Perceptions of AI-Assisted Case Assessment Among Social Work Professionals: A Quantitative Survey on Trust, Efficiency, and Ethical Concerns
TL;DR
The findings offer an empirically grounded basis for designing responsible AI integration frameworks in social services and conditionally endorse Large Language Models as decision-support instruments, contingent on transparency, ethical governance, and structured professional training.
Abstract
The progressive digitalization of human services administration has generated urgent questions about professional trust, algorithmic governance, and the appropriate institutional boundaries of automated decision support. Social work constitutes a particularly consequential arena for examining how practitioners evaluate AI-assisted tools, given the high-stakes relational and ethical demands that characterize the profession. This study examined the predictors of professional trust in AI-assisted case assessment among social work professionals, with a specific focus on the roles of prior AI experience, perceived efficiency, perceived quality, and ethical and bias concerns. A cross-sectional quantitative survey was administered to N = 127 social work professionals using validated Likert-format scales. An independent-samples t-test revealed that professionals with high prior AI experience demonstrated substantially greater trust than those with limited experience, t(125) = 8.24, p < .001, Cohen’s d = 1.45. Multiple regression analysis identified perceived efficiency (β = .31, p < .001, 95% CI [.17, .45]), AI experience (β = .29, p < .001, 95% CI [.15, .43]), and perceived assessment quality (β = .27, p < .001, 95% CI [.14, .40]) as significant positive predictors of trust, while ethical concerns exerted a significant negative effect (β = −.18, p = .014, 95% CI [−.32, −.04]). Perceived bias risk did not reach significance in the multivariate model (β = −.12, p = .081). The overall model explained 58% of the variance in trust (R² = .58, F(5,121) = 33.44, p < .001). Social work professionals conditionally endorse Large Language Models as decision-support instruments, contingent on transparency, ethical governance, and structured professional training. The findings offer an empirically grounded basis for designing responsible AI integration frameworks in social services.