PhD Student in Information Science at University of Colorado Boulder
Roles for computing in social changeRediet Abebe, Solon Barocas, Jon Kleinberg, Karen Levy, Manish Raghavan, and David G. Robinson. 2020. Roles for computing in social change. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (FAT* '20). Association for Computing Machinery, New York, NY, USA, 252–260. DOI:https://doi.org/10.1145/3351095.3372871
The authors explore how computing might intervene in political and social reform. The authors reject technosolutionism, the idea that computing can solve social issues and injustice. Rather, their "goal is to cut a path between solutionist andcritical perspectives by describing potential roles through whichcomputing work can support, rather than supplant, other ways ofunderstanding and addressing social problems" (pg. 253). They scope this work to exploring how technologies can be one of a larger part of contributing to positive social change. They suggest four roles for computing in social change: (1) computing as diagnostic; (2) computing as formalizer; (3)computing as rebuttal; and (4) computing as synecdoche.
Computing as DiagnosticDefinition: "Computing can help us measure social problems and diagnose how they manifest in technical systems" (pg. 253).
Computing can be used to diagnose and characterize social problems, providing evidentiary support for interrogating social problems and also computing itself. Audit studies uncovering biases in machine learning models are an example of this. These do not always propose computational solutions to social problems, but as tools for confronting systems.
Computing is not a unique method in diagnosing problems, but can be used in tandem with or alongside other methods. They also posit skepticism, as offering system performance metrics through diagnosis may lead individuals and organizations to to optimize towards those metrics, often distorting the original meaning of that metric. This can lead that metric to become less meaningful/useful over time (Goodhart’s Law/Campbell's Law). Diagnostic methods may also be impeded by IP law that black-boxes systems or constraints posed by researcher insitutions or the law at large (e.g., The Computer Fraud and Abuse Act). Another caveat is that diagnosis is not treatment.
Computing as FormalizerDefinition: "Computing requires explicit specification of inputs and goals, and can shape how social problems are understood" (pg. 254).
The authors point out the pitfalls of generalizable statements, which on one hand allow for flexible interpretation that may be beneficial, but on the other also allow for flexible interpretation that may uphold bias (e.g., "most qualified applicant" hiding discrimination in hiring). Embedding computing in high-stakes decision-making scenarios is increasingly moving generalized standards to more formal, rule-based models. Computing is defined by discrete rules, so offloading something like "most qualified applicant" to a model is to rank applicants based on defined specific roles reflecting "a range ofconcrete judgments about how job performance should be defined,measured, and forecasted" (pg. 254). The authors argue that the required simplistic formalization of algorithms can help interrogate social framings. What is the process of decision-making and why was it chosen? What is the threshold for an appropriate outcome and why? The process of formalization is a human-based one and articularing those decisions opens up avenues for exploration, contestation, and transparency. "The process of formalization can bring analytic clarity to policy debates by forcing stakeholders to be more precise about their goals and objectives" (pg. 255).
However, not all values can be formalized or quantified, and attempting to do so raises moral and ethical questions. Formalization as constructive intervention may also draw away from underlying policy goals. Further, approaches have tended to focus on individuals rather than systems.
Computing as RebuttalDefinition: "Computing can clarify the limits of technical interventions, and of policies premised on them" (pg. 256).
The authors pose technical experts as well-suited for preventing pursuing inappropriate computational interventions, by illustrating the limitations of those interventions. They can protest claims about computing's capabilities and neutrality and illuminate that technology may not be the best solution for a specific social aim. Further, critiques of computing practice may showcase broader critiques of law and policy. For example, much of fairness in computing have adopted legal frameworks of discrimination, which are only one perspective on anti-discrimination. Recent work criticizing metrics of fairness employed in this way also showcase the limitations of discrimination law.
One risk of computing as rebuttal is that, by exposing the limitations of computing, others will then focus on improving those limitations, despite the tool itself not satsifactorily addressing a social issue. This is salient in security use cases, like CCTV and facial recognition. As the authors point out, "critics point out that a surveillance technol-ogy does not work, one wonders if they would be thrilled if it did" (pg. 257). Another risk is that policymakers might write off computational approaches entirely, even when they are beneficial.
Computing as SynecdocheDefinition: "Computing can foreground long-standing social problems in a new way" (pg. 257).
Computing can act as a mechanism for bringing attention to old problems: "Computing can offer us a tractable focus through which to notice anew, and bring renewed attention to, old problems" (pg. 257). Technologies may make certain social problems more salient than they once were, opening avenues for new conversation about tackling them. However, the authors say computing can also be a scapegoat, where outrage at computing's role "may give us a convenient target for outrage at systemic injustice, but in a way that does not build momentum toward change" (pg. 257). This also does not give computing a pass, given, as Langdon Winner pointed out, computing allies naturally with certain sociopolitical systems.