Hybrid Algorithmic Governance in U.S. Welfare Administration State-and County-Level AI as a Case of Support–Control Convergence

Jul 16, 2026

Upload by Maeve

The article analyzes institutional conditions shaping how artificial intelligence systems in U.S. welfare administration function either as instruments of support or as instruments of control. Rather than treating welfare algorithms as care or surveillance, it argues support–control coexistence inside one system, with their balance shifting over time through support–control convergence and an institutional ratchet. Routine political and budget pressures make control effects measurable, while reversing support requires disproportionate external intervention. Empirical process tracing across six cases shows drift toward control as routine and reversal as exceptional, costly, and incomplete.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

Introduction

Competing traditions in welfare algorithm interpretation

Support–control convergence and the institutional ratchet

Comparative motivation and research questions

Contribution and scope

arXiv :2607 .04503v 1 [ cs .CY] 5 Jul 2026 Hybrid Algorithmic Governance in U.S. Welfare Administration: State-and County-Level AI as a Case of Support–Control Convergence Maxim Dedyaev National Research University Higher School of Economics, Moscow [https://orcid.org/0009-0002-2675-5033](https://orcid.org/0009-0002-2675-5033) [E-mail: maxim.ai.policy@gmail.com](E-mail: maxim.ai.policy@gmail.com) Abstract This article examines the institutional conditions under which artificial intelligence systems in U.S. welfare administration come to operate as instruments of support or as instruments of control. Rather than asking what welfare algorithms “really” are (tools of proactive assistance or infrastructures of surveillance) the article starts from the premise that support and control are co-present within the same system, while their relative balance shifts over time. This movement is conceptualized through the notion of support–control convergence and the model of an institutional ratchet. Routine budgetary and political pressures make control-oriented effects easily measurable and politically capitalizable, whereas a return toward support requires external intervention of disproportionate force, such as judicial compulsion, legislative prohibition, or public scandal. Empirically, the article draws on process tracing of six state-and county-level cases: NYSDOL fraud detection, Michigan MiDAS, Illinois Medicaid managed care, LA County homelessness prevention, the Allegheny Family Screening Tool, and Washington Foster Care. The findings show that the system’s orientation is shaped by institutional design, with the decisive parameter being the side on which the costs of algorithmic error are placed. Drift toward control is routine, while reversal is exceptional and costly. In the MiDAS case, activation required a single administrative decision, whereas reversal took nine years and a $20 million settlement; even then, the system did not return to a support-oriented configuration. Keywords: algorithmic governance; welfare administration; artificial intelligence; support–control convergence; public administration; process tracing; false-positive costs; administrative discretion; U.S. federalism 1 Introduction The use of artificial intelligence in public administration is no longer merely a matter of accelerating discrete services. It now concerns the architecture of state action itself. Within the logic of the “third wave” of digital-era governance, algorithmic systems have become a central component of public administration, redistributing discretion, responsibility, and power among public officials, models, and private technology vendors (Dunleavy & Margetts, 2025; Wirtz, Weyerer & Geyer, 2019) . This shift bears directly on decisions about who gains access to state resources and who remains outside them. The effects of such systems therefore depend on how institutions reorganize decision-making procedures from within (Zuiderwijk, Chen & Salem, 2021) . This is most visible in welfare administration, where two almost non-overlapping traditions have emerged for interpreting the same technologies. The critical tradition shows that predictive systems reproduce and intensify inequality by operating as an apparatus of discipline and surveillance (Eubanks, 2018; Citron, 2008; Gilman, 2020; Alston, 2019) . The managerial tradition, rooted in the idea of the proactive state, sees the same technologies as tools for early risk detection, cost reduction, and expanded access to services (Margetts & Dorobantu, 2019; Scholta et al., 2019; Madan & Ashok, 2023) . In this literature, such polarization is often framed as a dispute over what welfare algorithms are: instruments of care or instruments of control. This article starts from a different premise. What diverges is not merely how these systems are interpreted, but how the systems themselves develop over time. The same class of predictive model, operating in the same s

FAQs

What does the article mean by support–control convergence?

Support and control are treated as co-present within the same welfare AI system, while their relative balance shifts over time due to institutional conditions.

Why does drift toward control happen more routinely than reversal?

Routine budgetary and political pressures make control-oriented effects easier to measure and capitalize politically, while returning to support requires unusually strong external intervention.

How are the six empirical cases used in the study?

The article uses process tracing of six state- and county-level cases to show how system orientation depends on institutional design, especially which side bears the costs of algorithmic error.

Jul 16, 2026

Upload by Maeve