Despite the potential of self-supervised video alignment algorithms for advancing human-computer interaction, they remain largely inaccessible to practitioners without machine learning expertise. To bridge this gap, we introduce VideoAlign, an open-source toolkit designed to facilitate training and integration of video alignment approaches in interactive applications. VideoAlign offers guidance for training models to align videos, detecting and tagging specific events, and performing anomaly detection through an interactive system without requiring machine learning experience. In addition to implementing state-of-the-art alignment techniques, our toolkit introduces a novel Local-Alignment Contrastive (LAC) loss. Unlike global methods that compare entire video sequences, LAC aligns localized segments independently. This capability enables robust matching when video structure or timing varies, which is crucial for real-world interactive applications. We demonstrate how VideoAlign facilitates the creation of a wide variety of interactive applications through three application scenarios: a teacher-support tool for providing efficient video-feedback, a mixed reality application that tracks activity progress in real-time, and an anomaly detection tool to monitor cooking activities.

Anchoring - the choice of frame of reference for mixed reality (MR) interface elements - is a critical design decision involving trade-offs between accessibility, interaction comfort, and visual interference. Despite its importance, user preferences for anchoring across different mobility contexts and interface properties remain poorly understood, as prior work has largely focused on specific tasks or fixed interface configurations. We address this through a mixed-methods user study in which participants configure anchoring strategies across different mobility conditions and interface types. Combining behavioral analysis with structured qualitative inquiry, we analyze how participants select and reason about anchoring modes. Our results show a clear transition from world-anchored interfaces in stationary contexts to body-anchored interfaces during locomotion. However, no single body anchor consistently dominates, highlighting the personal nature of anchoring strategies. Our qualitative analysis reveals the factors users consider in their anchoring decision, including interface accessibility, stability during interaction, visual clutter, and individual mental models. These findings inform the design of adaptive and controllable MR interfaces and highlight the importance of supporting user customization.

The use of Artificial Intelligence (AI) in high-risk, decision-making scenarios presents technical, safety, and normative challenges; problems that may only be ameliorated by human oversight. However, notions of human oversight lack a common foundational understanding: oversight architectures are not well defined, the roles involved remain unclear, and implementation steps are opaque. Hence, researchers and practitioners struggle to determine how to design, implement, and evaluate systems that enable effective human oversight. This paper advances a practical framework for effective human oversight of AI systems, based on a cross-disciplinary perspective that draws on insights from computer science, human-computer interaction, psychology, philosophy, and law. The core contributions are: (1) a foundational framework, with a working definition, architecture and processes for effective human oversight of AI systems; (2) an initial template for documenting oversight architectures and processes, applied to diverse domains; and (3) a synthesis of open research challenges that need to be considered in the emerging field of effective human oversight of AI systems.

Interfaces for human oversight must effectively support users' situation awareness under time-critical conditions. We explore reinforcement learning (RL)-based UI adaptation to personalize alerting strategies that balance the benefits of highlighting critical events against the cognitive costs of interruptions. To enable learning without real-world deployment, we integrate models of users' gaze behavior to simulate attentional dynamics during monitoring. Using a delivery-drone oversight scenario, we present initial results suggesting that RL-based highlighting can outperform static, rule-based approaches and discuss challenges of intelligent oversight support.

Intelligent text entry (ITE) methods, such as word suggestions, are widely used in mobile typing, yet improving ITE systems is challenging because the cognitive mechanisms behind suggestion use remain poorly understood, and evaluating new systems often requires long-term user studies to account for behavioral adaptation. We present WSTypist, a reinforcement learning-based model that simulates how typists integrate word suggestions into typing. It builds on recent hierarchical control models of typing, but focuses on the cognitive mechanisms that underlie the high-level decision-making for effectively integrating word suggestions into manual typing: assessing efficiency gains, considering orthographic uncertainties, and including personal reliance on AI support. Our evaluations show that WSTypist simulates diverse human-like suggestion-use strategies, reproduces individual differences, and generalizes across different systems. Importantly, we demonstrate on four design cases how computational rationality models can be used to inform what-if analyses during the design process, by simulating how users might adapt to changes in the UI or in the algorithmic support, reducing the need for long-term user studies.

