研究目的

人工智能（Artificial intelligence，AI）在当下的发展被认为是重塑医学和全球卫生格局的一个重要里程碑，尤其是通过其改善全球健康差距的潜力。然而，与其潜在受益并行存在的是一系列复杂而具有挑战性的因素，这些因素不仅可能导致AI难以实现其承诺，甚至加剧全球卫生领域的不公正。本论文遵循“技术现象—权力结构—治理范式”的逻辑架构，核心研究问题是：AI对全球健康公平的挑战是什么？以及如何有效权衡和应对全球实践中公正困境？

研究方法

1. 人工智能技术风险与全球健康公平：问题识别。基于主题分析研究设计，根据主题相关性从4个数据库中筛选出378篇出版物，通过标题、摘要和全文审查以及参考文献补充检索后，最终纳入14篇文章进行主题分析。在反复阅读选定文章后，识别、分类和确定AI对全球健康公平的特有挑战，分析和解释AI在全球健康中的公平问题和上游决定因素。

2. 人工智能伦理规范缺陷与全球权力公正：话语批判。基于批判性话语分析方法，对2019年至2025年1月期间的18份全球性AI伦理文件以及5份具有代表性的大型科技企业伦理实践内容进行了讨论，旨在剖析全球AI伦理规范性文件和大型企业关于AI与全球健康公平的主导话语，揭示不同实体对AI全球影响的看法和塑造意图，为伦理分析论证提供全球背景。

3. 人工智能在全球健康中的价值重构和进路：伦理论证。基于实证生命伦理学研究方法和反思平衡方法，结合前两个阶段的研究，从受益风险、必要性、目的性和成本等方面对全球健康领域AI的公正问题进行伦理考量和价值判断。通过伦理分析和论证，构建AI在全球健康中的伦理矩阵模型、制衡机制和伦理冲突解决的优先级框架。

研究结果

1. 人工智能技术风险与全球健康公平：问题识别。尽管数据有限，但来自中低收入国家研究的稀缺性仍然明显。AI在全球健康中的特有公正主题确定为以下三类：（1）数据贫困和情境偏移。AI系统开发呈现群体偏向性以及在中低收入国家存在严重的模型泛化问题；（2）成本效益和健康公平。AI系统大规模部署的成本效益评估仍然有限，AI系统的使用可能加剧国家之间和国家内部现有的健康结果差异；（3）新技术殖民和潜在剥削。对利润的内在追求可能会导致AI技术公司优先开发适用于富裕国家疾病的工具，而非专注于解决在资源有限环境中使用AI相关的问题。

2. 人工智能伦理规范缺陷与全球权力公正：话语批判。全球文件在AI伦理认知上存在一定的共性和全球趋同。伦理原则往往集中在有限数量且短期影响上，长期伦理问题和全球性挑战存在被忽视或被边缘化的情况。大型科技公司主要关注技术系统安全性和个人层面的影响，对更广泛的社会责任或全球健康的强调有限。尽管一些AI规范和承诺可能是出于伦理的目的作出的，但行为主体在AI技术的开发和部署方面具有不确定性。因此，传统的伦理框架可能会在新的技术背景下失去适用性、权威性或可执行性。这种“消解”并非完全否定当前的伦理规范，而是需要重新审视、调整或重构伦理框架。

3. 人工智能在全球健康中的价值重构和进路：伦理论证。AI技术在特定健康挑战中具有不可替代性，但超出必要性限度的部署和应用会对人的主体性构成威胁，并扩大全球健康差距。全球健康中AI系统的大规模部署应进行严格的目的限制，即缩小健康差距并促进健康效益的最大化。考虑到AI可能对人的主体性、脆弱群体、生态正义和代际公平以及人类生存构成威胁，应建立基于人权的制衡机制，并充分考量中低收入国家卫生系统的承载能力和实际健康需求，以及根据差异原则进行AI的全球责任分配。

研究结论

AI是一项嵌入并受社会、政治和经济影响的技术，因此它本质上并非是中立或客观的，而是全球权力关系的镜像。虽然普遍观点认为AI将改善群体健康并减少健康差距，但由于巨大的数字鸿沟、健康不公平和社会结构性因素，AI在资源匮乏的环境中可能进一步加剧全球健康差距。AI是否促进代表性和健康公平不仅取决于技术本身，更取决于全球社会是否可以将卫生资源和权力向弱势群体倾斜以实现公平。因此，AI治理应聚焦于全球健康公平，以推动人类健康福祉和可持续发展目标的实现。

Objectives

Artificial intelligence (AI) is currently regarded as a significant milestone in reshaping the landscape of medicine and global health, particularly its potential to reduce global health disparities. However, alongside its potential benefits, a range of complex and challenging factors exist that may not only hinder AI from fulfilling its promises but also exacerbate injustices in global health. This thesis follows a logical framework of “technological phenomenon—power structure—governance paradigm.” The core research questions are: What challenges does AI posed by global health equity? And how to effectively balance and address the dilemmas of justice in global practice? 

