Transitioning Towards a Proactive Practice: A Longitudinal Field Study on the Implementation of a ML System in Adult Social Care

Document Title

Transitioning Towards a Proactive Practice: A Longitudinal Field Study on the Implementation of a ML System in Adult Social Care

Document Information

• Subject Area: Artificial Intelligence and Social Welfare, Human-Computer Interaction in Social Services
• Keywords: Machine Learning, Implementation Challenges, Adult Social Care, Field Study, Human-Computer Interaction, Preventive Care, Data Sharing, Technology Integration, Social Work Technology, Public Services

Research Background and Issues

• Identified Problems or Challenges:

  • Adult social care (ASC) in England is facing a crisis, with increasing demand and limited resources.
  • Policymakers advocate for the use of machine learning (ML) systems to improve services, yet specific implementation guidelines are lacking.
  • Historically, the implementation of information systems has often failed due to human-computer interaction issues, while existing research predominantly focuses on the design and use of ML systems, with limited exploration of their real-world implementation.

• Significance:

  • Promoting a shift in the social care sector from a "reactive" model to a "proactive preventive" approach can alleviate system pressures and improve the well-being of service users.
  • Successful implementation of technology in social care can serve as a valuable reference for the digital transformation of public services.

• Research Motivation and Related Work:

  • Responding to calls within the human-computer interaction (HCI) field to explore how algorithm design can translate into real-world impact.
  • While some HCI studies have examined the impact of algorithmic decision-making at work boundaries, there is a lack of in-depth discussion on the "implementation" phase bridging laboratory research and practical application.

Solution

• Research Methodology:

  • Conducting a longitudinal field study to observe the process of implementing an ML system within a county council in England through a cross-organizational team.
  • The study aims to address three core questions:
    1. How are machine learning systems implemented in social care organizations?
    2. What are the main challenges encountered during implementation?
    3. How do practitioners respond to these challenges?

• Proposed Approach:

  • Integrating the ML system into the "preventive workflow" of social care.
  • Utilizing HCI knowledge to redesign workflows, reducing negative impacts and enhancing practical benefits.
  • Employing interactive interviews, participatory observation, and analysis of practical documentation to ensure comprehensive and reliable research data.

• Implementation Steps and Key Technologies:

  • The design of the ML system encompasses the full process from data collection and natural language processing (NLP) to generating predictions (e.g., fall risk).
  • Workflow design focuses on the interaction between the county council and service users, as well as the actual service delivery pathways.
  • Addressing specific challenges and introducing "pragmatic solutions," such as optimizing data-sharing protocols, implementing population screening strategies, and developing fuzzy evaluation tools (e.g., "dip checks").

Research Findings

• Specific Outcomes:

  • Described the complex process of implementing ML systems in the social care domain, uncovering a range of technical, social, and legal challenges.
  • Provided practice-oriented solutions, such as expanding data sources, optimizing AI models, and enhancing trust in data sharing.
  • Developed an alternative workflow based on HCI knowledge, emphasizing natural interaction with service users through risk prediction to improve project outcomes.

• Comparative Advantages Over Existing Solutions:

  • Systematically covered various challenges within cross-organizational collaboration networks, offering a detailed depiction of the social context of technology implementation.
  • Highlighted the potential for combining HCI and social work research to improve technology design from a human-centered perspective.

• Experimental or Evaluation Results:

  • Identified common implementation challenges in ML projects (e.g., barriers to data sharing, unclear definitions of target populations, capacity limitations in preventive services).
  • Proposed flexible and practical solutions based on these challenges and validated the effectiveness of HCI knowledge in improving workflows.

• Limitations and Future Directions:

  • The study is limited to a single scenario; future research could extend to more social care environments to verify the generalizability of conclusions.
  • Explore specific pathways for integrating technology with organizational workflows, particularly potential issues arising from system expansion to broader data sources.

Summary and Recommendations

• Recommend advancing ML projects from a problem-oriented perspective rather than a technology-driven approach.
• Enhance data collection mechanisms, including creating proxy variables that directly reflect phenomena.
• Facilitate early communication on data sharing and strengthen cross-organizational coordination.
• Combine ML system design with workflow design, incorporating user-driven design principles.
• Improve project evaluation frameworks to ensure robust and ethically sound assessment standards.

Through this study, the authors not only deepen academic understanding of ML system implementation in social care but also provide practical guidance for policymakers and practitioners, offering value beyond mere technical discussions.