DDRIG: The Algorithmic Translation of Expertise: Credible Knowledge and Machine Learning in Medicine

在医疗诊断、药物开发和公共卫生等领域，利用人工智能 (AI) 和机器学习 (ML) 协助专家做出复杂的专业决策的情况正在顺利进行。该项目重点关注机器学习系统的一个特别有前景的应用：通过分析 CT 扫描和数字化病理切片等图像来支持医疗诊断。本研究将研究基于机器学习的医学图像分析系统的开发，追踪其生产、应用和监管。它将特别关注医学专家和政策制定者如何评估机器学习系统提出的诊断建议的可信度。该研究旨在促进机器学习工具的使用，以提高医疗保健的质量和可及性，并为政策制定者提供有关引入这些技术的信息。开发机器学习系统涉及将人类专业知识转化为新的算法形式。这项研究将调查这一过程中提出的有关诊断可信度的新问题。鉴于机器学习系统的内部运作复杂且在某种程度上难以理解，医学专家如何评估这些系统的可信度？机器学习系统的兴起将如何影响人类专家的可信度？当训练有素的专家（历来是复杂专业问题最可信的评判者）发现他们的判断受到人工智能系统的隐性挑战时，对专业知识的理解将会发生怎样的变化？为了探索这些问题，研究人员将在两家人工智能初创公司进行人种学研究，对工程师和临床医生进行半结构化访谈，并分析书面材料。通过分析在此背景下有关可信知识的谈判，该项目将提供有关如何评估人类和机器的可信度的见解。除了对理解机器学习系统可信度的直接影响外，该研究还旨在丰富专业社会学、医学社会学、数据研究和新兴技术治理方面的学术成果。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

The use of artificial intelligence (AI) and machine learning (ML) to assist experts in making sophisticated professional decisions is well under way in such areas as medical diagnosis, drug development, and public health. This project focuses on an especially promising application of ML systems: their use to support medical diagnoses by analyzing images, such as CT scans and digitized pathology slides. This research will study the development of ML-based medical image analysis systems, tracing their production, application, and regulation. It will pay special attention to how medical experts and policymakers assess the credibility of the diagnostic suggestions that ML systems make. The research aims to contribute to the use of ML tools to improve the quality and accessibility of healthcare and to inform policymaking about the introduction of these technologies. Developing an ML system involves translating human expertise into a new algorithmic form. This study will investigate novel questions raised by this process about the credibility of diagnosis. How can medical experts evaluate the credibility of ML systems, given that the internal workings of these systems are complex and, to some extent, inscrutable? How might the rise of ML systems affect the credibility of human experts? How will understanding of expertise change when well-trained experts, historically the most credible judges of complex professional questions, find their judgments implicitly challenged by AI systems? To explore these questions, the investigators will conduct ethnography at two AI startups, conduct semi-structured interviews with engineers and clinicians, and analyze written materials. By analyzing negotiations over credible knowledge in this context, the project will provide insights about how the credibility of the human and the machine are assessed. Beyond its immediate implications for understanding the credibility of ML systems, the study aims to enrich scholarship in the sociology of expertise, medical sociology, data studies, and the governance of emerging technologies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.