Professional self-representation involves constructing identities that reflect personal values while aligning with the norms of professional communities. Many people turn to generative AI for help, but misalignments between LLM outputs and self-understanding hinder authenticity and accuracy of the content. To explore how LLMs can support co-creation aligned, authentic self-representational content, we designed LL.me, a web-based probe based on bi-directional alignment that utilizes users’ resumes and guides them through iterative cycles of refining AI-generated self-representations. Our user study with 14 participants showed users engaged in identity work with the tool, re-framing content to emphasize their personal values, imparting tacit knowledge from their communities of practice, and leveraging system explainability features as a proxy for how the representation would be perceived by others. We demonstrate how LLM-based tools can facilitate a co-constructive process of identity formation, helping individuals actively shape their professional self-representations in collaboration with the AI.

Call agents, a representative group of emotion workers, must manage emotions under constrained autonomy, yet workplace stress sensing has primarily centered on knowledge work. We ask how the task‑aligned cycle of emotional labor, alternating customer interaction (CI) and non‑customer interaction (nCI), shapes stress and how it manifests in data. We conducted a month-long in-the-wild formative mixed-methods study with professional call agents, collecting structured task logs, environmental and behavioral signals, and per-call stress self-reports, followed by semi-structured interviews. Task logs, used as a new sensor modality, were incorporated as primary sensing signals, and task-related features were extracted by respecting CI boundaries for modeling. Our results showed that a short 5-minute windowing approach was comparable to task-aligned windowing using multimodal sensors, with task-related features being considered the most important across all generalized models. Personalized models improved further and shifted importance toward diverse data sources, revealing individual differences in preparation patterns. Interviews support those findings, reveal key modelling challenges, and highlight potential benefits of semi-automated self-tracking. We discuss implications for timing interventions at breakpoints suited for work patterns, and ethically deploying stress support for emotion workers.

Cancer surgery is a key treatment for gastrointestinal (GI) cancers, a group of cancers that account for more than 35% of cancer-related deaths worldwide, but postoperative complications are unpredictable and can be life-threatening. In this paper, we investigate how recent advancements in large language models (LLMs) can benefit remote patient monitoring (RPM) systems through clinical integration by designing RECOVER, an LLM-powered RPM system for postoperative GI cancer care. To closely engage stakeholders in the design process, we first conducted seven participatory design sessions with five clinical staff and interviewed five cancer patients to derive six major design strategies for integrating clinical guidelines and information needs into LLM-based RPM systems. We then designed and implemented RECOVER, which features an LLM-powered conversational agent for cancer patients and an interactive dashboard for clinical staff to enable efficient postoperative RPM. Finally, we used RECOVER as a pilot system to assess the implementation of our design strategies with four clinical staff and five patients, providing design implications by identifying crucial design elements, offering insights on responsible AI, and outlining opportunities for future LLM-powered RPM systems.

Large Language Models are often trained on data reflecting neurotypical norms, yet are increasingly deployed to support neurodivergent users in sensitive domains like job-seeking.We examine interactions between neurodivergent job-seekers and a GPT-4powered career support chatbot through the lens of misalignment.Through analysis of over 300 chat logs and interviews with 15 neurodivergent participants, we found that the chatbot frequently misrepresented users' skills, imposed neurotypical language and expectations, and provided generic or inappropriate advice-even when relevant user data was available.Participants expected the chatbot to interpret implicit insights from their data, however, they sometimes lacked the clarity or confidence to correct the system when it did not, revealing gaps in both AI design and user understanding of system function.Our findings underscore the need for bi-directional alignment between neurodivergent users and LLMs, and call for design strategies that integrate neurodivergent perspectives and preferences to ensure more authentic, personalized, and human-centered AI support.

Client-Service Representatives (CSRs) are vital to organizations.Frequent interactions with disgruntled clients, however, disrupt their mental well-being.To help CSRs regulate their emotions while interacting with uncivil clients, we designed Care-Pilot, an LLM-powered assistant, and evaluated its efficacy, perception, and use.Our comparative analyses between 665 human and Care-Pilotgenerated support messages highlight Care-Pilot's ability to adapt to and demonstrate empathy in various incivility incidents.Additionally, 143 CSRs assessed Care-Pilot's empathy as more sincere and actionable than human messages.Finally, we interviewed 20 CSRs who interacted with Care-Pilot in a simulation exercise.They reported that Care-Pilot helped them avoid negative thinking, recenter thoughts, and humanize clients; showing potential for bridging gaps in coworker support.Yet, they also noted deployment challenges and emphasized the indispensability of shared experiences.We discuss future designs and societal implications of AI-mediated emotional labor, underscoring empathy as a critical function for AI assistants for worker mental health.

