Stakeholder Guidance to Anticipate and Address Ethical Challenges in Applications of Machine Learning and Artificial Intelligence in Algorithmic Medicine: a Novel Empirical Approach

利益相关者指导预测和解决机器学习和人工智能在算法医学中的应用中的伦理挑战：一种新颖的经验方法

• 批准号: 10367404
• 负责人: Jane Paik Kim
• 金额: $ 15.74万
• 依托单位: STANFORD UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-08-17 至 2024-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10367404
• 关键词: Address; Administrative Supplement; Affect; Algorithms; Area; Artificial Intelligence; Attention; Behavior Therapy; Car Phone; Code; Communities; Computers; Data; Data Collection; Data Reporting; Decision Making; Development; Devices; Dimensions; Environment; Ethical Issues; Ethics; Exercise; Face; Failure; Feedback; Funding; Furniture; Future; Goals; Health; Health Technology; Healthcare; Individual; Instruction; Intervention; Interview; Learning; Legal; Location; Machine Learning; Manuscripts; Measurement; Measures; Medicine; Methodology; Methods; Moods; Morals; Movement; Mus; Nature; Occupational; Occupational Health; Outcome; Parents; Participant; Patient Self-Report; Patients; Personal Computers; Persons; Policies; Policy Maker; Precision Health; Privacy; Public Health; Published Comment; Qualitative Methods; Research; Research Personnel; Resources; Role; Societies; Stress; Structure; System; Technology; Thinking; Time; Translations; United States National Institutes of Health; Universities; Voice; Work; authority; base; behavioral health; clinical decision support; data privacy; design; digital; digital health; digital intervention; evidence base; experience; health management; improved; innovation; innovative technologies; interest; machine learning algorithm; novel; parent grant; population health; precision medicine; real world application; response; stress management; surveillance data; theories; tool; two-dimensional

PROJECT ABSTRACT The potential for artificial intelligence applications to enable more granular and pervasive measurement, prediction, and provide behavioral interventions offers immense promise in reaching the goal of precision health to maintain the overall health of populations. When applied to devices encountered in our everyday environment, (e.g. personal computers, mobile phones, computer mice, even office furniture such as sit-stand desks), machine learning algorithms can amplify the impact of technology on health improvement by its ability to passively sense stress, and to provide just-in-time behavioral interventions based on contextual data and self-reported user feedback. At the same time, the ethical dimensions of these innovative lines of work – some of which entail fundamental concerns about privacy and autonomy – require careful attention from the scientific community. Most critically, there has been little engagement with the end-users of such technologies as a major stakeholder group who are most affected by these learning systems and tools. This administrative supplement request is premised on the fact that the rationale for and unmet needs targeted in the scope and aims of the parent grant can be even more effectively met (i.e. not changed but enriched) by adding participants with direct exposure and personal experience of interacting with precision health technologies to the last stakeholder group in the parent grant (i.e. patients). By extending the patient group in Aim 1 to include those directly participating in cutting-edge research at the intersection of occupational and precision health research, the Aims and Scope of the parent grant remain unchanged, while the real-world application and impact of the products from the parent grant are substantially enhanced. Our Supplemental proposal incorporates precision health technologies involving behavioral interventions of stress management that use ML into the first Specific Aims of the parent R01. In Supplemental Aim 1, we will use semi-structured interviews and qualitative methods to articulate ethical issues in the context of the development, refinement, and application of machine learning in behavioral interventions as part of a precision health methodology, with particular attention to occupational health contexts. Specifically, our methodology elicits a wide range of viewpoints from participants by comparing two distinct types of machine learning applications (i.e. physical versus digital interventions), with two varying degrees of autonomy that users may exercise to accept or reject the AI-recommended interventions. Both of these applications present novel ethical questions regarding the decision-making role of ML/AI algorithms in behavioral health research and practice. This supplementary project leverages access to the exceptional machine learning research conducted at Stanford University, including work by NIH-funded investigators, and provides extensive, systematically collected data on ethical issues encountered and anticipated as a result of machine learning applications in precision, behavioral, and occupational health.

项目摘要 人工智能应用程序实现更细粒度和更普遍测量的潜力， 预测，并提供行为干预提供了巨大的承诺，在达到目标的精度 健康，以维护人民的整体健康。当应用到我们日常生活中遇到的设备时， 环境（例如，个人计算机、移动的电话、计算机鼠标，甚至办公家具，例如坐立两用 机器学习算法可以通过其能力放大技术对健康改善的影响， 被动地感知压力，并根据上下文数据提供及时的行为干预， 自我报告的用户反馈。与此同时，这些创新工作的道德层面-一些 其中涉及对隐私和自主权的基本关注-需要科学界的密切关注， 社区最关键的是，很少与这些技术的最终用户接触， 受这些学习系统和工具影响最大的主要利益攸关方群体。这一行政 提出补充请求的理由是，范围和 通过增加以下内容，可以更有效地实现父母补助金的目标（即不改变，而是丰富） 参与者直接接触并亲身体验与精确健康技术的互动， 父母补助金中的最后一个利益相关者群体（即患者）。通过扩展目标1中的患者群体， 直接参与职业和精准健康交叉领域前沿研究的人员 研究，父母补助金的目的和范围保持不变，而现实世界的应用和 来自母公司赠款的产品的影响大大增强。我们的补充建议 结合了精确的健康技术，涉及压力管理的行为干预， ML到父R 01的第一个特定目标中。在补充目标1中，我们将使用半结构化面试 和定性的方法来阐明道德问题的背景下，发展，完善， 机器学习在行为干预中的应用，作为精确健康方法的一部分， 特别注意职业健康方面。具体来说，我们的方法学涵盖了广泛的 通过比较两种不同类型的机器学习应用程序（即物理 与数字干预相比），用户可以行使两种不同程度的自主权来接受或拒绝 AI建议的干预措施。这两个应用程序都提出了新的伦理问题， ML/AI算法在行为健康研究和实践中的决策作用。这项补充 该项目利用了斯坦福大学进行的卓越机器学习研究， 包括NIH资助的研究人员的工作，并提供广泛的，系统收集的数据， 由于机器学习在精度、行为和 职业健康