked to write a short, eight sentence story. Participants are randomly assigned to one of three conditions: Human only, Human with 1 GenAI idea, and Human with 5 GenAI ideas. In our Human only baseline condition, writers are assigned the task with no mention of or access to GenAI. In the two GenAI conditions, we provide writers with the option to call upon a GenAI technology (OpenAI’s GPT-4 model) to provide a three-sentence starting idea to inspire their own story writing. In one of the two GenAI conditions (Human with 5 GenAI ideas), writers can choose to receive up to five GenAI ideas, each providing a possibly different inspiration for their story. After completing their story, writers are asked to self-evaluate their story on novelty, usefulness, and several emotional characteristics. In the second phase, th..., , # Dataset and Code for \"Generative artificial intelligence enhances creativity but reduces the diversity of novel content\" ### by Anil R. Doshi and Oliver P. Hauser ## Introduction We recommend downloading the file \"GenAI\_creativity\_data\_and\_scripts.zip\" which contains all data (raw and processed) as well as the analysis code. Then please follow the steps below. We provide two methods to perform the data analysis. 1. Compile all files. This method processes the raw csv files and performs the analysis. Requires some knowledge of Python and an API key to OpenAI. 2. Processed file analysis. This allows you to run the analysis on the already processed dta files. ## 1\. Compile all files method If you would like to compile all files, please follow these steps to ensure your machine is set up to run all the necessary scripts. ### Machine setup 1. Ensure Python is installed (tested with Python 3) 2. Install the following packages in Python * numpy * scipy * openai It may be nece...创造力是人类的核心特质。由强大的大语言模型（Large Language Models，LLMs）广泛普及的生成式AI（Generative AI），既可为人类提供新灵感以提升创造力，也可能让创作者锚定生成式AI的思路而削弱创造力。我们通过一项在线实验研究了生成式AI思路对短篇小说创作的因果影响：部分创作者可从大语言模型获取故事灵感。我们发现，获取生成式AI思路会让故事被评价为更具创造性、文笔更佳且更具观赏性，尤其在创造力较弱的创作者中效果更显著。然而，借助生成式AI创作的故事彼此间的相似度，高于仅由人类独立创作的故事。上述结果表明，个体创造力有所提升，但同时可能丧失集体层面的新颖性。这一动态过程类似于社会困境：借助生成式AI时，创作者个体境况更佳，但集体层面产出的新颖内容范畴却更狭窄。我们的研究结果对研究者、政策制定者及从业者具有启示意义…… 本数据集基于一项预先注册的两阶段在线实验研究。在研究的第一阶段，我们招募了N=293名参与者（即“创作者”），要求他们创作一篇由8个句子组成的短篇故事。参与者被随机分配至三种实验条件：仅人类创作组、获取1个生成式AI灵感组，以及获取5个生成式AI灵感组。在仅人类创作的基线条件中，创作者仅被分配创作任务，未提及或无法使用生成式AI。在两个生成式AI辅助创作条件中，我们为创作者提供调用生成式AI技术（OpenAI的GPT-4模型）的选项，以获取一段三句话的开篇灵感，用于启发自身的故事创作。在其中一个生成式AI条件（即获取5个生成式AI灵感组）中，创作者可选择获取最多5个生成式AI灵感，每个灵感均可为其故事创作提供差异化的启发。完成故事创作后，创作者需就故事的新颖性、实用性及多项情感特征进行自我评估。 在第二阶段，…… # 题为《生成式人工智能提升创造力却降低新颖内容多样性》的数据集与代码 作者：阿尼尔·R·多西（Anil R. Doshi）与奥利弗·P·豪泽（Oliver P. Hauser） ## 引言 我们建议下载文件"GenAI_creativity_data_and_scripts.zip"，其中包含所有原始与处理后的数据以及分析代码。请按照以下步骤操作。 我们提供两种数据分析方法： 1. 全文件编译法：该方法将处理原始CSV文件并执行分析，需具备一定Python编程知识以及OpenAI API密钥。 2. 已处理文件分析法：该方法可直接对已处理的dta格式文件执行分析。 ## 1. 全文件编译法 若您希望采用全文件编译法，请按照以下步骤配置您的设备，以运行所有必要脚本。 ### 设备配置 1. 确保已安装Python（已在Python 3环境下测试） 2. 在Python中安装以下依赖包： * numpy * scipy * openai 这可能需要……