When the Machine Is Not Enough: An Autoethnography of AI-Assisted Amateur Research and the Boundaries of Epistemic Authority
收藏资源简介：
Title:Analysis Data for "When the Machine Is Not Enough: An Autoethnography of AI-Assisted Amateur Research and the Boundaries of Epistemic Authority" Description:This deposit contains the R analysis script and coded output files for the keyword-based sentiment analysis of AI conversation logs reported in the accompanying manuscript, submitted to Science, Technology, & Human Values. The dataset comprises 4,459 assistant messages from 101 conversations with OpenAI's ChatGPT-4o across three cross-disciplinary research projects (Collatz conjecture, Information-Topological Cosmology, and Creative Singularity Triangle), conducted between 2024 and 2025. Each message was coded for encouragement, criticism, hedging, and superlative keywords in both Korean and English, following an iterative calibration procedure described in the manuscript's Supplementary Material. Files:- three_project_analysis_v3_2.R: R analysis script (requires tidyverse and lubridate). Reads per-project CSV files and produces coded output with keyword counts, tone classification (encouragement-only / criticism-only / mixed / neutral), and length-normalized metrics.- all_projects_coded_v3.csv: Per-message coding results for all 4,459 assistant messages, including keyword counts, tone classification, and length-normalized metrics (encouragement and criticism counts per 1,000 characters).- manuscript_table_v3.csv: Aggregated project-level statistics reported in Table 2 and related analyses in the main text. Note: The raw conversation logs (containing personal and bilingual dialogue) are available from the corresponding author upon reasonable request. Keywords:artificial intelligence, large language models, expertise, autoethnography, sentiment analysis, peer review, RLHF, sycophancy Related publications:Oh, Y.T. (2025a). Collatz Convergence through Logarithmic Descent and Modular Reduction. Zenodo. https://doi.org/10.5281/zenodo.15484231Oh, Y.T. (2025b). Information-Topological Cosmology: A Thermodynamic Alternative to ΛCDM Unified by Entropy Production. Zenodo. https://doi.org/10.5281/zenodo.15446634Oh, Y.T. (2025c). The Creative Singularity Triangle: A Participatory Model for Post-Technological Innovation. Zenodo. https://doi.org/10.5281/zenodo.15512484
标题：《当机器力有不逮：AI辅助业余研究的自传式民族志与认知权威的边界》分析数据集 描述：本数据存档包含用于AI对话日志基于关键词的情感分析的R分析脚本与编码输出文件，相关成果已提交至《科学、技术与人类价值（Science, Technology, & Human Values）》期刊的配套学术论文。 本数据集涵盖2024至2025年间，在三项跨学科研究项目（考拉兹猜想、信息拓扑宇宙学、创意奇点三角）中，与OpenAI的ChatGPT-4o开展的101轮对话所生成的4459条助手回复消息。依据论文补充材料中详述的迭代校准流程，所有消息均按韩、英双语，针对鼓励类、批评类、模糊限制语（Hedging）以及最高级类关键词完成编码。 文件说明： - three_project_analysis_v3_2.R：R分析脚本（需依赖tidyverse与lubridate扩展包），可读取各项目的CSV文件，并生成包含关键词计数、语气分类（仅鼓励/仅批评/混合/中性）以及长度归一化指标的编码输出结果。 - all_projects_coded_v3.csv：全部4459条助手消息的逐消息编码结果，涵盖关键词计数、语气分类以及长度归一化指标（每1000字符的鼓励与批评计数）。 - manuscript_table_v3.csv：论文正文表2及相关分析中报告的聚合项目级统计数据。 注：包含个人隐私信息与双语对话的原始对话日志，可经合理申请后向通讯作者获取。 关键词：人工智能、大语言模型（Large Language Model，LLM）、专业专长（Expertise）、自传式民族志（Autoethnography）、情感分析（Sentiment Analysis）、同行评审（Peer Review）、基于人类反馈的强化学习（RLHF）、奉承倾向（Sycophancy） 相关出版物： Oh, Y.T. (2025a). 《基于对数下降与模约简的考拉兹收敛性》. Zenodo. https://doi.org/10.5281/zenodo.15484231 Oh, Y.T. (2025b). 《信息拓扑宇宙学：以熵产生统一ΛCDM模型的热力学替代框架》. Zenodo. https://doi.org/10.5281/zenodo.15446634 Oh, Y.T. (2025c). 《创意奇点三角：后技术创新的参与式模型》. Zenodo. https://doi.org/10.5281/zenodo.15512484
