AI creates and exacerbates privacy risks, yet practitioners lack effective resources to identify and mitigate these risks. We present Privy, a tool that guides practitioners without privacy expertise through structured privacy impact assessments to: (i) identify relevant risks in novel AI product concepts, and (ii) propose appropriate mitigations. Privy was shaped by a formative study with 11 practitioners, which informed two versions — one LLM-powered, the other template-based. We evaluated these two versions of Privy through a between-subjects, controlled study with 24 separate practitioners, whose assessments were reviewed by 13 independent privacy experts. Results show that Privy helps practitioners produce privacy assessments that experts deemed high quality: practitioners identified relevant risks and proposed appropriate mitigation strategies. These effects were augmented in the LLM-powered version. Practitioners themselves rated Privy as being useful and usable, and their feedback illustrates how it helps overcome long-standing awareness, motivation, and ability barriers in privacy work.

The rise of Generative AI (GenAI) in knowledge workflows raises questions about its impact on critical thinking skills and practices.We survey 319 knowledge workers to investigate 1) when and how they perceive the enaction of critical thinking when using GenAI, and 2) when and why GenAI affects their effort to do so.Participants shared 936 first-hand examples of using GenAI in work tasks.Quantitatively, when considering both task-and user-specific factors, a user's task-specific self-confidence and confidence in GenAI are predictive of whether critical thinking is enacted and the effort of doing so in GenAI-assisted tasks.Specifically, higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking.Qualitatively, GenAI shifts the nature of critical thinking toward information verification, response integration, and task stewardship.Our insights reveal new design challenges and opportunities for developing GenAI tools for knowledge work.

Social media websites thrive on user engagement by employing Attention Capture Damaging Patterns (ACDPs), e.g., infinite scroll, that prey on cognitive vulnerabilities to distract users. Prior work has taxonomized these ACDPs, but we have yet to measure how the presence of ACDPs impacts perceived distraction nor how mechanisms that suppress ACDPs reduce distraction. We conducted a two-week, mixed-methods field study with 29 participants to model how people get distracted when browsing social media websites, and how ACDPs might play a role. In the first week of the study, we sample participants’ in-situ perceptions of distraction, subjective perceptions of the browsing session (e.g., satisfaction), and the presence/absence of ACDPs. Participants reported feeling distracted 28% of the time, and that subjective perceptions and some ACDPs (e.g., notifications) highly correlated with when they felt distracted. In the second week of the study, participants were given access to Purpose Mode — a browser extension that allows users to “toggle off” ACDPs. Participants reported feeling distracted only 7% of the time and spent 21 fewer daily minutes browsing these websites. We discovered that Purpose Mode empowered users to feel more in control over their social media browsing and made participants feel less irritated and frustrated.

Privacy is a key principle for developing ethical AI technologies, but how does including AI technologies in products and services change privacy risks? We constructed a taxonomy of AI privacy risks by analyzing 321 documented AI privacy incidents. We codified how the unique capabilities and requirements of AI technologies described in those incidents generated new privacy risks, exacerbated known ones, or otherwise did not meaningfully alter the risk. We present 12 high-level privacy risks that AI technologies either newly created (e.g., exposure risks from deepfake pornography) or exacerbated (e.g., surveillance risks from collecting training data). One upshot of our work is that incorporating AI technologies into a product can alter the privacy risks it entails. Yet, current approaches to privacy-preserving AI/ML (e.g., federated learning, differential privacy, checklists) only address a subset of the privacy risks arising from the capabilities and data requirements of AI.

The widespread use of Large Language Model (LLM)-based conversational agents (CAs), especially in high-stakes domains, raises many privacy concerns. Building ethical LLM-based CAs that respect user privacy requires an in-depth understanding of the privacy risks that concern users the most. However, existing research, primarily model-centered, does not provide insight into users’ perspectives. To bridge this gap, we analyzed sensitive disclosures in real-world ChatGPT conversations and conducted semi-structured interviews with 19 LLM-based CA users. We found that users are constantly faced with trade-offs between privacy, utility, and convenience when using LLM-based CAs. However, users’ erroneous mental models and the dark patterns in system design limited their awareness and comprehension of the privacy risks. Additionally, the human-like interactions encouraged more sensitive disclosures, which complicated users’ ability to navigate the trade-offs. We discuss practical design guidelines and the needs for paradigm shifts to protect the privacy of LLM-based CA users.

Many people are concerned about how their personal data is used for online behavioral advertising (OBA). Ad targeting explanations have been proposed as a way to reduce this concern by improving transparency. However, it is unclear when and why people might want ad targeting explanations. Without this insight, we run the risk of designing explanations that do not address real concerns. To bridge this gap, we conducted a four-week, mixed-methods field study with 60 participants to understand when and why people want targeting explanations for the ads they actually encountered while browsing the web. We found that users wanted explanations for around 30% of the 4,251 ads we asked them about during the study, and that subjective perceptions of how their personal data was collected and shared were highly correlated with when users wanted ad explanations. Often, users wanted these explanations to confirm or deny their own preconceptions about how their data was collected or the motives of advertisers. A key upshot of our work is that one-size-fits-all approaches to ad explanations are likely to fail at addressing people’s lived concerns about ad targeting; instead, more personalized explanations are needed.

