skip to main content

Title: Algorithmic Impact Assessments and Accountability: The Co-construction of Impacts
Algorithmic impact assessments (AIAs) are an emergent form of accountability for entities that build and deploy automated decision-support systems. These are modeled after impact assessments in other domains. Our study of the history of impact assessments shows that "impacts" are an evaluative construct that enable institutions to identify and ameliorate harms experienced because of a policy decision or system. Every domain has different expectations and norms about what constitutes impacts and harms, how potential harms are rendered as the impacts of a particular undertaking, who is responsible for conducting that assessment, and who has the authority to act on the impact assessment to demand changes to that undertaking. By examining proposals for AIAs in relation to other domains, we find that there is a distinct risk of constructing algorithmic impacts as organizationally understandable metrics that are nonetheless inappropriately distant from the harms experienced by people, and which fall short of building the relationships required for effective accountability. To address this challenge of algorithmic accountability, and as impact assessments become a commonplace process for evaluating harms, the FAccT community should A) understand impacts as objects constructed for evaluative purposes, B) attempt to construct impacts as close as possible to actual harms, more » and C) recognize that accountability governance requires the input of various types of expertise and affected communities. We conclude with lessons for assembling cross-expertise consensus for the co-construction of impacts and to build robust accountability relationships. « less
; ; ; ;
Award ID(s):
Publication Date:
Journal Name:
ACM Conference on Fairness, Accountability,and Transparency (FAccT ’21)
Sponsoring Org:
National Science Foundation
More Like this
  1. Algorithmic impact assessments (AIA) are increasingly being proposed as a mechanism for algorithmic accountability. These assessments are seen as potentially useful for anticipating, avoiding, and mitigating the negative consequences of algorithmic decision-making systems (ADS). At the same time, what an AIA would entail remains under-specified. While promising, AIAs raise as many questions as they answer. Choices about the methods, scope, and purpose of impact assessments structure the possible governance outcomes. Decisions about what type of effects count as an impact, when impacts are assessed, whose interests are considered, who is invited to participate, who conducts the assessment, the public availability of the assessment, and what the outputs of the assessment might be all shape the forms of accountability that AIA proponents seek to encourage. These considerations remain open, and will determine whether and how AIAs can function as a viable governance mechanism in the broader algorithmic accountability toolkit, especially with regard to furthering the public interest. Because AlAs are still an incipient governance strategy, approaching them as social constructions that do not require a single or universal approach offers a chance to produce interventions that emerge from careful deliberation.
  2. A quiet revolution is afoot in the field of law. Technical systems employing algorithms are shaping and displacing professional decision making, and they are disrupting and restructuring relationships between law firms, lawyers, and clients. Decision-support systems marketed to legal professionals to support e-discovery—generally referred to as “technology assisted review” (TAR)—increasingly rely on “predictive coding”: machine-learning techniques to classify and predict which of the voluminous electronic documents subject to litigation should be withheld or produced to the opposing side. These systems and the companies offering them are reshaping relationships between lawyers and clients, introducing new kinds of professionals into legal practice, altering the discovery process, and shaping how lawyers construct knowledge about their cases and professional obligations. In the midst of these shifting relationships—and the ways in which these systems are shaping the construction and presentation of knowledge—lawyers are grappling with their professional obligations, ethical duties, and what it means for the future of legal practice. Through in-depth, semi-structured interviews of experts in the e-discovery technology space—the technology company representatives who develop and sell such systems to law firms and the legal professionals who decide whether and how to use them in practice—we shed light on the organizational structures, professional rulesmore »and norms, and technical system properties that are shaping and being reshaped by predictive coding systems. Our findings show that AI-supported decision systems such as these are reconfiguring professional work practices. In particular, they highlight concerns about potential loss of professional agency and skill, limited understanding and thereby both over- and under reliance on decision-support systems, and confusion about responsibility and accountability as new kinds of technical professionals and technologies are brought into legal practice. The introduction of predictive coding systems and the new professional and organizational arrangements they are ushering into legal practice compound general concerns over the opacity of technical systems with specific concerns about encroachments on the construction of expert knowledge, liability frameworks, and the potential (mis)alignment of machine reasoning with professional logic and ethics. Based on our findings, we conclude that predictive coding tools—and likely other algorithmic systems lawyers use to construct knowledge and reason about legal practice— challenge the current model for evaluating whether and how tools are appropriate for legal practice. As tools become both more complex and more consequential, it is unreasonable to rely solely on legal professionals—judges, law firms, and lawyers—to determine which technologies are appropriate for use. The legal professionals we interviewed report relying on the evaluation and judgment of a range of new technical experts within law firms and, increasingly, third-party vendors and their technical experts. This system for choosing technical systems upon which lawyers rely to make professional decisions—e.g., whether documents are responsive, or whether the standard of proportionality has been met—is no longer sufficient. As the tools of medicine are reviewed by appropriate experts before they are put out for consideration and adoption by medical professionals, we argue that the legal profession must develop new processes for determining which algorithmic tools are fit to support lawyers’ decision making. Relatedly, because predictive coding systems are used to produce lawyers’ professional judgment, we argue they must be designed for contestability— providing greater transparency, interaction, and configurability around embedded choices to ensure decisions about how to embed core professional judgments, such as relevance and proportionality, remain salient and demand engagement from lawyers, not just their technical experts.« less
