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

{"title":"POTs","authors":"B. Kulynych, R. Overdorf, C. Troncoso, Seda F. Gürses","doi":"10.1145/3351095.3372853","DOIUrl":"https://doi.org/10.1145/3351095.3372853","url":null,"abstract":"Algorithmic fairness aims to address the economic, moral, social, and political impact that digital systems have on populations through solutions that can be applied by service providers. Fairness frameworks do so, in part, by mapping these problems to a narrow definition and assuming the service providers can be trusted to deploy countermeasures. Not surprisingly, these decisions limit fairness frameworks' ability to capture a variety of harms caused by systems. We characterize fairness limitations using concepts from requirements engineering and from social sciences. We show that the focus on algorithms' inputs and outputs misses harms that arise from systems interacting with the world; that the focus on bias and discrimination omits broader harms on populations and their environments; and that relying on service providers excludes scenarios where they are not cooperative or intentionally adversarial. We propose Protective Optimization Technologies (POTs). POTs, provide means for affected parties to address the negative impacts of systems in the environment, expanding avenues for political contestation. POTs intervene from outside the system, do not require service providers to cooperate, and can serve to correct, shift, or expose harms that systems impose on populations and their environments. We illustrate the potential and limitations of POTs in two case studies: countering road congestion caused by traffic beating applications, and recalibrating credit scoring for loan applicants.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"34 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":"123863481","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":"An empirical study on the perceived fairness of realistic, imperfect machine learning models","authors":"Galen Harrison, Julia Hanson, Christine Jacinto, Julio Ramirez, Blase Ur","doi":"10.1145/3351095.3372831","DOIUrl":"https://doi.org/10.1145/3351095.3372831","url":null,"abstract":"There are many competing definitions of what statistical properties make a machine learning model fair. Unfortunately, research has shown that some key properties are mutually exclusive. Realistic models are thus necessarily imperfect, choosing one side of a trade-off or the other. To gauge perceptions of the fairness of such realistic, imperfect models, we conducted a between-subjects experiment with 502 Mechanical Turk workers. Each participant compared two models for deciding whether to grant bail to criminal defendants. The first model equalized one potentially desirable model property, with the other property varying across racial groups. The second model did the opposite. We tested pairwise trade-offs between the following four properties: accuracy; false positive rate; outcomes; and the consideration of race. We also varied which racial group the model disadvantaged. We observed a preference among participants for equalizing the false positive rate between groups over equalizing accuracy. Nonetheless, no preferences were overwhelming, and both sides of each trade-off we tested were strongly preferred by a non-trivial fraction of participants. We observed nuanced distinctions between participants considering a model \"unbiased\" and considering it \"fair.\" Furthermore, even when a model within a trade-off pair was seen as fair and unbiased by a majority of participants, we did not observe consensus that a machine learning model was preferable to a human judge. Our results highlight challenges for building machine learning models that are perceived as fair and broadly acceptable in realistic situations.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"17 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":"116258999","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 concept of fairness in the GDPR: a linguistic and contextual interpretation","authors":"Gianclaudio Malgieri","doi":"10.1145/3351095.3372868","DOIUrl":"https://doi.org/10.1145/3351095.3372868","url":null,"abstract":"There is a growing attention on the notion of fairness in the GDPR in the European legal literature. However, the principle of fairness in the Data Protection framework is still ambiguous and uncertain, as computer science literature and interpretative guidelines reveal. This paper looks for a better understanding of the concept of fairness in the data protection field through two parallel methodological tools: linguistic comparison and contextual interpretation. In terms of linguistic comparison, the paper analyses all translations of the world \"fair\" in the GDPR in the EU official languages, as the CJEU suggests in CILFIT Case for the interpretation of the EU law. The analysis takes into account also the translation of the notion of fairness in other contiguous fields (e.g. at Article 8 of the EU Charter of fundamental rights or in the Consumer field, e.g. Unfair terms directive or Unfair commercial practice directive). In general, the notion of fairness is translated with several different nuances (in accordance or in discordance with the previous Data protection Directive and with Article 8 of the Charter) In some versions different words are used interchangeably (it is the case of French, Spanish and Portuguese texts), in other versions there seems to be a specific rationale for using different terms in different parts of the GDPR (it is the case of German and Greek version). The analysis reveals three mean semantic notions: correctness (Italian, Swedish, Romanian), loyalty (French, Spanish, Portuguese and the German version of \"Treu und Glaube\") and equitability (French, Spanish and Portuguese). Interestingly, these three notions have common roots in the Western legal history: the Roman law notion of \"bona fide\". Taking into account both the value of \"bona fide\" in the