Method

1. AI Technology Risks and Global Health Equity: Issue Identification. Using a thematic analysis research design, 378 publications were screened from four databases based on thematic relevance. Following title, abstract, and full-text reviews, supplemented by snowball sampling, 14 articles were included for thematic analysis. Through iterative reading of the selected articles, unique challenges posed by AI to global health equity were identified, categorized, and defined. The study analyzed and interpreted the equity issues and upstream determinants of AI in global health.

2. Deficiencies in AI Ethics Norms and Global Power Justice: Discourse Analysis. Using a critical discourse analysis approach, this study examined 18 global AI ethics documents and 5 ethical practices of representative large-scale technology companies from 2019 to January 2025. The aim is to dissect the dominant discourses in global AI ethics guidelines and major corporations regarding the issue of AI and global health equity, and to reveal the perceptions and shaping intentions of different entities towards the global impact of AI. This research provided a global context for ethical analysis and argumentation.

3. Reconceptualizing the Value of AI in Global Health and Pathways Forward: Ethical Justification. Based on the research methods of empirical bioethics and reflective equilibrium, and integrating the findings from the previous two stages, this study conducted ethical considerations and value judgments on the equity issues of AI in global health from aspects such as benefit-risk, necessity, purposefulness, and cost. Through ethical analysis and justification, an ethical matrix of AI in global health, a system of checks and balances, and a priority framework for resolving ethical conflicts were constructed.

Results

1. AI Technology Risks and Global Health Equity: Issue Identification. Despite limited data, the scarcity of research from low- and middle-income countries (LMICs) remains evident. The unique justice themes of AI in global health have been identified in three categories: (1) Data Poverty and Contextual Bias: AI system development exhibits group bias and severe model generalization issues in LMICs due to data poverty and contextual shifts. (2) Cost-Effectiveness and Health Equity: Cost-effectiveness assessments for large-scale deployment of AI systems are still limited, and the use of AI systems may exacerbate existing health outcome disparities both between and within countries. (3) New Technological Colonialism and Potential Exploitation: The inherent pursuit of profit may lead AI technology companies to prioritize tools for diseases in affluent countries rather than focusing on issues related to using AI in resource-limited settings.

2. Deficiencies in AI Ethics Norms and Global Power Justice: Discourse Analysis. Global AI ethics frameworks reveal emerging commonalities and a trend toward convergence. However, ethical principles often focus on a limited number of issues with short-term impacts, whereas long-term ethical concerns and global challenges are frequently overlooked or marginalized. Large technology firms primarily focus on technical system safety and individual-level effects, with limited emphasis on broader social responsibility or global health. Although some AI norms and commitments are ethically motivated, uncertainties persist in the actions of stakeholders during AI development and deployment. Consequently, traditional ethical frameworks may lose applicability, authority, or enforceability in novel technological contexts. This “dissolution” does not wholly invalidate existing norms but underscores the need for critical re-examination, refinement, or reconstruction of ethical frameworks.

3. Reconceptualizing the Value of AI in Global Health and Pathways Forward: Ethical Justification. While AI technology is irreplaceable in addressing specific health challenges, its deployment and application beyond the necessary limits can threaten human subjectivity and exacerbate global health disparities. The large-scale deployment of AI systems in global health should be subject to strict purpose limitation, namely narrowing health disparities and maximizing health benefits. Considering that AI may pose threats to human subjectivity, vulnerable groups, the ecological justice, intergenerational equity, and human survival, a human rights-based system of checks and balances should be established. The capacity of health systems in LMICs, their actual health needs, and the global responsibility allocation of AI based on the principle of difference should be fully considered.

Conclusions

AI is a technology embedded in and influenced by social, political, and economic factors. Therefore, it is not inherently neutral or objective but rather a reflection of global power relations. Although the common perception is that AI will improve population health and reduce health disparities, the reality is more complex. Due to the significant digital divide, existing health inequities, and structural social factors, AI may further exacerbate global health disparities in resource-poor settings. Whether AI promotes representation and health equity depends not only on the technology itself but more importantly on whether global society can reallocate health resources and power towards vulnerable groups to achieve fairness. Thus, AI governance should focus on global health equity to promote human health and well-being and the achievement of sustainable development goals.