The false promise of risk assessments: epistemic reform and the limits of fairness

By Ben Green. 2020

97 citations recorded by the source corpus.

Risk assessments have proliferated in the United States criminal justice system. The theory of change motivating their adoption involves two key assumptions: first, that risk assessments will reduce human biases by making objective decisions, and second, that risk assessments will promote criminal justice reform. In this paper I interrogate both of these assumptions, concluding that risk assessments are an ill-advised tool for challenging the centrality and legitimacy of incarceration within the criminal justice system. First, risk assessments fail to provide objectivity, as their use creates numerous sites of discretion. Second, risk assessments provide no guarantee of reducing incarceration; instead, they risk legitimizing the criminal justice system's structural racism. I then consider, via an "epistemic reform," the path forward for criminal justice reform. I reinterpret recent results regarding the "impossibility of fairness" as not simply a tension between mathematical metrics but as evidence of a deeper tension between notions of equality. This expanded frame challenges the formalist, colorblind proceduralism at the heart of the criminal justice system and suggests a more structural approach to reform. Together, this analysis highlights how algorithmic fairness narrows the scope of judgments about justice and how "fair" algorithms can reinforce discrimination.

Subfields: Cognitive Neuroscience, Safety Research, Sociology and Political Science, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Safety Research

In Sociology and Political Science

In AI Ethics, Fairness & Society

Citations in the corpus

Re-imagining Algorithmic Fairness in India and Beyond (2021, background): “The community also often repurposes terminology from US legal domains, such as “disparate impact”, “disparate treatment”, and “equal opportunity”, or use them as points of triangulation, in order to compare technical properties of fairness through analogy with legal concepts [81, 82].”
Measurement and Fairness (2019, critical): “Indeed, by using arrests as a proxy for crimes committed, COMPAS can only exacerbate racist policing practices, rather than transcending them [7, 13, 23, 37, 39].”
It's COMPASlicated: The Messy Relationship between RAI Datasets and Algorithmic Fairness Benchmarks (2021, background): “The “impossibility of fairness” principle 237 [71, 72, 73, 74] gets at the fact that in an unequal society, equal treatment is guaranteed to result in 238 unequal outcomes [67], making the thoughtful selection of fairness definitions and their associated 239 values all the more critical.”
Harms from Increasingly Agentic Algorithmic Systems (2023, background): “…the FATE community has extensively studied the ways in which automated decision-making can disempower individuals, by impairing human decision-making [Green and Chen 2019; Green 2021] or subjecting individuals to oppressive institutions [Barabas et al. 2020, 2018; Green 2020; Zilka et al. 2022].”
Directional Bias Amplification (2021, methodology): “This setting has two important differences compared to computer vision detection tasks: 1) The training labels typically come from arrest records and suffer from problems like historical and selection bias (Suresh & Guttag, 2019; Olteanu et al., 2019; Green, 2020), and 2) the task is to predict futu”
Algorithmic Bias in Education (2021, background): “Work by Green and colleagues (Green, 2020; Green & Hu, 2018; Green & Viljoen, 2020), for example, suggests that the conflation of fairness defined as statistical measures of group parity with fairness in society creates its own obstacles to addressing social inequities.”