The future of work is rapidly changing, with remote and hybrid settings blurring the boundaries between professional and personal life. To understand how work rhythms vary across different work settings, we conducted a month-long study of 65 software developers, collecting anonymized computer activity data as well as daily ratings for perceived stress, productivity, and work setting. In addition to confirming the double-peak pattern of activity at 10:00 am and 2:00 pm observed in prior research, we observed a significant third peak around 9:00 pm. This third peak was associated with higher perceived productivity during remote days but increased stress during onsite and hybrid days, highlighting a nuanced interplay between work demands and work settings. Additionally, we found strong correlations between computer activity, productivity, and stress, including an inverted U-shaped relationship where productivity peaked at around six hours of computer activity before declining on more active days. These findings provide new insights into evolving work rhythms and highlight the impact of different work settings on productivity and stress.

Advances in mobile and wearable technologies have enabled the potential to passively monitor a person's mental, behavioral, and affective health. These approaches typically rely on longitudinal self-reported outcomes, e.g., depression, stress, and anxiety, to train machine learning models. However, the continuous need to self-report various internal states adds a significant burden on the participants, often resulting in attrition, missing labels, or insincere responses. In this work, we introduce the Scale Scores Simulation using Mental Models (SeSaMe) framework to alleviate participants' burden in digital mental health studies. By leveraging pre-trained large language models (LLMs), SeSaMe enables the simulation of participants' responses on psychological scales. In SeSaMe, researchers can prompt LLMs with information on participants' internal behav-ioral dispositions, enabling LLMs to construct mental models of participants to simulate their responses on psychological scales. As part of the framework, we provide four evaluation metrics to assess the effectiveness of the simulated responses. We demonstrate an application of SeSaMe, where we use GPT-4 to simulate responses on one scale using responses from another as behavioral information. We use SeSaMe's evaluation metrics to assess the alignment between human and SeSaMe-simulated responses to psychological scales. Then, we present multiple experiments to inspect the utility of SeSaMe-simulated responses as ground truth in training machine-learning models by replicating established depression and anxiety screening tasks with passive sensing data from a previous study. Our results indicate SeSaMe to be a promising approach, but its alignment may vary across scales and specific prediction objectives. We also observed that model performance with simulated data was on par with using the real data for training in most evaluation scenarios. We conclude by discussing the potential implications of SeSaMe in addressing some challenges with ground-truth collection in passive sensing studies.

Algorithmic estimations of worker behavior are gaining popularity. Passive Sensing–enabled AI (PSAI) systems leverage behavioral traces from workers’ digital tools to infer their experience. Despite their conceptual promise, the practical designs of these systems elicit tensions that lead to workers resisting adoption. This paper teases apart the monolithic representation of PSAI by investigating system components that maximize value and mitigate concerns. We conducted an interactive online survey using the Experimental Vignette Method. Using Linear Mixed-effects Models we found that PSAI systems were more acceptable when sensing digital time use or physical activity, instead of visual modes. Inferences using language were only acceptable in work-restricted contexts. Compared to insights into performance, workers preferred insights into mental wellbeing. However, they resisted systems that automatically forwarded these insights to others. Our findings provide a template to reflect on existing systems and plan future implementations of PSAI to be more worker-centered.

We are witnessing an emergence in Passive Sensing enabled AI (PSAI) to provide dynamic insights for performance and wellbeing of information workers. Hybrid work paradigms have simultaneously created new opportunities for PSAI, but have also fostered anxieties of misuse and privacy intrusions within a power asymmetry. At this juncture, it is unclear if those who are sensed can find these systems acceptable. We conducted scenario-based interviews of 28 information workers to highlight their perspectives as data subjects in PSAI. We unpack their expectations using the Contextual Integrity framework of privacy and information gathering. Participants described appropriateness of PSAI based on its impact on job consequences, work-life boundaries, and preservation of flexibility. They perceived that PSAI inferences could be shared with selected stakeholders if they could negotiate the algorithmic inferences. Our findings help envision worker-centric approaches to implementing PSAI as an empowering tool in the future of work.

Information workers often struggle to balance their time for a variety of activities like focused work, communication, and caring. This study analyzes the impact of a commercially available computer-assisted time protection intervention that automatically and preemptively schedules calendar time for self-determined activities. We analyzed the behaviors and self-reports of workers in two naturalistic studies. First, we studied 27 workers who were already using Computer-Assisted Protected Time (CAP time) and found that they mainly used it for focused work. Second, we analyzed the effect of CAP time as a randomized intervention on 89 workers who never had CAP time and found that those with it self-reported an increase in performance, job resources, and immersion. In both studies, workers with CAP time exhibited a rearrangement of activities leading to an overall reduction in work activity. This study highlights new opportunities for intelligent time-management interventions and the importance of protected time at work.

