[论文解读] Sociotechnical Challenges of Machine Learning in Healthcare and Social Welfare

Tyler Reinmund, Lars Kunze+1 more|arXiv (Cornell University)|Jan 16, 2026

Artificial Intelligence in Healthcare and Education被引用 0

一句话总结

论文提出一个框架，识别在以机器学习为驱动的护理路径中共有11个社会技术性挑战，并给出一个过程模型，描述这些挑战如何通过设计、约束和实践而出现。它结合实地调研、纵向研究综述和共创工作坊，以实现对医疗护理场景中ML部署的精确分析。

ABSTRACT

Sociotechnical challenges of machine learning in healthcare and social welfare are mismatches between how a machine learning tool functions and the structure of care practices. While prior research has documented many such issues, existing accounts often attribute them either to designers' limited social understanding or to inherent technical constraints, offering limited support for systematic description and comparison across settings. In this paper, we present a framework for conceptualizing sociotechnical challenges of machine learning grounded in qualitative fieldwork, a review of longitudinal deployment studies, and co-design workshops with healthcare and social welfare practitioners. The framework comprises (1) a categorization of eleven sociotechnical challenges organized along an ML-enabled care pathway, and (2) a process-oriented account of the conditions through which these challenges emerge across design and use. By providing a parsimonious vocabulary and an explanatory lens focused on practice, this work supports more precise analysis of how machine learning tools function and malfunction within real-world care delivery.

研究动机与目标

• Identify and categorize sociotechnical challenges arising during the use of ML tools in healthcare and social welfare.
• Develop a parsimonious vocabulary to describe how ML tools function and malfunction in real-world care practice.
• Propose a process-oriented account of conditions that foster these challenges across design and use.

提出的方法

• Phase 1 grounds the framework in Technologies-in-Practice to connect design and use via facilities, interpretive schemes, and norms.
• Phase 2 conducts fieldwork in a fall-prevention ML deployment with 100+ hours of observation and 53 interviews.
• Phase 3 reviews 22 longitudinal ML deployment studies across care pathways.
• Phase 4 runs 5 co-design workshops with 15 practitioners to validate and extend the model.

实验结果

研究问题

• RQ1 What sociotechnical challenges emerge when ML tools are integrated into healthcare and social welfare care pathways?
• RQ2 How do design, constraints, and practice interact to produce these challenges?
• RQ3 What vocabulary best captures the phenomena to enable cross-setting analysis?

主要发现

• Identified 11 sociotechnical challenges aligned along stages of the ML-enabled care pathway.
• Challenges often overlap in practice despite clear analytical distinctions.
• A three-path process model explains how challenges emerge: determining design, grappling with constraints, and deviating from scripts.
• Fieldwork-informed framework aims to support precise analysis of ML deployment beyond mere technical failure.
• Workshops validate a dimension for distinguishing when challenges arise during ML-tool interaction.