Human services organizations and the responsible integration of AI: Considering ethics and contextualizing risk(s)

Context matters more than blanket AI policies: The paper argues against a binary "adopt or reject AI" approach, showing instead that ethical risks vary significantly depending on the specific use case and implementation context within human services organizations—meaning some AI applications may be responsibly integrated while others require stricter limitations.

Multiple ethical dimensions require assessment: Beyond commonly discussed concerns like model bias, the authors identify several interconnected ethical issues including professional judgment displacement, environmental impact, and data labeler exploitation, suggesting organizations need multidimensional risk frameworks rather than single-issue evaluations.

Local large language models offer a practical solution: The paper highlights locally-run AI models as a promising approach that can address common ethical concerns—particularly around data privacy and control—while still allowing human services organizations to benefit from AI capabilities.

Start small and build evidence: Rather than large-scale AI rollouts, the authors recommend organizations begin with lower-risk applications and use careful experimentation to develop empirical understanding of how AI affects their specific contexts and client outcomes before expanding deployment.

Professional oversight and data sensitivity are key risk factors: The dimensional risk assessment approach emphasizes that applications requiring high professional judgment or involving sensitive client data present greater ethical challenges and therefore demand stronger safeguards and human oversight mechanisms.