REGULATION OF ARTIFICIAL INTELLIGENCE: BETWEEN THE SWITCH AND INNOVATION – IS BRAZIL PREPARED? Supplementary Material: Detailed Document Selection and Coding Protocol

This dataset -Supplementary Material A: Detailed Document Selection and Coding Protocol from the study “Regulation of Artificial Intelligence: Between the Switch and Innovation – Is Brazil Prepared?” — provides a systematically organized corpus of 52 official and analytical documents on AI governance across eight jurisdictions: Brazil (focal case), European Union, United States, United Kingdom, India, Chile, Mexico, and China. The dataset was constructed through a transparent five-phase retrieval process involving government portals, legal databases, regulatory repositories, and cross-referenced citation tracking. Inclusion criteria ensured institutional relevance, public accessibility, temporal validity (2018–2025), and thematic focus on AI governance. Documents were selected from governmental, regulatory, academic, civil society, and media sources to ensure multi-perspective representation. Weighting schemes (e.g., 60% government vs. 40% independent sources) adjusted for jurisdictional disparities in data availability. Each document was coded in MAXQDA using a hierarchical framework grounded in Lawrence Lessig’s four regulatory modalities — law, architecture, market, and social norms — complemented by institutional theory, economic regulation, and OECD’s Regulatory Impact Assessment (RIA) methodology. Subcodes captured compliance mechanisms, risk classification, incentives, fairness, and governance coordination. Inter-coder reliability was established through double-coding and reconciliation logs, with triangulation across institutional types. The dataset’s central table (Table SP1, pp. 4–11) classifies all documents by country, source institution, document title, and type (legal, regulatory, strategic, analytical, or media). For Brazil, ten documents cover Bill 2338/2023, ANPD technical notes, civil-society assessments (e.g., Data Privacy Brasil, CGI.br), and media commentary. The EU section lists six foundational sources including the AI Act, GDPR, and Ethics Guidelines for Trustworthy AI. The U.S. subset includes the AI Bill of Rights, NIST’s AI Risk Management Framework, and executive policy orders. The UK materials span the National AI Strategy and AI Regulation White Paper (2023). Comparable sets exist for India (NITI Aayog frameworks), Chile (MinCiencia 2024 policy updates), Mexico (Agenda IA 2030 and INAI guidelines), and China (CAC regulations and MIIT standards). Analytically, the dataset supports comparative cross-country evaluation of regulatory architectures, innovation incentives, and societal norms shaping AI governance. Brazil’s expanded corpus enabled a detailed OECD-style RIA of Bill 2338/2023, assessing problem definition, policy alternatives, stakeholder impacts, feasibility, and cost–benefit distribution. Validation procedures included expert reviews and cross-benchmarking with OECD and ECLAC reports.本数据集——作为研究《人工智能规制：在转型与创新之间——巴西是否已做好准备？》的补充材料A：详细文件遴选与编码规程——提供了一套系统整理的语料库，涵盖巴西（核心研究案例）、欧盟、美国、英国、印度、智利、墨西哥及中国共8个法域的52份人工智能（Artificial Intelligence，AI）治理相关官方文件与分析材料。 本数据集通过透明的五阶段检索流程构建完成，检索渠道涵盖政府门户网站、法律数据库、监管知识库以及交叉引用追踪。遴选标准确保了文件的制度相关性、公共可获取性、时间有效性（2018年至2025年），以及聚焦人工智能治理的主题方向。文件遴选自政府、监管机构、学术团体、公民社会及媒体渠道，以保障多视角的代表性。通过权重分配方案（如政府来源占比60%，独立来源占比40%），对不同法域的数据可得性差异进行了调整。 所有文件均采用MAXQDA编码工具，基于劳伦斯·莱斯格（Lawrence Lessig）提出的四大规制模态——法律、架构、市场与社会规范，结合制度理论、经济规制理论以及经济合作与发展组织（Organisation for Economic Co-operation and Development，OECD）的规制影响评估（Regulatory Impact Assessment，RIA）方法，构建了层级化编码框架。子编码涵盖合规机制、风险分类、激励措施、公平性及治理协调等维度。通过双编码与协调日志确立了编码者间信度，并通过不同制度类型的三角验证予以巩固。 本数据集的核心表格（SP1表，第4至11页）按照国家、来源机构、文件标题及类型（法律文件、监管文件、战略文件、分析文件或媒体材料）对所有文件进行了分类。针对巴西，收录的10份文件涵盖第2338/2023号法案、巴西国家数据保护局（Agência Nacional de Proteção de Dados，ANPD）技术说明、公民社会评估报告（如巴西数据隐私组织、巴西互联网治理委员会CGI.br）以及媒体评论。欧盟板块收录了包括《人工智能法案》《通用数据保护条例》（General Data Protection Regulation，GDPR）及《可信赖人工智能伦理指南》在内的6份基础性文件。美国板块包含《人工智能权利法案》、美国国家标准与技术研究院（National Institute of Standards and Technology，NIST）人工智能风险管理框架以及行政政策指令。英国相关材料涵盖《国家人工智能战略》及2023年《人工智能规制白皮书》。其他法域亦设有可比数据集：印度包含印度国家转型研究院（NITI Aayog）框架文件，智利包含智利科学与技术部（MinCiencia）2024年政策更新文件，墨西哥包含《人工智能2030议程》及墨西哥国家透明度、信息获取与个人数据保护委员会（Instituto Nacional de Transparencia, Acceso a la Información y Protección de Datos Personales，INAI）指南，中国包含中国网络空间管理局（Cyberspace Administration of China，CAC）规制文件及工业和信息化部（Ministry of Industry and Information Technology，MIIT）标准。 从分析层面而言，本数据集可用于开展各国规制架构、创新激励及塑造人工智能治理的社会规范的比较性跨国评估。针对巴西的扩充语料库支持了基于OECD规制影响评估方法的第2338/2023号法案详细分析，涵盖问题界定、政策备选方案、利益相关方影响、可行性及成本收益分配等维度。验证流程包含专家评审以及与OECD和拉丁美洲经济委员会（Economic Commission for Latin America and the Caribbean，ECLAC）报告的交叉基准比对。