Human-in-the-loop optimization identifies optimal interface designs by iteratively observing user performance. However, it often requires numerous iterations due to the lack of prior information. While recent approaches have accelerated this process by leveraging previous optimization data, collecting user data remains costly and often impractical. We present a conceptual framework, Human-in-the-Loop Optimization with Model-Informed Priors (HOMI), which augments human-in-the-loop optimization with a training phase where the optimizer learns adaptation strategies from diverse, synthetic user data generated with predictive models before deployment. To realize HOMI, we introduce Neural Acquisition Function+ (NAF+), a Bayesian optimization method featuring a neural acquisition function trained with reinforcement learning. NAF+ learns optimization strategies from large-scale synthetic data, improving efficiency in real-time optimization with users. We evaluate HOMI and NAF+ with mid-air keyboard optimization, a representative VR input task. Our work presents a new approach for more efficient interface adaptation by bridging in situ and in silico optimization processes.

Proactive assistance with large language models (LLMs) has received growing attention in the human computer interaction (HCI) community. However, most past work on proactive LLMs' assistance has focused on adult users and task-oriented settings, leaving open how such systems could support children, whose interests and needs are often expressed through gaze and other nonverbal behaviors rather than explicit requests. In this study, we focus on two key challenges of proactive assistance in children's picture exploration: when to provide assistance and what assistance to provide based on children's nonverbal behaviors. To address these challenges, we introduce Ollie, a gaze-informed proactive artificial intelligence (AI) assistant that offers short narrative descriptions based on where a child is looking. Ollie uses children's gaze to estimate their attention, identify their current visual focus, and select a related picture region for the LLM to verbally describe. In a within-subject experiment, we compared gaze-informed assistance with random assistance. Results show that gaze-informed assistance kept children's attention on their current focus for a longer period of time, and guided them more effectively to related picture regions. Children, parents, and a participating kindergarten teacher viewed Ollie positively and consider that it better matched children's interests when compared with the random assistance. This work shows the feasibility of using gaze as an implicit input for proactive AI assistance for children and provides design implications for future child-centered AI systems.

The growing use of AI-generated responses in everyday tools raises concerns about how subtle features such as supporting detail or tone of confidence may shape people’s beliefs. To understand this, we conducted a pre-registered online experiment (N=304) investigating how the detail and confidence of AI-generated responses influence belief change. We introduce an analysis framework with two targeted measures: belief switch and belief shift. These measures distinguish between users changing their initial stance after AI input and the extent to which they adjust their conviction toward or away from the AI’s stance, thereby quantifying not only categorical changes but also more subtle, continuous adjustments in belief strength that indicate reinforcement or weakening of existing beliefs. Using this framework, we find that detailed responses with medium confidence are associated with the largest overall belief changes. Highly confident messages tend to induce belief shifts but fewer stance reversals. Our results also show that task type (fact-checking versus opinion evaluation), prior conviction, and perceived stance agreement further modulate the extent and direction of belief change. These results illustrate how different properties of AI responses interact with user beliefs in often subtle but potentially consequential ways and raise practical as well as ethical considerations for the design of LLM-powered systems.

The growing use of AI-generated responses in everyday tools raises concern about how subtle features such as supporting detail or tone of confidence may shape people's beliefs. To understand this, we conducted a pre-registered online experiment (N = 304) investigating how the detail and confidence of AI-generated responses influence belief change. We introduce an analysis framework with two targeted measures: belief switch and belief shift. These distinguish between users changing their initial stance after AI input and the extent to which they adjust their conviction toward or away from the AI's stance, thereby quantifying not only categorical changes but also more subtle, continuous adjustments in belief strength that indicate a reinforcement or weakening of existing beliefs. Using this framework, we find that detailed responses with medium confidence are associated with the largest overall belief changes. Highly confident messages tend to elicit belief shifts but induce fewer stance reversals. Our results also show that task type (fact-checking versus opinion evaluation), prior conviction, and perceived stance agreement further modulate the extent and direction of belief change. These findings illustrate how different properties of AI responses interact with user beliefs in subtle but potentially consequential ways and raise practical as well as ethical considerations for the design of LLM-powered systems.

