Papers

Human Delegation Behavior in Human-AI Collaboration: The Effect of Contextual Information

The integration of artificial intelligence (AI) into human decision-making processes at the workplace presents both opportunities and challenges. One promising approach to leverage existing complementary capabilities is allowing humans to delegate individual instances of decision tasks to AI. However, enabling humans to delegate instances effectively requires them to assess several factors. One key factor is the analysis of both their own capabilities and those of the AI in the context of the given task. In this work, we conduct a behavioral study to explore the effects of providing contextual information to support this delegation decision. Specifically, we investigate how contextual information about the AI and the task domain influence humans' delegation decisions to an AI and their impact on the human-AI team performance. Our findings reveal that access to contextual information significantly improves human-AI team performance in delegation settings. Finally, we show that the delegation behavior changes with the different types of contextual information. Overall, this research advances the understanding of computer-supported, collaborative work and provides actionable insights for designing more effective collaborative systems.

Delusionized? Potential Harms of Proprioceptive Manipulations through Hand Redirection in Virtual Reality

To enhance interactions in VR, hand redirection (HR)-based illusion techniques apply offsets between the virtual and real-world position of users’ hands. While adaptation to such HR offsets is recognized, their impact on proprioception accuracy remains unexplored. However, deploying HR without understanding its potential effects on proprioception accuracy may pose risks to users in real-life situations. To investigate this, we conducted an experiment with 22 participants, studying the influence of prolonged exposure to unnoticeable HR offsets on proprioceptive accuracy during hand-reaching in VR. Our results show that proprioceptive accuracy declines significantly after prolonged exposure to redirected hand interactions. However, short-time exposure to unaltered hand interactions can – yet only partially – restore normal levels. Thus, we advocate being aware of potential risks arising from prolonged exposure to visual-proprioceptive offsets to ensure users’ safety.

Will Health Experts Adopt a Clinical Decision Support System for Game-Based Digital Biomarkers? Investigating the Impact of Different Explanations on Perceived Ease-of-Use, Perceived Usefulness, and Trust

This paper explores the adoption of a clinical decision support system (cDSS) utilizing game-based digital biomarkers for diagnosing mild cognitive impairment (MCI). Specifically, it investigates how different explanation methods, with a focus on data-centric explanations, impact perceived ease-of-use, perceived usefulness, and trust among healthcare professionals (HCPs). Through a qualitative study with 12 HCPs, we assess their interactions with an explainable AI (XAI)-enriched cDSS. The findings indicate that HCPs are open to adopting XAI-enriched cDSS to communicate the outcomes of game-based digital biomarkers. HCPs preferred to receive key diagnostic information in an easily digestible format. Both local explanations of intra-personal evolutionary data and global overview of normative data were found to be valuable for interpreting digital biomarkers. HCPs tended to trust the machine learning algorithms as a black box, but they considered the dataset used for training the model and the outcome prediction to be crucial. Therefore, presenting the uncertainty alongside the prediction was deemed important. These insights underscore the importance of designing cDSS tools that foster trust through clear, actionable explanations, paving the way for improved decision-making in clinical contexts.

"Is This Seat Accessible for Me?": An Autoethnography of a Person With a Mobility Disability Using Interactive Seat Plans for Public Events

Spectating sports matches or concerts is a popular activity, but these public live events have yet to become more accessible to people with disabilities. Inspecting the corresponding interactive seat plan before purchasing tickets online can be necessary to avoid or prepare for barriers at these venues. Unfortunately, these representations often lack valuable accessibility information. To explore how this can affect the disabled community, we leverage autoethnography to provide an in-depth introspective account through the lens of a person with a mobility disability. We apply Thematic Analysis to synthesise field notes from his research diary. The crafted themes showcase the lacking accessibility support in seat plans and illustrate the first author’s adaptation strategies to facilitate accessible experiences. We further contextualise his social relationships as a key factor throughout this process. Grounded in these results, we reflect on the provision of accessibility information, the categorisation of seats, and interdependent relationships within and through these systems.