current European legal contexts and also a contextual interpretation of the role of fairness in the GDPR, the preliminary conclusions is that fairness refers to a substantial balancing of interests among data controllers and data subjects. The approach of fairness is effect-based: what is relevant is not the formal respect of procedures (in terms of transparency, lawfulness or accountability), but the substantial mitigation of unfair imbalances that create situations of \"vulnerability\". Building on these reflections, the paper analyses how the notion of fairness and imbalance are related to the idea of vulnerability, within and beyond the GDPR. In sum, the article suggests that the best interpretation of the fairness principles in the GDPR (taking into account both the notion of procedural fairness and of fair balancing) is the mitigation of data subjects' vulnerabilities through specific safeguards and measures.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"11 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":"132691295","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":"Multi-layered explanations from algorithmic impact assessments in the GDPR","authors":"M. Kaminski, Gianclaudio Malgieri","doi":"10.1145/3351095.3372875","DOIUrl":"https://doi.org/10.1145/3351095.3372875","url":null,"abstract":"Impact assessments have received particular attention on both sides of the Atlantic as a tool for implementing algorithmic accountability. The aim of this paper is to address how Data Protection Impact Assessments (DPIAs) (Art. 35) in the European Union (EU)'s General Data Protection Regulation (GDPR) link the GDPR's two approaches to algorithmic accountability---individual rights and systemic governance--- and potentially lead to more accountable and explainable algorithms. We argue that algorithmic explanation should not be understood as a static statement, but as a circular and multi-layered transparency process based on several layers (general information about an algorithm, group-based explanations, and legal justification of individual decisions taken). We argue that the impact assessment process plays a crucial role in connecting internal company heuristics and risk mitigation to outward-facing rights, and in forming the substance of several kinds of explanations.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"9 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":"114035931","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":"Toward situated interventions for algorithmic equity: lessons from the field","authors":"Michael A. Katell, Meg Young, Dharma Dailey, Bernease Herman, Vivian Guetler, Aaron Tam, Corinne Binz, Daniella Raz, P. Krafft","doi":"10.1145/3351095.3372874","DOIUrl":"https://doi.org/10.1145/3351095.3372874","url":null,"abstract":"Research to date aimed at the fairness, accountability, and transparency of algorithmic systems has largely focused on topics such as identifying failures of current systems and on technical interventions intended to reduce bias in computational processes. Researchers have given less attention to methods that account for the social and political contexts of specific, situated technical systems at their points of use. Co-developing algorithmic accountability interventions in communities supports outcomes that are more likely to address problems in their situated context and re-center power with those most disparately affected by the harms of algorithmic systems. In this paper we report on our experiences using participatory and co-design methods for algorithmic accountability in a project called the Algorithmic Equity Toolkit. The main insights we gleaned from our experiences were: (i) many meaningful interventions toward equitable algorithmic systems are non-technical; (ii) community organizations derive the most value from localized materials as opposed to what is \"scalable\" beyond a particular policy context; (iii) framing harms around algorithmic bias suggests that more accurate data is the solution, at the risk of missing deeper questions about whether some technologies should be used at all. More broadly, we found that community-based methods are important inroads to addressing algorithmic harms in their situated contexts.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"87 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":"116159445","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":"Robustness in machine learning explanations: does it matter?","authors":"Leif Hancox-Li","doi":"10.1145/3351095.3372836","DOIUrl":"https://doi.org/10.1145/3351095.3372836","url":null,"abstract":"The explainable AI literature contains multiple notions of what an explanation is and what desiderata explanations should satisfy. One implicit source of disagreement is how far the explanations should reflect real patterns in the data or the world. This disagreement underlies debates about other desiderata, such as how robust explanations are to slight perturbations in the input data. I argue that robustness is desirable to the extent that we're concerned about finding real patterns in the world. The import of real patterns differs according to the problem context. In some contexts, non-robust explanations can constitute a moral hazard. By being clear about the extent to which we care about capturing real patterns, we can also determine whether the Rashomon Effect is a boon or a bane.