Social media platforms pose privacy risks when data is used in unexpected ways (e.g., for advertising or data sharing with partners). Using a custom browser extension and an online survey with 195 Facebook users, we investigated (1) whether participants’ expected values of their Facebook privacy settings were (mis)aligned with their actual settings; (2) demographic differences in privacy expectation-setting mismatches; and (3) participants' privacy concerns and trust towards Facebook.Our study presents a current and comprehensive analysis of Facebook users' privacy settings. We find that expectation-setting mismatches are prevalent: all participants had at least one mismatch; many had multiple, often expecting their settings to be more restrictive than they were. We also found that Facebook's default values are not aligned with people's expectations and/or actual settings, which suggests that those defaults are ineffective. Furthermore, mismatches differed along certain demographic variables.Participants' trust in Facebook decreased after they became aware of mismatches and their actual settings. Our empirical findings indicate that, despite increased public awareness, media scrutiny, and regulatory attention regarding privacy issues, there is still a substantial and concerning disconnect between how private people perceive their social media data to be and how exposed their data actually is, opening them up to both interpersonal and institutional privacy risks. We discuss design and public policy implications of our findings.

Describing Privacy Enhancing Technologies (PETs) to the general public is challenging but essential to convey the privacy protections they provide. Existing research has explored the explanation of differential privacy in health contexts. Our study adapts well-performing textual descriptions of local differential privacy from prior work to a new context and broadens the investigation to the descriptions of additional PETs. Specifically, we develop user-centric textual descriptions for popular PETs in ad tracking and analytics, including local differential privacy, federated learning with and without local differential privacy, and Google's Topics. We examine the applicability of previous findings to these expanded contexts, and evaluate the PET descriptions with quantitative and qualitative survey data (n=306). We find that adapting a process- and implications-focused approach to the ad tracking and analytics context achieved similar effects in facilitating user understanding compared to health contexts, and that our descriptions developed with this process+implications approach for the additional, understudied PETs help users understand PETs' processes. We also find that incorporating an implications statement into PET descriptions did not hurt user comprehension but also did not achieve a significant positive effect, which contrasts prior findings in health contexts. We note that the use of technical terms as well as the machine learning aspect of PETs, even without delving into specifics, led to confusion for some respondents. Based on our findings, we offer recommendations and insights for crafting effective user-centric descriptions of privacy-enhancing technologies.

Personalization improves user experience by tailoring interactions relevant to each user’s background and preferences. However, personalization requires information about users that platforms often collect without their awareness or their enthusiastic consent. Here, we study how the transparency of AI inferences on users’ personal data affects their privacy decisions and sentiments when sharing data for personalization. We conducted two experiments where participants (N=877) answered questions about themselves for personalized public arts recommendations. Participants indicated their consent to let the system use their inferred data and explicitly provided data after awareness of inferences. Our results show that participants chose restrictive consent decisions for sensitive and incorrect inferences about them and for their answers that led to such inferences. Our findings expand existing privacy discourse to inferences and inform future directions for shaping existing consent mechanisms in light of increasingly pervasive AI inferences.

As content moderation becomes a central aspect of all social media platforms and online communities, interest has grown in how to make moderation decisions contestable. On social media platforms where individual communities moderate their own activities, the responsibility to address user appeals falls on volunteers from within the community. While there is a growing body of work devoted to understanding and supporting the volunteer moderators' workload, little is known about their practice of handling user appeals. Through a collaborative and iterative design process with Reddit moderators, we found that moderators spend considerable effort in investigating user ban appeals and desired to directly engage with users and retain their agency over each decision. To fulfill their needs, we designed and built AppealMod, a system that induces friction in the appeals process by asking users to provide additional information before their appeals are reviewed by human moderators. In addition to giving moderators more information, we expected the friction in the appeal process would lead to a selection effect among users, with many insincere and toxic appeals being abandoned before getting any attention from human moderators. To evaluate our system, we conducted a randomized field experiment in a Reddit community of over 29 million users that lasted for four months. As a result of the selection effect, moderators viewed only 30% of initial appeals and less than 10% of the toxically worded appeals; yet they granted roughly the same number of appeals when compared with the control group. Overall, our system is effective at reducing moderator workload and minimizing their exposure to toxic content while honoring their preference for direct engagement and agency in appeals.

Tech companies that rely on ads for business argue that users have control over their data via ad privacy settings. However, these ad settings are often hidden. This work aims to inform the design of findable ad controls and study their impact on users’ behavior and sentiment. We iteratively designed ad control interfaces that varied in the setting’s (1) entry point (within ads, at the feed’s top) and (2) level of actionability, with high actionability directly surfacing links to specific advertisement settings, and low actionability pointing to general settings pages (which is reminiscent of companies’ current approach to ad controls). We built a Chrome extension that augments Facebook with our experimental ad control interfaces and conducted a between-subjects online experiment with 110 participants. Results showed that entry points within ads or at the feed’s top, and high actionability interfaces, both increased Facebook ad settings’ findability and discoverability, as well as participants’ perceived usability of them. High actionability also reduced users’ effort in finding ad settings. Participants perceived high and low actionability as equally usable, which shows it is possible to design more actionable ad controls without overwhelming users. We conclude by emphasizing the importance of regulation to provide specific and research-informed requirements to companies on how to design usable ad controls.