Scrolling in the Deep: Analysing Contextual Influences on Intervention Effectiveness during Infinite Scrolling on Social Media

Infinite scrolling on social media platforms is designed to encourage prolonged engagement, leading users to spend more time than desired, which can provoke negative emotions. Interventions to mitigate infinite scrolling have shown initial success, yet users become desensitized due to the lack of contextual relevance. Understanding how contextual factors influence intervention effectiveness remains underexplored. We conducted a 7-day user study (N=72) investigating how these contextual factors affect users' reactance and responsiveness to interventions during infinite scrolling. Our study revealed an interplay, with contextual factors such as being at home, sleepiness, and valence playing significant roles in the intervention's effectiveness. Low valence coupled with being at home slows down the responsiveness to interventions, and sleepiness lowers reactance towards interventions, increasing user acceptance of the intervention. Overall, our work contributes to a deeper understanding of user responses toward interventions and paves the way for developing more effective interventions during infinite scrolling.

Exploring Flow in Real-World Knowledge Work Using Discrete cEEGrid Sensors

Flow, a state of deep task engagement, is associated with optimal experience and well-being, making its detection a prolific HCI research focus. While physiological sensors show promise for flow detection, most studies are lab-based. Furthermore, brain sensing during natural work remains unexplored due to the intrusive nature of traditional EEG setups. This study addresses this gap by using wearable, around-the-ear EEG sensors to observe flow during natural knowledge work, measuring EEG throughout an entire day. In a semi-controlled field experiment, participants engaged in academic writing or programming, with their natural flow experiences compared to those from a classic lab paradigm. Our results show that natural work tasks elicit more intense flow than artificial tasks, albeit with smaller experience contrasts. EEG results show a well-known quadratic relationship between theta power and flow across tasks, and a novel quadratic relationship between beta asymmetry and flow during complex, real-world tasks.

It's a Match - Enhancing the Fit between Users and Phishing Training through Personalisation

Effective training is essential for enhancing users' ability to detect phishing attempts. Personalised training offers huge potential to more closely align training content with individuals' needs and skill levels. In an online study, we assigned N=342 participants to personalised training or a random training variant to compare their effectiveness. The personalisation was based on a phishing proficiency score calculated from factors such as detection ability, knowledge, and security attitude. After training, the participants demonstrated greater proficiency, with an increased ability to detect phishing emails and higher security attitudes. These effects were most pronounced in the personalised condition, demonstrating the potential of personalisation to improve training outcomes. Overall, personalised training levelled the playing field, efficiently bringing all groups, regardless of their initial proficiency, to a comparable and desired post-training phishing proficiency level. Finally, we derived recommendations for designing personalised phishing training content and assigning users to suitable training programmes.

Work Hard, Play Harder: Intense Games Enable Recovery from High Mental Workload Tasks

Playing games has been shown to be an effective method of post-work recovery. Previous research has shown that gameplay with high cognitive involvement is effective for recovery. This finding conflicts with models of mental workload (MWL), which suggest that people feel best when cycling between high and low MWL. To unpack the relationship between recovery and mental workload, we designed a lab experiment where 40 participants experienced different combinations of high and low MWL while undertaking both work tasks and recovery gameplay, and we collected both self-report and physiological (fNIRS) data. Results showed that high and low MWL games created different impacts on recovery, depending on the MWL of the prior work task. While fNIRS measurements of MWL varied as expected during work tasks, experience of MWL when playing games was not evident in the prefrontal cortex. We conclude by discussing the relationship between mental workload and theories of recovery.

Closing the Loop between User Stories and GUI Prototypes: An LLM-Based Assistant for Cross-Functional Integration in Software Development

Graphical user interfaces (GUIs) are at the heart of almost every software we encounter. GUIs are often created through a collaborative effort involving UX designers, product owners, and software developers, constantly facing changing requirements. Historically, problems in GUI development include a fragmented, poorly integrated tool landscape and high synchronization efforts between stakeholders. Recent approaches suggest using large language models (LLMs) to recognize requirements fulfillment in GUIs and automatically propose new GUI components. Based on ten interviews with practitioners, this paper proposes an LLM-based assistant as a Figma plug-in that bridges the gap between user stories and GUI prototyping. We evaluated the prototype with 40 users and 40 crowd-workers, showing that the effectiveness of GUI creation is improved by using LLMs to detect requirements' completion and generate new GUI components. We derive design rationales to support cross-functional integration in software development, ensuring that our plug-in integrates well into established processes.

