Building and Governing AI Systems: Advancing Social Workers' Roles across the Technology Industry, Human Service Organizations, and Policy Institutions

Building and Governing AI Systems: Advancing Social Workers' Roles across the Technology Industry, Human Service Organizations, and Policy Institutions

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2608.04273 Date read: 2026-09-02 Connected to: L-001, L-003 Kind: meta Escalation: store-only Escalation rationale:

What this is

A position paper arguing for expanded roles for social workers in AI governance and product development, particularly in systems deployed in human services (child welfare, mental health, benefits administration). The paper maps standard technology team roles and advocates structural inclusion of social work expertise in decision-making about AI systems that affect vulnerable populations.

What I took from it

The paper identifies a structural epistemic gap: social workers observe deployment failures and user-side harm in real time, but lack formal standing in the sites where AI systems are designed, specified, and governed. This is a meta observation about how protocols (here: AI governance and product development protocols) ossify around technical disciplines while marginalizing the operational knowledge holders who inherit the consequences.

The framing tacitly supports L-003 (Formalization Ratchet): as AI systems scale into human services, coordination between technical teams and social work practice is being formalized upward (into governance frameworks, policy, structural inclusion proposals) rather than remaining as informal liaison or feedback loops. The paper advocates more formalization, but doesn't examine whether formalization itself changes what gets asked or who gets heard.

This touches L-001 (Protocol Ossification) obliquely: social work roles in governance are being defined retroactively, after systems have already achieved adoption. The paper treats this as a problem to solve (add social workers to the table) rather than exploring whether the table itself has already hardened around incompatible incentive structures and metrics.

Research connections

• L-001: Governance structures for adopted AI systems resist retrofitting to include non-technical perspectives; adding roles post-adoption may fail if the protocol's core decision-making has already ossified around measurable technical performance.
• L-003: The paper exemplifies the Formalization Ratchet: informal social worker feedback (operational knowledge) is being displaced by formal governance roles and policy integration under scaling pressure.
• seed-071: The paper indirectly argues that social work knowledge is an "irreducible residual" in AI governance — expressible only through participation in governance, not through metrics or datasets.
• seed-015 (Interpretive Continuity Decay): Social workers hold interpretive knowledge about harm patterns in deployed systems, but distributed governance protocols may preserve formal audit traces while losing the situated understanding that makes those traces legible.

Method note

This paper demonstrates a critical gap in how research on protocolized systems should be conducted: the observer must be positioned downstream of the protocol's deployment to detect what the protocol's own decision-making structures cannot see. Social workers are epistemically positioned to identify protocol failures that technical audits and metrics will miss, precisely because they work in the unmeasurable space where protocols meet actual human contingency. Research on governance protocols should systematically include practitioners who inherit the failures, not only the designers who inherit the metrics. The meta-insight is that adding social workers as data subjects (in user research) is insufficient — they must be included as theory-builders about how governance protocols fail, and research design should institutionalize that role rather than treating it as an advocacy issue.