Contact tracers assist in containing the spread of highly infectious diseases such as COVID-19 by engaging community members who receive a positive test result in order to identify close contacts. Many contact tracers rely on community member’s recall for those identifications, and face limitations such as unreliable memory. To investigate how technology can alleviate this challenge, we developed a visualization tool using de-identified location data sensed from campus WiFi and provided it to contact tracers during mock contact tracing calls. While the visualization allowed contact tracers to find and address inconsistencies due to gaps in community member’s memory, it also introduced inconsistencies such as false-positive and false-negative reports due to imperfect data, and information sharing hesitancy. We suggest design implications for technologies that can better highlight and inform contact tracers of potential areas of inconsistencies, and further present discussion on using imperfect data in decision making.

When modeling passive data to infer individual mental wellbeing, a common source of ground truth is self-reports. But these tend to represent the psychological facet of mental states, which might not align with the physiological facet of that state. Our paper demonstrates that when what people “feel” differs from what people “say they feel”, we witness a semantic gap that limits predictions. We show that predicting mental wellbeing with passive data (offline sensors or online social media) is related to how the ground-truth is measured (objective arousal or self-report). Features with psycho-social signals (e.g., language) were better at predicting self-reported anxiety and stress. Conversely, features with behavioral signals (e.g., sleep), were better at predicting stressful arousal. Regardless of the source of ground truth, integrating both signals boosted prediction. To reduce the semantic gap, we provide recommendations to evaluate ground truth measures and adopt parsimonious sensing.

Organizational culture (OC) encompasses the underlying beliefs, values, and practices that are unique to an organization. However, OC is inherently subjective and a coarse construct, and therefore challenging to quantify. Alternatively, self-initiated workplace reviews on online platforms like Glassdoor provide the opportunity to leverage the richness of language to understand OC. In as much, first, we use multiple job descriptors to operationalize OC as a word vector representation. We validate this construct with language used in 650k different Glassdoor reviews. Next, we propose a methodology to apply our construct on Glassdoor reviews to quantify the OC of employees by sector. We validate our measure of OC on a dataset of 341 employees by providing empirical evidence that it helps explain job performance. We discuss the implications of our work in guiding tailored interventions and designing tools for improving employee functioning.

The ubiquitous use of social media enables researchers to obtain self-recorded longitudinal data of individuals in real-time. Because this data can be collected in an inexpensive and unobtrusive way at scale, social media has been adopted as a “passive sensor” to study human behavior. However, such research is impacted by the lack of homogeneity in the use of social media, and the engineering challenges in obtaining such data. This paper proposes a statistical framework to leverage the potential of social media in sensing studies of human behavior, while navigating the challenges associated with its sparsity. Our framework is situated in a large-scale in-situ study concerning the passive assessment of psychological constructs of 757 information workers wherein of four sensing streams was deployed - bluetooth beacons, wearable, smartphone, and social media. Our framework includes principled feature transformation and machine learning models that predict latent social media features from the other passive sensors. We demonstrate the efficacy of this imputation framework via a high correlation of 0.78 between actual and imputed social media features. With the imputed features we test and validate predictions on psychological constructs like personality traits and affect. We find that adding the social media data streams, in their imputed form, improves the prediction of these measures. We discuss how our framework can be valuable in multimodal sensing studies that aim to gather comprehensive signals about an individual's state or situation.

Organizations often strive to recruit and retain individuals who would be a "good fit" with their core values, beliefs and practices. Person-Organization (P-O congruence is known to explain employee satisfaction, commitment and absenteeism. This paper proposes a new measure of P-O fit by empirically investigating the similarity of routine within an organization. This measure of routine fit is motivated by the theory of entrainment, which refers to the synchrony of individual and community behaviors. We use unobtrusive bluetooth sensing to examine how the concurrence of latent activity patterns is related to job performance and wellbeing. Routine fit echoes traditional constructs of congruence as it is significantly related to higher task performance and lower workplace deviance. Additionally however, it is also related to greater stress and higher arousal. Prior work in organizational psychology have used single-occasion survey instruments to infer uni-dimensional models of fit. These methods are limited by subjective perceptions of employees. In contrast, we demonstrate a data-driven and multidimensional approach to study normative routines in an organization as a measure of P-O fit. We discuss the potential of our approach in designing technologies that understand the congruence of employee routines and positively impact employee functioning at the workplace.