  3. Assessing the ecological and economic impacts of non-native species is crucial to providing managers and policymakers with the information necessary to respond effectively. Most non-native species have minimal impacts on the environment in which they are introduced, but a small fraction are highly deleterious. The definition of ‘damaging’ or ‘high-impact’ varies based on the factors determined to be valuable by an individual or group, but interpretations of whether non-native species meet particular definitions can be influenced by the interpreter’s bias or level of expertise, or lack of group consensus. Uncertainty or disagreement about an impact classification may delay or otherwise adversely affect policymaking on management strategies. One way to prevent these issues would be to have a detailed, nine-point impact scale that would leave little room for interpretation and then divide the scale into agreed upon categories, such as low, medium, and high impact. Following a previously conducted, exhaustive search regarding non-native, conifer-specialist insects, the authors independently read the same sources and scored the impact of 41 conifer-specialist insects to determine if any variation among assessors existed when using a detailed impact scale. Each of the authors, who were selected to participate in the working group associated with this studymore »because of their diverse backgrounds, also provided their level of expertise and uncertainty for each insect evaluated. We observed 85% congruence in impact rating among assessors, with 27% of the insects having perfect inter-rater agreement. Variance in assessment peaked in insects with a moderate impact level, perhaps due to ambiguous information or prior assessor perceptions of these specific insect species. The authors also participated in a joint fact-finding discussion of two insects with the most divergent impact scores to isolate potential sources of variation in assessor impact scores. We identified four themes that could be experienced by impact assessors: ambiguous information, discounted details, observed versus potential impact, and prior knowledge. To improve consistency in impact decision-making, we encourage groups to establish a detailed scale that would allow all observed and published impacts to fall under a particular score, provide clear, reproducible guidelines and training, and use consensus-building techniques when necessary.« less
  4. Reproductive coercion (RC) is a type of intimate partner violence (IPV) characterized by partner interference with contraception or reproductive decision-making. Despite sexual minority people’s vulnerability to other forms of IPV, limited research has examined reproductive coercion in this population. Research on behavioral health impacts of reproductive coercion is also lacking, especially for sexual minorities. This study addressed these gaps by examining the occurrence and behavioral health impact of reproductive coercion among emerging adults, including both heterosexual and sexual minority people. In 2020, survey data were collected with emerging adults (aged 18–24 years) at a university in the southeastern United States (N = 387). We conducted bivariate analyses to explore associates of reproductive coercion and logistic regression to identify risk factors. We then conducted linear regression to investigate relationships between reproductive coercion and behavioral health outcomes, in the full sample and in separate models based on sexual attraction. One in 15 sampled students (6.3%) had experienced reproductive coercion. Sexual attraction was a significant risk factor for reproductive coercion, with increased rates among plurisexual (i.e., attracted to people of multiple genders) compared to monosexual students, in both bivariate (χ2(2) = 7.57, p = .023) and regression analyses ( B = 1.25, p = .012). Reproductive coercion was associated with worse behavioral health outcomesmore »in bivariate and regression analyses (anxiety: B = 3.77, p = .001; depression: B = 3.26, p = .010; alcohol use: B = 2.32, p < .001). In separate linear regression models based on sexual attraction, RC was significantly associated with behavioral health indicators for participants who were attracted only to men and those attracted to people of multiple genders but not for participants who were attracted only to women. Findings indicated increased RC risk for plurisexual students compared to monosexuals and showed significant associations between RC and behavioral health outcomes, with differential effects based on sexual attraction. These novel findings support the need for continued research on RC, inclusive of nuanced conceptualizations of sexuality.

    « less
  5. The increasing impact of algorithmic decisions on people’s lives compels us to scrutinize their fairness and, in particular, the disparate impacts that ostensibly color-blind algorithms can have on different groups. Examples include credit decisioning, hiring, advertising, criminal justice, personalized medicine, and targeted policy making, where in some cases legislative or regulatory frameworks for fairness exist and define specific protected classes. In this paper we study a fundamental challenge to assessing disparate impacts in practice: protected class membership is often not observed in the data. This is particularly a problem in lending and healthcare. We consider the use of an auxiliary data set, such as the U.S. census, to construct models that predict the protected class from proxy variables, such as surname and geolocation. We show that even with such data, a variety of common disparity measures are generally unidentifiable, providing a new perspective on the documented biases of popular proxy-based methods. We provide exact characterizations of the tightest possible set of all possible true disparities that are consistent with the data (and possibly additional assumptions). We further provide optimization-based algorithms for computing and visualizing these sets and statistical tools to assess sampling uncertainty. Together, these enable reliable and robust assessmentsmore »of disparities—an important tool when disparity assessment can have far-reaching policy implications. We demonstrate this in two case studies with real data: mortgage lending and personalized medicine dosing. This paper was accepted by Hamid Nazerzadeh, Guest Editor for the Special Issue on Data-Driven Prescriptive Analytics.« less