Monitoring interfaces are crucial for dynamic, high-stakes tasks where effective user attention is essential. Visual highlights can guide attention effectively, but may also introduce unintended disruptions. To investigate this, we examined how visual highlights affect users’ gaze behavior in a drone monitoring task, focusing on when, how long, and how much attention they draw. We found that highlighted areas exhibit distinct temporal characteristics compared to nonhighlighted ones, quantified using normalized saliency (NS) metrics. We found that highlights elicited immediate responses, with NS peaking quickly, but this shift came at the cost of reduced search efforts elsewhere, potentially impacting situational awareness. To predict these dynamic changes and support interface design, we developed the Highlight-Informed Saliency Model, which provides granular predictions of NS over time. These predictions enable evaluations of highlight effectiveness and inform the optimal timing and deployment of highlights in future monitoring interface designs, particularly for time-sensitive tasks.

Word suggestions are commonly used when people type on mobile devices. However, how users adjust their typing behavior and visual attention to integrate the use of word suggestions and whether they are effective in doing so remains unclear, mainly due to the lack of gaze data in realistic settings. In this paper, we conduct an eye-tracking study of word suggestion users transcribing and composing text on their own phones and keyboards. Our analysis reveals that users frequently checked the suggestion list without picking a suggestion, yielding a 68% failure rate. Screen recordings show that only about half of these Failed Suggestions can be attributed to the algorithm's performance. In 43.6% of cases, users typed the word manually even though they fixated on the correctly suggested word. We analyze the dynamics of users' checking behavior and quantify the time cost of checking for word suggestions. Overall, we find that despite using word suggestions on a daily basis, users' checking behavior is not well aligned with the performance of the suggestion algorithm, resulting in a decrease of typing speed. These findings have implications for the design of intelligent text entry systems and AI support in general, and our WS-Gaze dataset will support future research in this important direction.

In time-critical detection tasks, such as drone monitoring, a key condition for users to effectively leverage AI assistance is to find an appropriate trade-off between making fast decisions and verifying AI suggestions, which we refer to as appropriate user reliance. However, assessing such reliance is often oversimplified by focusing solely on task outcomes, potentially overlooking whether users properly verify AI messages. We collected eye-tracking data from an AI-assisted monitoring task and developed a gaze-based reliance model: RelEYEance, to assess the extent of user reliance on AI-suggested alarms. We found that gaze patterns related to verification behaviors distinguish between appropriate reliance, over-reliance, and under-reliance, influencing task performance. We validated our model in a second user study, showing it can reliably detect users' over- and under-reliance at run-time, which could be used e.g. for issuing intervention messages. The results demonstrate the potential for real-time human-AI reliance assessment, facilitating adaptive reliance calibration.

This study examines the role of visual highlights in guiding user attention in drone monitoring tasks, employing a simulated interface for observation. The experiment results show that such highlights can significantly expedite the visual attention on the corresponding area. Based on this observation, we leverage both the temporal and spatial information in the highlight to develop a new saliency model: the highlight-informed saliency model (HISM), to infer the visual attention change in the highlight condition. Our findings show the effectiveness of visual highlights in enhancing user attention and demonstrate the potential of incorporating these cues into saliency prediction models.

Robust frame-wise embeddings are essential to perform video analysis and understanding tasks. We present a self-supervised method for representation learning based on aligning temporal video sequences. Our framework uses a transformer-based encoder to extract frame-level features and leverages them to find the optimal alignment path between video sequences. We introduce the novel Local-Alignment Contrastive (LAC) loss, which combines a differentiable local alignment loss to capture local temporal dependencies with a contrastive loss to enhance discriminative learning. Prior works on video alignment have focused on using global temporal ordering across sequence pairs, whereas our loss encourages identifying the best-scoring subsequence alignment. LAC uses the differentiable Smith-Waterman (SW) affine method, which features a flexible parameterization learned through the training phase, enabling the model to adjust the temporal gap penalty length dynamically. Evaluations show that our learned representations outperform existing state-of-the-art approaches on action recognition tasks.

This study examines the role of visual highlights in guiding user attention in drone monitoring tasks, employing a simulated interface for observation. The experiment results show that such highlights can significantly expedite the visual attention on the corresponding area. Based on this observation, we leverage both the temporal and spatial information in the highlight to develop a new saliency model: the highlight-informed saliency model (HISM), to infer the visual attention change in the highlight condition. Our findings show the effectiveness of visual highlights in enhancing user attention and demonstrate the potential of incorporating these cues into saliency prediction models.

