In response to calls for open data and growing privacy threats, organizations are increasingly adopting privacy-preserving techniques that add noise to published datasets. These techniques seek to protect privacy of data subjects while enabling useful analyses. With expert feedback, we developed empirically-driven documentation explaining the noise characteristics of two Wikipedia pageview datasets: one using rounding (heuristic privacy) and another using differential privacy (DP, formal privacy). We then used these documents to conduct a task-based contextual inquiry (n=15) exploring how data users—largely unfamiliar with these methods—perceive, interact with, and interpret privacy-preserving noise during data analysis.

Decentralized social media protocols enable users in independent, user-hosted servers (i.e., instances) to interact with each other while they self-govern. This community-based model of social media governance opens up new opportunities for tailored decision-making about information flows - i.e., what user data is shared to whom and when - and in turn, for protecting user privacy. To better understand how community governance shapes privacy expectations on decentralized social media, we conducted a semi-structured interview with 23 users of the Fediverse, a decentralized social media network. Our findings illustrate important factors that shape a community's understandings of information flows, such as rules and proactive efforts from admins who are perceived as trustworthy. We highlight ''governance frictions'' between communities that raise new privacy risks due to incompatibilities in values, security practices, and software. Our findings point to the unique challenges of decentralized social media, suggest design opportunities to address frictions, and outline the role of participatory decision-making to realize the full potential of decentralization.

Differential privacy is a popular privacy-enhancing technology that has been deployed both by industry and government agencies. Unfortunately, existing explanations of differential privacy fail to set accurate privacy expectations for data subjects, which depend on the choice of deployment model. We design and evaluate new explanations of differential privacy for the local and central models, drawing inspiration from prior work explaining other privacy-enhancing technologies such as encryption. We reflect on the challenges in evaluating explanations and on the tradeoffs between qualitative and quantitative evaluation strategies. These reflections offer guidance for other researchers seeking to design and evaluate explanations of privacy-enhancing technologies.

Differential privacy (DP) has the potential to enable privacy-preserving analysis on sensitive data, but requires analysts to judiciously spend a limited "privacy loss budget" ϵ across queries. Analysts conducting exploratory analyses do not, however, know all queries in advance and seldom have DP expertise. Thus, they are limited in their ability to specify ϵ allotments across queries prior to an analysis. To support analysts in spending ϵ efficiently, we propose a new interactive analysis paradigm, Measure-Observe-Remeasure, where analysts "measure" the database with a limited amount of ϵ, observe estimates and their errors, and remeasure with more ϵ as needed.We instantiate the paradigm in an interactive visualization interface which allows analysts to spend increasing amounts of ϵ under a total budget. To observe how analysts interact with the Measure-Observe-Remeasure paradigm via the interface, we conduct a user study that compares the utility of ϵ allocations and findings from sensitive data participants make to the allocations and findings expected of a rational agent who faces the same decision task. We find that participants are able to use the workflow relatively successfully, including using budget allocation strategies that maximize over half of the available utility stemming from ϵ allocation. Their loss in performance relative to a rational agent appears to be driven more by their inability to access information and report it than to allocate ϵ.

Arguments that machine learning (ML) is facing a reproducibility and replication crisis suggest that some published claims in research cannot be taken at face value. Concerns inspire analogies to the replication crisis affecting the social and medical sciences. A deeper understanding of what reproducibility concerns in supervised ML research have in common with the replication crisis in experimental science puts the new concerns in perspective, and helps researchers avoid "the worst of both worlds," where ML researchers begin borrowing methodologies from explanatory modeling without understanding their limitations and vice versa. We contribute a comparative analysis of concerns about inductive learning that arise in causal attribution as exemplified in psychology versus predictive modeling as exemplified in ML. We identify common themes in reform discussions, like overreliance on asymptotic theory and non-credible beliefs about real-world data generating processes. We argue that in both fields, claims from learning are implied to generalize outside the specific environment studied (e.g., the input dataset or subject sample, modeling implementation, etc.) but are often difficult to refute due to underspecification of key parts of the learning pipeline. We conclude by discussing risks that arise when sources of errors are misdiagnosed and the need to acknowledge the role of human inductive biases in learning and reform.

The artificial intelligence research community is continuing to grapple with the ethics of its work by encouraging researchers to discuss potential positive and negative consequences. Neural Information Processing Systems (NeurIPS), a top-tier conference for machine learning and artificial intelligence research, first required a statement of broader impact in 2020. In 2021, NeurIPS updated their call for papers such that 1) the impact statement focused on negative societal impacts and was not required but encouraged, 2) a paper checklist and ethics guidelines were provided to authors, and 3) papers underwent ethics reviews and could be rejected on ethical grounds. In light of these changes, we contribute a qualitative analysis of 231 impact statements and all publicly-available ethics reviews. We describe themes arising around the ways in which authors express agency (or lack thereof) in identifying or mitigating negative consequences and assign responsibility for mitigating negative societal impacts. We also characterize ethics reviews in terms of the types of issues raised by ethics reviewers (falling into categories of policy-oriented and non-policy-oriented), recommendations ethics reviewers make to authors (e.g., in terms of adding or removing content), and interaction between authors, ethics reviewers, and original reviewers (e.g., consistency between issues flagged by original reviewers and those discussed by ethics reviewers). Finally, based on our analysis we make recommendations for how authors can be further supported in engaging with the ethical implications of their work.

Abstract Organizations often collect private data and release aggregate statistics for the public’s benefit. If no steps toward preserving privacy are taken, adversaries may use released statistics to deduce unauthorized information about the individuals described in the private dataset. Differentially private algorithms address this challenge by slightly perturbing underlying statistics with noise, thereby mathematically limiting the amount of information that may be deduced from each data release. Properly calibrating these algorithms—and in turn the disclosure risk for people described in the dataset—requires a data curator to choose a value for a privacy budget parameter,ɛ. However, there is little formal guidance for choosingɛ, a task that requires reasoning about the probabilistic privacy–utility tradeoff. Furthermore, choosingɛin the context of statistical inference requires reasoning about accuracy trade-offs in the presence of both measurement error and differential privacy (DP) noise. We presentVisualizingPrivacy (ViP), an interactive interface that visualizes relationships betweenɛ, accuracy, and disclosure risk to support setting and splittingɛamong queries. As a user adjustsɛ, ViP dynamically updates visualizations depicting expected accuracy and risk. ViP also has an inference setting, allowing a user to reason about the impact of DP noise on statistical inferences. Finally, we present results of a study where 16 research practitioners with little to no DP background completed a set of tasks related to settingɛusing both ViP and a control. We find that ViP helps participants more correctly answer questions related to judging the probability of where a DP-noised release is likely to fall and comparing between DP-noised and non-private confidence intervals.

The computer science research community and the broader public have become increasingly aware of negative consequences of algorithmic systems. In response, the top-tier Neural Information Processing Systems (NeurIPS) conference for machine learning and artificial intelligence research required that authors include a statement of broader impact to reflect on potential positive and negative consequences of their work. We present the results of a qualitative thematic analysis of a sample of statements written for the 2020 conference. The themes we identify broadly fall into categories related to how consequences are expressed (e.g., valence, specificity, uncertainty), areas of impacts expressed (e.g., bias, the environment, labor, privacy), and researchers' recommendations for mitigating negative consequences in the future. In light of our results, we offer perspectives on how the broader impact statement can be implemented in future iterations to better align with potential goals.