“Mirror, mirror in the call“: Exploring the Ambivalent Nature of the Self-view in Video Meeting Systems with Self-reported and Eye-tracking Data

Video meeting systems offer great potential for work and life, but they can also have negative effects. One reason is the presence of technical stimuli that do not exist in the physical world. A prominent example is the self-view feature, a mirrored image of oneself shown during the video meeting. The self-view feature comes with a trade-off between the advantage of enhancing control and the disadvantage of increasing cognitive load of its users. So far, research is scarce when it comes to understanding this ambivalent nature and studies mostly relied on self-reported data without considering the actual interaction with the self-view. To address this gap, we conducted an experimental study with 57 participants and two design variants (with/without self-view), analyzed user perceptions through surveys and interviews, and explored gaze patterns using eye-tracking technology. Results reveal varying perceptions of cognitive load and control among self-view users and between the design variants, highlighting its ambivalent nature. We see differences in how participants interact with the self-view. In a cluster analysis, we identify three user groups (Benefiting Users, Cognitively Challenged Users, Control Losing Users). These groups also show differences in visual behavior, especially median fixation duration, and user characteristics. Based on our findings, we outline design recommendations for more flexible and intelligent design solutions by considering user groups and their individual differences.

The Impact of Imperfect XAI on Human-AI Decision-Making

Explainability techniques are rapidly being developed to improve human-AI decision-making across various cooperative work settings. Consequently, previous research has evaluated how decision-makers collaborate with imperfect AI by investigating appropriate reliance and task performance with the aim of designing more human-centered computer-supported collaborative tools. Several human-centered explainable AI (XAI) techniques have been proposed in hopes of improving decision-makers' collaboration with AI; however, these techniques are grounded in findings from previous studies that primarily focus on the impact of incorrect AI advice. Few studies acknowledge the possibility for the explanations to be incorrect even if the AI advice is correct. Thus, it is crucial to understand how imperfect XAI affects human-AI decision-making. In this work, we contribute a robust, mixed-methods user study with 136 participants to evaluate how incorrect explanations influence humans' decision-making behavior in a bird species identification task taking into account their level of expertise and an explanation's level of assertiveness. Our findings reveal the influence of imperfect XAI and humans' level of expertise on their reliance on AI and human-AI team performance. We also discuss how explanations can deceive decision-makers during human-AI collaboration. Hence, we shed light on the impacts of imperfect XAI in the field of computer-supported cooperative work and provide guidelines for designers of human-AI collaboration systems.

AutoAugHAR: Automated Data Augmentation for Sensor-based Human Activity Recognition

Zhou 等人提出 AutoAugHAR 自动化数据增强框架，自动搜索最优增强策略提升基于传感器的人类活动识别性能。

Effects of Uncertain Trajectory Prediction Visualization in Highly Automated Vehicles on Trust, Situation Awareness, and Cognitive Load

Colley 等人研究高度自动驾驶汽车中不确定轨迹预测可视化形式对驾驶员信任度、态势感知和认知负荷的影响机制。

Eco Is Just Marketing: Unraveling Everyday Barriers to the Adoption of Energy-Saving Features in Major Home Appliances

Systematic Comparison of Ear Temperature Probing Positions for Continuous Wearable Vital Sign Monitoring

In recent years, research on sensory ear-worn devices ("earables") explored measuring body temperature at different locations in and around the ears. While the tympanic membrane (eardrum) is a well-established medical gold-standard position, earables face challenges integrating body temperature sensors because of the limited space available, the diversity of earphone designs, and the obstruction of the eardrum by the audio hardware components. Therefore, to understand the trade-offs in accuracy between different sensor positions in and around the ears, we contribute the first systematic comparison of locations around the ear. Based on related work, existing earable form factors, and a pre-study with thermal ear images from four subjects, we selected five positions to compare to tympanic temperature: concha, ear canal, and three positions spread behind the ear. Subsequently, we developed a custom earable with six optical temperature sensors that can measure all positions simultaneously. In a study with 12 participants at room temperature and settled isothermal conditions, we find that compared to the tympanic membrane, the mean temperature difference was 0.30°C colder at the concha (sufficient according to the American Society for Testing and Materials) and ear canal, and 0.6°C colder at positions behind the ear. Exposing participants to varying environmental conditions and physical movements resulted in unreliable measurements which could not be calibrated.