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"9 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":"115120543","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":"From the total survey error framework to an error framework for digital traces of humans: translation tutorial","authors":"Indira Sen, Fabian Flöck, Katrin Weller, Bernd Weiss, Claudia Wagner","doi":"10.1145/3351095.3375669","DOIUrl":"https://doi.org/10.1145/3351095.3375669","url":null,"abstract":"The digital traces of hundreds of millions of people offer increasingly comprehensive pictures of both individuals and groups on different platforms, but also allow inferences about broader target populations beyond those platforms. Studying the errors that can occur when digital traces are used to learn about humans and social phenomena is essential. Many similar errors also affect survey estimates, which survey designers have been addressing for decades, most notably using the Total Survey Error Framework (TSE). In this tutorial, we first introduce the audience to the concepts and guidelines of the TSE and how they are applied by survey practitioners in the social sciences. Second, we introduce our own conceptual framework to diagnose, understand, and avoid errors that may occur in studies that are based on digital traces of humans.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"25 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":"128359227","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":"Algorithmically encoded identities: reframing human classification","authors":"Dylan Baker, A. Hanna, Emily L. Denton","doi":"10.1145/3351095.3375687","DOIUrl":"https://doi.org/10.1145/3351095.3375687","url":null,"abstract":"Our aim with this workshop is to provide a venue within which the FAT* community can thoughtfully engage with identity and the categories which are imposed on people as part of making sense of their identities. Most people have nuanced and deeply personal understandings of what identity categories mean to them; however, sociotechnical systems must, through a set of classification decisions, reduce the nuance and complexity of those identities into discrete categories. The impact of misclassifications can range from the uncomfortable (e.g. displaying ads for items that aren't desirable) to devastating (e.g. being denied medical care; being evaluated as having a high risk of criminal recidivism). However, even the act of being classified can force an individual into categories which feel foreign and othering. Through this workshop, we hope to connect participants' personal understandings of identity to how identity is 'seen' and categorized by sociotechnical systems.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"29 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":"134222898","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 case for voter-centered audits of search engines during political elections","authors":"Eni Mustafaraj, Emma Lurie, Claire Devine","doi":"10.1145/3351095.3372835","DOIUrl":"https://doi.org/10.1145/3351095.3372835","url":null,"abstract":"Search engines, by ranking a few links ahead of million others based on opaque rules, open themselves up to criticism of bias. Previous research has focused on measuring political bias of search engine algorithms to detect possible search engine manipulation effects on voters or unbalanced ideological representation in search results. Insofar that these concerns are related to the principle of fairness, this notion of fairness can be seen as explicitly oriented toward election candidates or political processes and only implicitly oriented toward the public at large. Thus, we ask the following research question: how should an auditing framework that is explicitly centered on the principle of ensuring and maximizing fairness for the public (i.e., voters) operate? To answer this question, we qualitatively explore four datasets about elections and politics in the United States: 1) a survey of eligible U.S. voters about their information needs ahead of the 2018 U.S. elections, 2) a dataset of biased political phrases used in a large-scale Google audit ahead of the 2018 U.S. election, 3) Google's \"related searches\" phrases for two groups of political candidates in the 2018 U.S. election (one group is composed entirely of women), and 4) autocomplete suggestions and result pages for a set of searches on the day of a statewide election in the U.S. state of Virginia in 2019. We find that voters have much broader information needs than the search engine audit literature has accounted for in the past, and that relying on political science theories of voter modeling provides a good starting point for informing the design of voter-centered audits.","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"34 4 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":"115662891","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":"Measuring justice in machine learning","authors":"Alan Lundgard","doi":"10.1145/3351095.3372838","DOIUrl":"https://doi.org/10.1145/3351095.3372838","url":null,"abstract":"How can we build more just machine learning systems? To answer this question, we need to know both what justice is and how to tell whether one system is more or less just than another. That is, we need both a definition and a measure of justice. Theories of distributive justice hold that justice can be measured (in part) in terms of the fair distribution of benefits and burdens across people in society. Recently, the field known as fair machine learning has turned to John Rawls's theory of distributive justice for inspiration and operationalization. However, philosophers known as capability theorists have long argued that Rawls's theory uses the wrong measure of justice, thereby encoding biases against people with disabilities. If these theorists are right, is it possible to operationalize Rawls's theory in machine learning systems without also encoding its biases? In this paper, I draw on examples from fair machine learning to suggest that the answer to this question is no: the capability theorists' arguments against Rawls's theory carry over into machine learning systems. But capability theorists don't only argue that Rawls's theory uses the wrong measure, they also offer an alternative measure. Which measure of justice is right? And has fair machine learning been using the wrong one?","PeriodicalId":377829,"journal":{"name":"Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency","volume":"155 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":"123501708","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}