TL;DR
This paper provides a comprehensive overview of how algorithms are used in the public sector, moving beyond the hype to examine the real-world implications. It argues that understanding these systems requires looking at their entire lifecycle, from problem formulation to evaluation, and highlights the critical social, political, and ethical challenges that arise, especially at the local government level.
What Problem Does This Paper Address?
While algorithms are increasingly discussed in the private sector, their use in public services often flies under the radar, despite their significant impact on citizens' lives. This paper addresses the gap in understanding the social implications of these public-sector algorithmic systems. It aims to provide a roadmap for researchers to investigate these systems, focusing on the complex interplay between technology, governance, and societal impact, particularly at the municipal level in the US. The authors highlight that defining "algorithm" itself is a challenge, and that these systems are not merely technical tools but sociotechnical systems embedded in social and institutional contexts.
Methods
The paper employs a "life-cycle approach" to analyze public-sector algorithmic systems. This framework breaks down the process into stages: problem formulation, data acquisition, procurement, governance, deployment, and evaluation. This approach is presented as an analytic tool rather than a strict chronological representation, allowing for a structured examination of the tensions, constraints, and dynamics at each stage. The authors synthesize existing empirical and analytical work from social science research, focusing on the political, legal, economic, and social dimensions of these systems, rather than solely their computational aspects.
Datasets
Not discussed in the paper. The paper focuses on a conceptual and empirical survey of existing literature and practices, rather than presenting new empirical research with specific datasets.
Key Insights and Ideas
- "Algorithmic Federalism": The adoption of algorithmic solutions at the local level is often shaped by policies and funding from higher levels of government, creating a dynamic where local governments may adopt systems not entirely of their own volition.
- "Procurement Mindset": The way governments acquire algorithmic systems, often through private vendors, can lead to a "hollowing out" of public power, where policy choices are embedded in contracts and left to third-party developers, potentially obscuring accountability.
- "Wicked Problems" and Algorithmic Simplification: Complex social issues ("wicked problems") are often "carved off" into more manageable, measurable components to fit algorithmic solutions, which can lead to superficial fixes that don't address root causes.
- Data as Both Resource and Constraint: Government data enables algorithmic systems but also shapes what problems can be addressed and how. Issues like selection bias and historical bias in data can lead to the reproduction or exacerbation of existing inequalities.
- Function Creep: Algorithmic systems can expand beyond their original intended uses without open debate, leading to unaccountable changes in how technology is deployed and data is used.
- Transparency is Multidimensional and Difficult: Achieving meaningful transparency for algorithmic systems is complex, involving various objects (code, data, decisions) and forms (proactive vs. reactive), and is often hindered by trade secrets and contractual limitations.
Key Results
This paper is a survey and conceptual framework, not an empirical study with quantitative results. Its "results" are the insights and arguments derived from synthesizing existing research. The key takeaway is that public-sector algorithms are deeply embedded in complex sociotechnical systems, and their impacts are shaped by a multitude of factors beyond the technology itself, including policy, procurement, governance, and human interaction.
Limitations and Future Work
The authors acknowledge that empirical work on the impact and efficacy of public-sector algorithmic systems remains "nascent." They highlight several open questions that require further empirical research, particularly concerning:
- The social implications of municipal algorithmic systems across various stages of their lifecycle.
- The effectiveness of different governance mechanisms (transparency, impact assessments, bans).
- The long-term impacts of various approaches to algorithm reform and opposition.
- The dynamics of how systems are evaluated, maintained, and put to new uses over time.
The paper implicitly calls for more interdisciplinary research that bridges technical understanding with social science perspectives to better understand and govern these systems.
Summary updated