Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency最新文献

{"title":"When not to design, build, or deploy","authors":"Solon Barocas, Asia J. Biega, Benjamin Fish, Jędrzej Niklas, Luke Stark","doi":"10.1145/3351095.3375691","DOIUrl":"https://doi.org/10.1145/3351095.3375691","url":null,"abstract":"Recent debate within the FAT* community has focused on how the field conceptualizes the problems it seeks to address, what approach the field should take in attempting to address these problems, and whether the field should even pursue some of the proposed remedies. Questions regarding when not to design, build, or deploy a technology are perhaps the most common expression of this trend. Identifying the problems to address is inextricably linked to the broader question of how to collectively make decisions about what technologies our societies need and want.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123702939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The impact of overbooking on a pre-trial risk assessment tool","authors":"K. Lum, Chesa Boudin, Megan Price","doi":"10.1145/3351095.3372846","DOIUrl":"https://doi.org/10.1145/3351095.3372846","url":null,"abstract":"Pre-trial risk assessment tools are used to make recommendations to judges about appropriate conditions of pre-trial supervision for people who have been arrested. Increasingly, there is concern about whether these models are operating fairly, including concerns about whether the models' input factors are fair measures of one's criminal activity. In this paper, we assess the impact of booking charges that do not result in a conviction on a popular risk assessment tool, the Arnold Public Safety Assessment. Using data from a pilot run of the tool in San Francisco, CA, we find that booking charges that do not result in a conviction (i.e. charges that are dropped or end in an acquittal) increased the recommended level of pre-trial supervision in around 27% of cases evaluated by the tool.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128563174","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"FlipTest","authors":"Emily Black, Samuel Yeom, Matt Fredrikson","doi":"10.1145/3351095.3372845","DOIUrl":"https://doi.org/10.1145/3351095.3372845","url":null,"abstract":"We present FlipTest, a black-box technique for uncovering discrimination in classifiers. FlipTest is motivated by the intuitive question: had an individual been of a different protected status, would the model have treated them differently? Rather than relying on causal information to answer this question, FlipTest leverages optimal transport to match individuals in different protected groups, creating similar pairs of in-distribution samples. We show how to use these instances to detect discrimination by constructing a flipset: the set of individuals whose classifier output changes post-translation, which corresponds to the set of people who may be harmed because of their group membership. To shed light on why the model treats a given subgroup differently, FlipTest produces a transparency report: a ranking of features that are most associated with the model's behavior on the flipset. Evaluating the approach on three case studies, we show that this provides a computationally inexpensive way to identify subgroups that may be harmed by model discrimination, including in cases where the model satisfies group fairness criteria.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120893721","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Implications of AI (un-)fairness in higher education admissions: the effects of perceived AI (un-)fairness on exit, voice and organizational reputation","authors":"Frank Marcinkowski, Kimon Kieslich, C. Starke, Marco Lünich","doi":"10.1145/3351095.3372867","DOIUrl":"https://doi.org/10.1145/3351095.3372867","url":null,"abstract":"Algorithmic decision-making (ADM) is becoming increasingly important in all areas of social life. In higher education, machine-learning systems have manifold uses because they can efficiently process large amounts of student data and use these data to arrive at effective decisions. Despite the potential upsides of ADM systems, fairness concerns are gaining momentum in academic and public discourses. The criticism largely focuses on the disparate effects of ADM. That is, algorithms may not serve as objective and fair decision-makers but, rather, reproduce biases existing within the respective training data. This study adopted a different approach by focusing on individual perceptions of fairness. Specifically, we looked at two different dimensions of perceived fairness: (i) procedural fairness and (ii) distributive fairness. Using cross-sectional survey data (n = 304) from a large German university, we tested whether students' assessments of fairness differ with respect to algorithmic vs. human decision-making (HDM) within the higher education context. Furthermore, we investigated whether fairness perceptions have subsequent effects on three different outcome variables, which are hugely important for universities: (1) exit, (2) voice, and (3) organizational reputation. The results of our survey suggest that participants evaluated ADM higher than HDM in terms of both procedural and distributive fairness. Concerning the subsequent effects of fairness perceptions, we find that (1) distributive fairness as well as procedural fairness perceptions have a negative impact on the intention to protest against an ADM system, whereas (2) only procedural fairness perceptions negatively affect the likelihood of exiting. Finally, (3) distributive fairness, but not procedural fairness perceptions have a positive effect on organizational reputation. For universities aiming to implement ADM systems, it is crucial, therefore, to take possible fairness issues and their further implications into account.