Past work has explored various ways for online platforms to leverage crowd wisdom for misinformation detection and moderation. Yet, platforms often relegate governance to their communities, and limited research has been done from the perspective of these communities and their moderators. How is misinformation currently moderated in online communities that are heavily self-governed? What role does the crowd play in this process, and how can this process be improved? In this study, we answer these questions through semi-structured interviews with Reddit moderators. We focus on a case study of COVID-19 misinformation. First, our analysis identifies a general moderation workflow model encompassing various processes participants use for handling COVID-19 misinformation. Further, we show that the moderation workflow revolves around three elements: content facticity, user intent, and perceived harm. Next, our interviews reveal that Reddit moderators rely on two types of crowd wisdom for misinformation detection. Almost all participants are heavily reliant on reports from crowds of ordinary users to identify potential misinformation. A second crowd--participants' own moderation teams and expert moderators of other communities--provide support when participants encounter difficult, ambiguous cases. Finally, we use design probes to better understand how different types of crowd signals---from ordinary users and moderators---readily available on Reddit can assist moderators with identifying misinformation. We observe that nearly half of all participants preferred these cues over labels from expert fact-checkers because these cues can help them discern user intent. Additionally, a quarter of the participants distrust professional fact-checkers, raising important concerns about misinformation moderation.

In conventional software development, user experience (UX) designers and engineers collaborate through separation of concerns (SoC): designers create human interface specifications, and engineers build to those specifications. However, we argue that Human-AI systems thwart SoC because human needs must shape the design of the AI interface, the underlying AI sub-components, and training data. How do designers and engineers currently collaborate on AI and UX design? To find out, we interviewed 21 industry professionals (UX researchers, AI engineers, data scientists, and managers) across 14 organizations about their collaborative work practices and associated challenges. We find that hidden information encapsulated by SoC challenges collaboration across design and engineering concerns. Practitioners describe inventing ad-hoc representations exposing low-level design and implementation details (which we characterize as leaky abstractions) to “puncture” SoC and share information across expertise boundaries. We identify how leaky abstractions are employed to collaborate at the AI-UX boundary and formalize a process of creating and using leaky abstractions.

Social media platforms aspire to create online experiences where users can participate safely and equitably. However, women around the world experience widespread online harassment, including insults, stalking, aggression, threats, and non-consensual sharing of sexual photos. This article describes women's perceptions of harm associated with online harassment and preferred platform responses to that harm. We conducted a survey in 14 geographic regions around the world (N = 3,993), focusing on regions whose perspectives have been insufficiently elevated in social media governance decisions (e.g. Mongolia, Cameroon). Results show that, on average, women perceive greater harm associated with online harassment than men, especially for non-consensual image sharing. Women also prefer most platform responses compared to men, especially removing content and banning users; however, women are less favorable towards payment as a response. Addressing global gender-based violence online requires understanding how women experience online harms and how they wish for it to be addressed. This is especially important given that the people who build and govern technology are not typically those who are most likely to experience online harms.

Affirmative consent is the idea that someone must ask for, and earn, enthusiastic approval before interacting with someone else. For decades, feminist activists and scholars have used affirmative consent to theorize and prevent sexual assault. In this paper, we ask: Can affirmative consent help to theorize online interaction? Drawing from feminist, legal, and HCI literature, we introduce the feminist theory of affirmative consent and use it to analyze social computing systems. We present affirmative consent’s five core concepts: it is voluntary, informed, revertible, specific, and unburdensome. Using these principles, this paper argues that affirmative consent is both an explanatory and generative theoretical framework. First, affirmative consent is a theoretical abstraction for explaining various problematic phenomena in social platforms—including mass online harassment, revenge porn, and problems with content feeds. Finally, we argue that affirmative consent is a generative theoretical foundation from which to imagine new design ideas for consentful socio-technical systems.

Social signals are crucial when we decide if we want to interact with someone online. However, social signals are typically limited to the few that platform designers provide, and most can be easily manipulated. In this paper, we propose a new idea called synthesized social signals (S3s): social signals computationally derived from an account's history, and then rendered into the profile. Unlike conventional social signals such as profile bios, S3s use computational summarization to reduce receiver costs and raise the cost of faking signals. To demonstrate and explore the concept, we built Sig, an extensible Chrome extension that computes and visualizes S3s. After a formative study, we conducted a field deployment of Sig on Twitter, targeting two well-known problems on social media: toxic accounts and misinformation. Results show that Sig reduced receiver costs, added important signals beyond conventionally available ones, and that a few users felt safer using Twitter as a result. We conclude by reflecting on the opportunities and challenges S3s provide for augmenting interaction on social platforms.

Resolving disputes in a timely manner is crucial for any online production group. We present an analysis of Requests for Comments (RfCs), one of the main vehicles on Wikipedia for formally resolving a policy or content dispute. We collected an exhaustive dataset of 7,316 RfCs on English Wikipedia over the course of 7 years and conducted a qualitative and quantitative analysis into what issues affect the RfC process. Our analysis was informed by 10 interviews with frequent RfC closers. We found that a major issue affecting the RfC process is the prevalence of RfCs that could have benefited from formal closure but that linger indefinitely without one, with factors including participants' interest and expertise impacting the likelihood of resolution. From these findings, we developed a model that predicts whether an RfC will go stale with 75.3% accuracy, a level that is approached as early as one week after dispute initiation.