MLP-HAR: Boosting Performance and Efficiency of HAR Models on Edge Devices with Purely Fully Connected Layers

Neural network models have demonstrated exceptional performance in wearable human activity recognition (HAR) tasks. However, the increasing size or complexity of HAR models significantly impacts their deployment on wearable devices with limited computational power. In this study, we introduce a novel HAR model architecture named Multi-Layer Perceptron-HAR (MLP-HAR), which contains solely fully connected layers. This model is specifically designed to address the unique characteristics of HAR tasks, such as multi-modality interaction and global temporal information. The MLP-HAR model employs fully connected layers that alternately operate along the modality and temporal dimensions, enabling multiple fusions of information across these dimensions. Our proposed model demonstrates comparable performance with other state-of-the-art HAR models on six open-source datasets, while utilizing significantly fewer learnable parameters and exhibiting lower model complexity. Specifically, the complexity of our model is at least ten times smaller than that of the TinyHAR model and several hundred times smaller than the benchmark model DeepConvLSTM. Additionally, due to its purely fully connected layer-based architecture, MLP-HAR offers the advantage of ease of deployment. To substantiate these claims, we report the inference time performance of MLP-HAR on the Samsung Galaxy Watch 5 PRO and the Arduino Portenta H7 LITE, comparing it against other state-of-the-art HAR models.

SafeARCross: Augmented Reality Collision Warnings and Virtual Traffic Lights for Pedestrian Safety

Augmented Reality (AR) holds great potential for enhancing pedestrian’s urban experiences; however, its use in road traffic poses safety concerns due to potential distractions from interacting with AR interfaces. This paper investigates the effectiveness of AR applications for assisting pedestrians in crossing scenarios, against traditional crossing methods, by incorporating a collision warning system that uses an arrow to indicate the direction of a potential danger, and a virtual traffic light showing whether it is safe to cross. By leveraging Vehicle-to-Everything (V2X) communications within the living lab of Aveiro, Portugal, we conducted a user study to evaluate involved workloads, perceived safety and system usability in a realistic scenario. The findings from our study involving 20 participants reveal significant improvements in pedestrians’ perceived safety and a decrease in the perceived workload when using AR for pedestrian crossings, with both collision warning systems and virtual traffic lights demonstrating excellent usability.

Language Cues for Expressing Artificial Personality: A Systematic Literature Review for Conversational Agents

Users attribute artificial personality (AP) to conversational agents (CAs) based on perceived language respectively verbal cues. This review synthesizes studies on this topic, encompassing research not only on chat- and voicebots but also on social robots, drawing from interdisciplinary databases. This approach led to an identification of 200 verbal signals, nearly four times more as in previous reviews. The signals were classified according to the personality dimensions of the BFM as well as its facets. Besides, the relevance of theories of personality other than the BFM are discussed. Furthermore, six methodological challenges in the empirical study of verbal cues expressing AP are identified. Practical implications include providing practitioners an overview of verbal signals, while offering opportunities for research improvement based on identified challenges. Enhanced understanding of verbal signals related to AP aids in evaluating implementation quality, not only in rule-based CAs but also in LLM-based systems.

A Living Framework for Understanding Cooperative Games

Playing cooperative games is recognised as a positive social activity. Yet, we have limited means to rigorously define or communicate the structures that govern these experiences, hindering attempts at consolidating knowledge and limiting the potential of design efforts. In this work, we introduce the Living Framework for Cooperative Games (LFCG), a framework derived from a multi-step systematic analysis of 129 cooperative games with contributions of eleven researchers. We describe how LFCG can be used as a tool for analyses and ideation, and as a shared language for describing a game’s design. LFCG is published as a web application to facilitate use and appropriation. It supports the creation, dissemination and aggregation of game reports and specifications; and enables stakeholders to extend and publish custom versions. Lastly, we discuss using a research-driven approach for formalising game structures and the advantages of community contributions for consolidation and reach.

Better Together: The Interplay Between a Phishing Awareness Video and a Link-centric Phishing Support Tool

Two popular approaches for helping consumers avoid phishing threats are phishing awareness videos and tools supporting users in identifying phishing emails. Awareness videos and tools have each been shown on their own to increase people's phishing detection rate. Videos have been shown to be a particularly effective awareness measure; link-centric warnings have been shown to provide effective tool support. However, it is unclear how these two approaches compare to each other. We conducted a between-subjects online experiment (n=409) in which we compared the effectiveness of the NoPhish video and the TORPEDO tool and their combination. Our main findings suggest that the TORPEDO tool outperformed the NoPhish video and that the combination of both performs significantly better than just the tool. We discuss the implications of our findings for the design and deployment of phishing awareness measures and support tools.