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"1163 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121047177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Ethics on the ground: from principles to practice","authors":"Marguerite Barry, Aphra Kerr, Oliver Smith","doi":"10.1145/3351095.3375684","DOIUrl":"https://doi.org/10.1145/3351095.3375684","url":null,"abstract":"Surveys of public attitudes show that people believe it is possible to design ethical AI. However the everyday professional development context can offer minimal space for ethical reflection or oversight, creating a significant gap between public expectations and the performance of ethics in practice. This 2-part workshop includes an offsite visit to Telefónica Innovation Alpha and uses storytelling and theatre methods to examine how and where ethical reflection happens on the ground. It will explore the gaps in expectations and identify alternative approaches to more effective ethical performance. Bringing social scientists, data scientists, designers, civic rights activists and ethics consultants together to focus on AI/ML in the health context, it will foster critical and creative activities that will bring to the surface the structural, disciplinary, social and epistemological challenges to effective ethical performance in practice. Participants will explore and enact where, when and how meaningful interventions can happen.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"122 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131372413","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The meaning and measurement of bias: lessons from natural language processing","authors":"Abigail Z. Jacobs, Su Lin Blodgett, Solon Barocas, Hal Daumé, Hanna M. Wallach","doi":"10.1145/3351095.3375671","DOIUrl":"https://doi.org/10.1145/3351095.3375671","url":null,"abstract":"The recent interest in identifying and mitigating bias in computational systems has introduced a wide range of different---and occasionally incomparable---proposals for what constitutes bias in such systems. This tutorial introduces the language of measurement modeling from the quantitative social sciences as a framework for examining how social, organizational, and political values enter computational systems and unpacking the varied normative concerns operationalized in different techniques for measuring \"bias.\" We show that this framework helps to clarify the way unobservable theoretical constructs---such as \"creditworthiness,\" \"risk to society,\" or \"tweet toxicity\"---are turned into measurable quantities and how this process may introduce fairness-related harms. In particular, we demonstrate how to systematically assess the construct validity and reliability of these measurements to detect and characterize specific types of harms, which arise from mismatches between constructs and their operationalizations. We then take a critical look at existing approaches to examining \"bias\" in NLP models, ranging from work on embedding spaces to machine translation and hate speech detection. We show that measurement modeling can help uncover the implicit constructs that such work aims to capture when measuring \"bias.\" In so doing, we illustrate the limits of current \"debiasing\" techniques, which have obscured the specific harms whose measurements they implicitly aim to reduce. By introducing the language of measurement modeling, we provide the FAT* community with a framework for making explicit and testing assumptions about unobservable theoretical constructs embedded in computational systems, thereby clarifying and uniting our understandings of fairness-related harms.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"72 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121756462","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Model agnostic interpretability of rankers via intent modelling","authors":"Jaspreet Singh, Avishek Anand","doi":"10.1145/3351095.3375234","DOIUrl":"https://doi.org/10.1145/3351095.3375234","url":null,"abstract":"A key problem in information retrieval is understanding the latent intention of a user's under-specified query. Retrieval models that are able to correctly uncover the query intent often perform well on the document ranking task. In this paper we study the problem of interpretability for text based ranking models by trying to unearth the query intent as understood by complex retrieval models. We propose a model-agnostic approach that attempts to locally approximate a complex ranker by using a simple ranking model in the term space. Given a query and a blackbox ranking model, we propose an approach that systematically exploits preference pairs extracted from the target ranking and document perturbations to identify a set of intent terms and a simple term based ranker that can faithfully and accurately mimic the complex blackbox ranker in that locality. Our results indicate that we can indeed interpret more complex models with high fidelity. We also present a case study on how our approach can be used to interpret recently proposed neural rankers.