Visual highlighting can guide user attention in complex interfaces. However, its effectiveness under limited attentional capacities is underexplored. This paper examines the joint impact of visual highlighting (permanent and dynamic) and dual-task-induced cognitive load on gaze behaviour. Our analysis, using eye-movement data from 27 participants viewing 150 unique webpages reveals that while participants' ability to attend to UI elements decreases with increasing cognitive load, dynamic adaptations (i.e., highlighting) remain attention-grabbing. The presence of these factors significantly alters what people attend to and thus what is salient. Accordingly, we show that state-of-the-art saliency models increase their performance when accounting for different cognitive loads. Our empirical insights, along with our openly available dataset, enhance our understanding of attentional processes in UIs under varying cognitive (and perceptual) loads and open the door for new models that can predict user attention while multitasking.

We explore a novel transcription task in mobile text entry research, presenting stimuli within LLM-generated conversational contexts to improve participant engagement and phrase memorability. We conducted two studies: an eye-tracking study examining participants’ attention when presented with conversational contexts alongside stimuli, and an experiment comparing LLM-generated and human-generated prompt-response pairs in transcription tasks, involving both high and low memorability stimuli. Key findings reveal that presenting conversational contexts improves recall for low memorability phrases and results in fewer uncorrected errors during transcription. No significant effects were observed on other basic text entry metrics, or participant subjective appraisals of engagement with the novel task, suggesting it can be used safely as an alternative to the traditional transcription task. We discuss the potential of LLMs in improving text entry evaluation methods, including generating diverse linguistic styles, emotionally loaded contexts, and even simulating entire evaluation processes. Our study highlights the need for systematic approaches to generate and evaluate LLM outputs for research purposes, and for proposing new metrics and evaluation methods.

Mobile word suggestions can slow down typing, yet are still widely used. To investigate the apparent benefits beyond speed, we analyzed typing behavior of 15,162 users of mobile devices. Controlling for natural typing speed (a confounding factor not considered by prior work), we statistically show that slower typists use suggestions more often but are slowed down by doing so. To better understand how these typists leverage suggestions -- if not to improve their speed -- we extract eight usage strategies, including completion, correction, and next-word prediction. We find that word characteristics, such as length or frequency, along with the strategy, are predictive of whether a user will select a suggestion. We show how to operationalize our findings by building and evaluating a predictive model of suggestion selection. Such a model could be used to augment existing suggestion algorithms to consider people's strategic use of word predictions beyond speed and keystroke savings.

This paper proposes a new approach for online UI adaptation that aims to overcome the limitations of the most commonly used UI optimization method involving multiple objectives: weighted sum optimization. Weighted sums are highly sensitive to objective formulation, limiting the effectiveness of UI adaptations. We propose ParetoAdapt, an adaptation approach that uses online multi-objective optimization with a posteriori articulated preferences—that is, articulation of preferences after the optimization has concluded—to make UI adaptation robust to incomplete and inaccurate objective formulations. It offers users a flexible way to control adaptations by selecting from a set of Pareto optimal adaptation proposals and adjusting them to fit their needs. We showcase the feasibility and flexibility of ParetoAdapt by implementing an online layout adaptation system in a state-of-the-art 3D UI adaptation framework. We further evaluate its robustness and run-time in simulation-based experiments that allow us to systematically change the accuracy of the estimated user preferences. We conclude by discussing how our approach may impact the usability and practicality of online UI adaptations.

Adaptive user interfaces can improve experiences in Extended Reality (XR) applications by adapting interface elements according to the user's context. Although extensive work explores different adaptation policies, XR creators often struggle with their implementation, which involves laborious manual scripting. The few available tools are underdeveloped for realistic XR settings where it is often necessary to consider conflicting aspects that affect an adaptation. We fill this gap by presenting AUIT, a toolkit that facilitates the design of optimization-based adaptation policies. AUIT allows creators to flexibly combine policies that address common objectives in XR applications, such as element reachability, visibility, and consistency. Instead of using rules or scripts, specifying adaptation policies via adaptation objectives simplifies the design process and enables creative exploration of adaptations. After creators decide which adaptation objectives to use, a multi-objective solver finds appropriate adaptations in real-time. A study showed that AUIT allowed creators of XR applications to quickly and easily create high-quality adaptations.