Speech emotion recognition (SER) is increasingly adopted in interactive systems, yet its deployment raises persistent concerns regarding privacy and user trust. Although differential privacy (DP) offers formal mathematical guarantees with quantifiable performance trade-offs, most existing SER implementations evaluate these trade-offs solely through technical metrics, leaving open the question of whether they align with users' perceptions and expectations. This disconnect between technical privacy assurances and user acceptance motivates the need for privacy design approaches that foreground user understanding, control, and trust. Guided by a research-through-design approach, we first conducted a formative study ($N = 16$) to examine users' privacy concerns and expectations surrounding SER technologies. Building on these insights, we designed a two-stage differentially private SER framework and instantiated it in an interactive online platform that integrates DP at both the model training and recognition stages with user-level privacy controls. Through an A/B study ($N = 32$), we investigated how varying privacy configurations influence perceived usefulness, trust, security, and overall acceptance. Our findings indicate that transparent and configurable privacy mediation can significantly enhance user trust and perceived security without materially compromising recognition performance. By conceptualizing privacy as a design material rather than a purely technical constraint, this work articulates user-centered principles for privacy mediation in emotion-aware AI systems and contributes to broader discussions on responsible interaction design, human-data relations, and the governance of affective technologies. RESEARCH HIGHLIGHTS A formative study with 16 participants identified key concerns about unclear data use, insufficient privacy guarantees, and privacy--performance trade-offs in speech emotion recognition.
As AI systems become increasingly prevalent, older adults engage with them through diverse approaches, ranging from active use to deliberate non-use. But how does trust in AI differ across these orientations? Our explanatory sequential mixed-methods study, including a survey (n = 217) and interviews (n = 20), reveals a paradox: older adults who actively use AI and those who intentionally choose not to report similarly moderate levels of trust, yet arrive at that trust through fundamentally different pathways. Among self-identified users, greater AI familiarity predicts higher trust, mediated by growing confidence through hands-on experience. Among non-users, the pattern reverses: the most AI-familiar individuals report the lowest trust, as knowledge without direct engagement fosters skepticism rather than confidence. Interview findings unpack the sense-making behind these patterns, revealing two distinct logics: practice-based trust, built through iterative use, error calibration, and continuous verification; and delegated trust, maintained by routing AI-related decisions through accountable human intermediaries while personally opting out. We discuss implications for inclusive AI design and literacy that support older adults' distinct pathways to trust, validating non-use as an informed decision rather than a deficit. RESEARCH HIGHLIGHTS Older AI users and non-users report similarly moderate trust, yet reach it through different sense-making logics.Familiarity raises trust among those who use AI but deepens skepticism among those who do not, driving trust in opposite directions.What distinguishes the two groups is whether they see AI errors as calibration cues or moral breaches.Users build resilient, practice-based trust through iterative use, error calibration, and continuous verification of AI outputs.Non-users sustain a delegated trust, granting AI conditional legitimacy only when an accountable person or institution vouches for its outputs.Self-reported use and non-use form a porous spectrum rather than a clean binary, complicating how we measure and design for AI engagement.
While many students with attention-deficit/hyperactivity disorder (ADHD) already use productivity and focus tools to navigate attention challenges in university life, these tools remain limited. Existing approaches typically rely on a static mapping of applications to activities (e.g., labeling Facebook as social media and Microsoft Word as writing), which are often rigid and outdated, failing to distinguish between study-related and nonstudy-related activities within the same application. Additionally, while some tools provide feedback, it is usually retrospective, limiting support for real-time awareness needed in the context of ADHD. These limitations point to the need for tools that (1) detect distractions in real time and (2) provide immediate, point-of-performance feedback. Toward this end, we developed a comprehensive approach that (1) analyzes content from user screenshots to determine task focus and (2) provides immediate feedback to promote awareness of off-task shifts, allowing for real-time adjustments without external supervision. This approach includes an off-task detection model trained on the training partition of 14,533 in-lab screenshots from a manually labeled dataset of 14,995 screenshots, which achieved 96% accuracy on in-lab data and 84.4% on the remaining 462 intentionally challenging real-world data. We evaluated this approach with 11 university students with ADHD that we asked to work on their homework with and without our tool for 2 hours each and conducted interviews afterward. Participants appreciated the real-time feedback and reported feeling more focused, with benefits continuing even after the tool was removed. Our findings suggest that this approach offers a flexible resource for students with ADHD to manage their attention during homework; however, its effectiveness as a supportive aid relies on ensuring that students retain full agency over when, and for how long, they choose to engage with the system.
Designing emotion-aware interactive systems requires sensitivity to the social and cultural contexts in which emotional experience is formed and expressed. Many existing approaches in human-computer interaction focus on technical emotion detection at the level of individual participants, while paying limited attention to how emotions emerge through everyday activities, social relationships, and material environments. This paper presents a methodological approach that embeds activity theory (AT) within a user-centred design (UCD) process to support the design of emotion-aware interactive systems. We report on a multi-stage design inquiry consisting of a literature review, a participatory workshop, iterative prototyping, and situated public engagement. AT constructs were used proactively to guide design activities and reflective analysis across all stages of the process. The findings show that emotional meaning is not a stable internal state but is continuously shaped through interactions among people, artifacts, norms, and social settings. Embedding AT within UCD enabled designers to attend to these relational and contextual dynamics throughout the design process. This paper presents and demonstrates a theory-informed design approach and contributes design knowledge for emotion-aware interactive systems by showing that such systems should move beyond technical detection alone to support socially situated and relational emotional expression. RESEARCH HIGHLIGHTS We introduce AT-UCD, a design approach that embeds Activity Theory within user-centered design to guide the development of contextually sensitive, emotion-aware interactive systems. Through a literature review, participatory workshop, iterative prototyping, and situated public engagement, we demonstrate how Activity Theory can proactively inform design decisions throughout the design process. Our findings reveal how emotional meaning is shaped through relationships, material artefacts, social norms, and everyday practices. This work offers design guidance for supporting socially situated emotional expression, shared interpretation, and context-sensitive interaction.
The Artificial Social Agent Questionnaire (ASAQ) evaluates human interaction experiences with Artificial Social Agents (ASAs), including virtual agents, chatbots, and robots. Allowing researchers to make claims about its results depends on the validity of the questionnaire. Concurrent validity refers to the degree to which a questionnaire agrees with other established instruments that measure related constructs. To examine the concurrent validity of ASAQ, we used eight established questionnaires, including the Godspeed Questionnaire Series, the Unified Theory of Acceptance and Use of Technology, and the Working Alliance Inventory. We selected these measures because of their conceptual overlap with ASAQ. A diverse sample of 666 participants evaluated ten agents they had recently interacted with using both the ASAQ and the established questionnaires. The results show moderate to high correlations for over 80% of ASAQ's 24 constructs and dimensions, providing evidence for its concurrent validity. Our findings can help researchers judge whether the ASAQ is an appropriate and valid instrument to evaluate their social agent.