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"64 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130803785","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Whose side are ethics codes on?: power, responsibility and the social good","authors":"A. Washington, Rachel Kuo","doi":"10.1145/3351095.3372844","DOIUrl":"https://doi.org/10.1145/3351095.3372844","url":null,"abstract":"The moral authority of ethics codes stems from an assumption that they serve a unified society, yet this ignores the political aspects of any shared resource. The sociologist Howard S. Becker challenged researchers to clarify their power and responsibility in the classic essay: Whose Side Are We On. Building on Becker's hierarchy of credibility, we report on a critical discourse analysis of data ethics codes and emerging conceptualizations of beneficence, or the \"social good\", of data technology. The analysis revealed that ethics codes from corporations and professional associations conflated consumers with society and were largely silent on agency. Interviews with community organizers about social change in the digital era supplement the analysis, surfacing the limits of technical solutions to concerns of marginalized communities. Given evidence that highlights the gulf between the documents and lived experiences, we argue that ethics codes that elevate consumers may simultaneously subordinate the needs of vulnerable populations. Understanding contested digital resources is central to the emerging field of public interest technology. We introduce the concept of digital differential vulnerability to explain disproportionate exposures to harm within data technology and suggest recommendations for future ethics codes..","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127606628","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Creating community-based tech policy: case studies, lessons learned, and what technologists and communities can do together","authors":"Hannah Sassaman, Jennifer Lee, Jenessa Irvine, Shankar Narayan","doi":"10.1145/3351095.3375689","DOIUrl":"https://doi.org/10.1145/3351095.3375689","url":null,"abstract":"What are the core ways the field of data science can center community voice and power throughout all the processes involved in conceptualizing, creating, and disseminating technology?? What are the most possible and most urgent ways communities can shape the field of algorithmic decision-making to center community power in the next few years? This interactive workshop will highlight some of the following lessons learned through our combined experience engaging with communities challenging technology in Seattle and Philadelphia, cities in the United States. We will discuss the historical context of disproportionate impacts of technology on marginalized and vulnerable communities; case studies including criminal justice risk assessments, face surveillance technologies, and surveillance regulations; and work in small-group and break-out sessions to engage questions about when and where technologists hold power, serve as gatekeepers, and can work in accountable partnership with impacted communities. By the end of the session, we hope that participants will learn how to actively center diverse communities in creating technology by examining successes, challenges, and ongoing work in Seattle and Philadelphia, through the following lessons we have learned: • that communities, policy-makers, and technologists need to work intimately together to lift up each other's' goals • that communities need to gain data justice and data literacy to understand and independently audit how a system is impacting them • that scientific analyses of algorithmic bias are powerful but heard most clearly when lifted up by local community members and stakeholders in decisions where algorithms might be deployed • that anecdotal stories of harm are most impactful on decisionmakers when tied to rigorous scientific analysis and examples from other communities that amplify and ground those stories • that communities and community goals and standards are often not heard in conversations between data scientists and people who deploy algorithms, as well as in decision-makers' conversations about what policy should look like • and that we need to begin to craft what it means for those with the least power in conversations about algorithmic fairness - those judged by those tools - to have far more, or even the most power in the future of their design or implementation.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121293918","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Algorithmic realism: expanding the boundaries of algorithmic thought","authors":"Ben Green, Salomé Viljöen","doi":"10.1145/3351095.3372840","DOIUrl":"https://doi.org/10.1145/3351095.3372840","url":null,"abstract":"Although computer scientists are eager to help address social problems, the field faces a growing awareness that many well-intentioned applications of algorithms in social contexts have led to significant harm. We argue that addressing this gap between the field's desire to do good and the harmful impacts of many of its interventions requires looking to the epistemic and methodological underpinnings of algorithms. We diagnose the dominant mode of algorithmic reasoning as \"algorithmic formalism\" and describe how formalist orientations lead to harmful algorithmic interventions. Addressing these harms requires pursuing a new mode of algorithmic thinking that is attentive to the internal limits of algorithms and to the social concerns that fall beyond the bounds of algorithmic formalism. To understand what a methodological evolution beyond formalism looks like and what it may achieve, we turn to the twentieth century evolution in American legal thought from legal formalism to legal realism. Drawing on the lessons of legal realism, we propose a new mode of algorithmic thinking---\"algorithmic realism\"---that provides tools for computer scientists to account for the realities of social life and of algorithmic impacts. These realist approaches, although not foolproof, will better equip computer scientists to reduce algorithmic harms and to reason well about doing good.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"178 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133513003","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}