Using a mixed-methods approach over six weeks, we studied 30 smartphone users’ task choices, task execution and effort devoted to two commercial mobile crowdsourcing platforms in the wild. We focused on the influence of activity contexts, characterized by breakpoint situations and activity attributes. In line with their stated preferences, the participants were more likely to proactively perform mobile crowdsourcing tasks during transitions between activities than during an ongoing activity and during long breaks, respectively. Their task choices were influenced by various activity attributes, and more impacted by their current and preceding activities than their upcoming ones. Two of our three target outcomes, task execution and task choice, were also influenced by individuals’ stress and energy levels. Our qualitative data provide further insights into participants’ decisions about which crowdsourcing tasks to perform and when; and our results’ implications for the design of future mobile crowdsourcing task-prompting mechanisms are also discussed.

Smartphone users do not deal with notifications strictly in the order they are displayed, but sometimes read them from the middle, suggesting a mismatch between current systems’ display order and users’ needs. We therefore used mixed methods to investigate 34 smartphone users’ desired notification display order and related it with users’ self-reported order of attendance. Classifying using these two orders as dimensions, we obtained seven types of notifications, which helped us not only highlight the distinct attributes but understand the implied roles of these seven types of notifications, as well as the implied meaning of display orders. This is especially manifested in our identification of three main mismatches between the two orders. Qualitative findings reveal several meanings that participants attached to particular positions when arranging notifications. We offer design implications for notification systems, including calling for two-dimensional notification layout to support the multi-purpose roles of smartphone notifications we identified.

Researchers have long attempted to estimate instant-messaging (IM) users’ attentiveness, responsiveness, and interruptibility. Yet, IM users’ self-presentation of their receptivity, and their perceptions of automated adjustment/revelation of their receptivity status (e.g., Facebook Messenger’s green dot that deems a user to be “active”), remain under-explored. We therefore told our 43 participants that our IM app, IMStatus, was capable of automatically estimating and adjusting their receptivity status to responsive, attentive, or interruptible based on their smartphone activity. These statuses were also presented to their IM contacts in three different styles. Over a two-week period, the participants rarely chose the status interruptible, and when they did, it was usually to indicate low availability. Textual presentation was usually chosen to express statuses precisely, especially at high and low extremes of receptivity; while graphical and numeric presentations were preferred when self-perceived receptivity levels were more ambiguous. Conflicts between recipients’ and senders’ perspectives are also discussed.

AI-enabled smart-home agents that automate household routines are increasingly viable, but the design space of how and what such systems should communicate with their users remains underexplored. Through a user-enactment study, we identified various interpretations of and feelings toward such a system's confidence in its automated acts. That confidence and their own mental models influenced what and how the participants wanted the system to communicate, as well as how they would assess, diagnose, and subsequently improve it. Automated acts resulted from false predictions were not generally considered improper, provided that they were perceived as reasonable or potentially useful. The participants' improvement strategies were of four general types, all of which will be discussed. Factors affecting their preferred levels of involvement in automated acts and their interest in system confidence were also identified. We conclude by making practical design recommendations for the user-system communication design spaces of smart-home routine assistants.

This study examines the characteristics of mobile instant-messaging users' relationships with their social contacts and the effects of both relationship and interruption context on four measures of receptivity: Attentiveness, Responsiveness, Interruptibility, and Opportuneness. Overall, interruption context overshadows relationship characteristics as predictors of all four of these facets of receptivity; this overshadowing was most acute for Interruptibility and Opportuneness, but existed for all factors. In addition, while Mobile Maintenance Expectation and Activity Engagement were negatively correlated with all receptivity measures, each such measure had its own set of predictors, highlighting the conceptual differences among the measures. Finally, delving more deeply into potential relationship effects, we found that a single, simple closeness question was as effective at predicting receptivity as the 12-item Unidimensional Relationship Closeness Scale.

History of conversations through instant messaging (IM) contains abundant information about the communication patterns of the dyad, including conversation partners' mutual responsiveness to messages. We have, however, not seen many examinations of using such information in modeling mobile users' responsiveness in IM communication. In this paper, we present an in-the-wild study, in which we leverage participants' IM messaging logs to build models predicting their general responsiveness. Our models based on data from 33 IM user achieved an accuracy of up to 71% (AUROC). In particular, we show that 90-day IM-communication patterns, in general, outperformed their 14-day equivalent in our prediction models, indicating better coherence between long-term IM patterns with their general communication experience.