The Datafication of Care in Public Homelessness Services

Research Background and Issues

• What problems or challenges did the authors identify?

  • Public homelessness systems in North America are shifting toward data-driven approaches, utilizing standardized and algorithmic tools to assess client needs and allocate resources. However, these systems face several challenges in implementation, including:
    • Insufficient data quality and representativeness, which limits the effectiveness of algorithmic models.
    • Clients' reluctance to share information and distrust of the system, further impacting data accuracy and reliability.
    • Inconsistent data practices among different service providers, potentially leading to service duplication and a deterioration of client experiences.
    • Inherent biases and limitations of algorithmic risk assessment tools (e.g., VI-SPDAT) that remain unaddressed.

• Why is this issue important?

  • Homelessness is a complex social challenge, and leveraging data and AI technologies to optimize resource allocation can improve service efficiency and client outcomes. However, flawed data practices and blind reliance on algorithmic tools may exacerbate inequities for vulnerable populations and negatively impact social policies.

• Research Motivation and Related Work

  • This study is inspired by the SIGCHI community's research on public-sector algorithms, extending critical analyses of high-stakes sociotechnical systems like homelessness assistance systems. While prior studies have highlighted the biases and societal harms of AI tools, they lack an in-depth understanding of frontline workers' data practices.

Solutions

• What methods or solutions did the authors propose?

  • The authors proposed a "systemic perspective" to study how frontline workers collect and practice data in homelessness support systems, framing it as a continuous, relationship-based process rather than a static task isolated from specific contexts.
  • They advocated for a shift from predictive risk assessment models to holistic, human-centered needs assessment approaches.

• What is innovative about this solution?

  • Unlike many homelessness systems that rely on static risk assessment tools like VI-SPDAT, the city studied in this research employs a more dynamic and relationship-oriented assessment process, emphasizing iterative evaluations based on client trust and ongoing interactions. This approach avoids the bias of categorizing clients solely based on "risk scores."
  • The study delves into how frontline workers use heuristic decision-making to navigate uncertainties and conflicts in data collection, including balancing client privacy with policy-driven data requirements.

• What are the implementation steps and key techniques used?

  • A three-month ethnographic research method was employed, involving observations and interviews at three key service points within the city's homelessness system: a call center, a street outreach team, and shelters.
  • The authors recorded and analyzed interviews with 31 employees and conducted 60 hours of field observations with 21 staff members, exploring the tensions between data practices and organizational goals.

Research Findings

• What specific findings were achieved?

  • The study identified three core "care-oriented" goals in data practices:
    1. Matching Goal: Collecting sufficient client information to allocate resources quickly.
    2. Client Privacy and Autonomy Goal: Respecting clients' decisions not to provide information.
    3. Equity Goal: Meeting policy requirements, particularly tracking service outcomes across different demographic groups and reducing systemic discrimination.
  • The findings highlighted that data inconsistencies (e.g., clients providing different answers to the same questions) stem from client distrust of the system, limitations in inter-organizational data sharing, and resource constraints.

• What are its advantages compared to existing solutions?

  • Compared to traditional risk assessment tools (e.g., fixed-model AI tools), this relationship-based data collection and evaluation approach better addresses the dynamic needs of clients in real-world contexts, avoiding the mechanistic flaws of prioritizing clients solely through single-score mechanisms.
  • Its systemic and iterative approach enables frontline workers to gradually build trust with clients, improving data completeness and authenticity.

• What were the experimental or evaluation results?

  • Data practices were influenced by diverse factors, including spatial (e.g., shelter settings), technological (e.g., paper-based vs. digital record-keeping tools), and personnel (e.g., differences in work experience) dimensions.
  • In client assessments, data was treated as dynamic and evolving rather than static "truth." This poses challenges for the design of future algorithmic tools, particularly in accounting for the temporality and evolution of data.

• Limitations and Future Directions

  • This study focused on a single Canadian city, and its data practices may not fully apply to homelessness service systems in other institutional or cultural contexts.
  • Future research should incorporate the perspectives of clients with lived experiences of homelessness and explore similar algorithmic challenges in other public service domains.

Conclusion and Implications

• This study underscores the importance of developing human-centered data practices, particularly in fields facing complex sociotechnical challenges like homelessness assistance. As AI tools are increasingly adopted, the authors advocate rethinking existing risk assessment methods, shifting from "deficit-based categorization" to frameworks centered on dynamic, context-sensitive assessments of human needs.
• Future research could further explore how to empower frontline workers with the ability to interpret and challenge AI model outputs, building more equitable and transparent sociotechnical systems that advocate for the rights and opportunities of vulnerable populations in public systems.