Representation shapes public attitudes and behaviors. With the recent advances and rapid adoption of LLMs, the way these systems are introduced will negotiate societal expectations for their role in high-stakes domains like health. Yet it remains unclear whether current narratives present a balanced view. We analyzed five prominent discourse channels (news, research press, YouTube, TikTok, and Reddit) over a two-year period on lexical style, informational content, and symbolic representation. Discussions were generally positive and episodic, with positivity increasing over time. Risk communication was unthorough and often reduced to information quality incidents, while explanations of LLMs’ generative nature were rare. Compared with professional outlets, TikTok and Reddit highlighted wellbeing applications and showed greater variations in tone and anthropomorphism but little attention to risks. We discuss implications for public discourse as a diagnostic tool in identifying literacy and governance gaps, and for communication and design strategies to support more informed LLM engagement.

Recent breakthroughs in large language models (LLMs) have generated both interest and concern about their potential adoption as information sources or communication tools across different domains. In public health, where stakes are high and impacts extend across diverse populations, adopting LLMs poses unique challenges that require thorough evaluation. However, structured approaches for assessing potential risks in public health remain under-explored. To address this gap, we conducted focus groups with public health professionals and individuals with lived experience to unpack their concerns, situated across three distinct and critical public health issues that demand high-quality information: infectious disease prevention (vaccines), chronic and wellbeing care (opioid use disorder), and community health and safety (intimate partner violence). We synthesize participants' perspectives into a risk taxonomy, identifying and contextualizing the potential harms LLMs may introduce when positioned alongside traditional health communication. This taxonomy highlights four dimensions of risk to individuals, human-centered care, information ecosystems, and technology accountability. For each dimension, we unpack specific risks and offer example reflection questions to help practitioners adopt a risk-reflexive approach. By summarizing distinctive LLM characteristics and linking them to identified risks, we discuss the need to revisit prior mental models of information behaviors and complement evaluations with external validity and domain expertise through lived experience and real-world practices. Together, this work contributes a shared vocabulary and reflection tool for people in both computing and public health to collaboratively reflect and assess risks in deciding when to employ LLM capabilities (or not) and how to mitigate harm.

During times of crisis, heightened anxiety and fear create fertile ground for hate speech and misinformation, as people are more likely to fall for and be influenced by it. This paper looks into the interwoven relationship between anti-Asian hatred and COVID-19 misinformation amid the pandemic. By analyzing 785,798 Asian hate tweets and surveying 308 diverse participants, this empirical study explores how hateful content portrays the Asian community, including its truthfulness and targets, as well as what makes such portrayals harmful. We observed a high prevalence of misinformative hate speech that was lengthier, less emotional, and expressed more motivational drives than general hate speech. Overall, we found that anti-Asian rhetoric was characterized by an antagonism and inferiority framing, with misinformative hate underscoring antagonism and general hate emphasizing calls for action. Among all entities being explicitly criticized, China and the Chinese were constantly named to assign blame, with misinformative hate more likely to finger-point than general hate. Our survey results indicated that hateful messages with misinformation, demographic targeting, or divisive references were perceived as significantly more damaging. Individuals who placed less importance on free speech, had personal encounters with hate speech, or believed in the natural origin of COVID-19 were more likely to perceive higher severity. Taken together, this work highlights the distinct compositions of hate within misinformative hate speech that influences perceived harmfulness and adds to the complexity of defining and moderating harmful content. We discuss the implications for designing more context- and culture-sensitive counter-strategies and building more adaptive and explainable moderation approaches.

Gaurav Verma, Rynaa Grover, Jiawei Zhou, Binny Mathew, Jordan Kraemer, Munmun Choudhury, Srijan Kumar. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Today, clinicians have limited visibility into the quality of homework exercises that occur outside of the clinical context; however, understanding patient performance in these exercises is essential for guiding patient-centered care. To address this, we present the Clinician Homework Review (CHR), a unique measure and interface that displays similarity ratings calculated using sensor-captured patient-generated data (sPGD; i.e. heart rate, phone usage, ambient noise, and physical activity) for therapeutic exercises outside of the clinical setting within the post-traumatic stress disorder (PTSD) treatment context. Through concept testing sessions with 10 clinicians, we examine how sPGD can be leveraged to measure and investigate what contributes to patient performance in a therapeutic exercise. We also share in-depth information regarding clinician interpretation and planned use of data displayed by CHR in clinical sessions with patients. We frame our results in the context of situated objectivity and propose the notion of "perceived reference weight," which describes the significance attributed to contextualized data. In doing so, we support clinical decision-making in PTSD therapy.

Large language models have abilities in creating high-volume human-like texts and can be used to generate persuasive misinformation. However, the risks remain under-explored. To address the gap, this work first examined characteristics of AI-generated misinformation (AI-misinfo) compared with human creations, and then evaluated the applicability of existing solutions. We compiled human-created COVID-19 misinformation and abstracted it into narrative prompts for a language model to output AI-misinfo. We found significant linguistic differences within human-AI pairs, and patterns of AI-misinfo in enhancing details, communicating uncertainties, drawing conclusions, and simulating personal tones. While existing models remained capable of classifying AI-misinfo, a significant performance drop compared to human-misinfo was observed. Results suggested that existing information assessment guidelines had questionable applicability, as AI-misinfo tended to meet criteria in evidence credibility, source transparency, and limitation acknowledgment. We discuss implications for practitioners, researchers, and journalists, as AI can create new challenges to the societal problem of misinformation.

Past research has demonstrated that accounts of trusted others can provide additional context into real world behavior relevant to clinical decision-making and patient engagement. Our research investigates the Social Sensing System, a concept which leverages trusted other feedback for veterans in therapy for PTSD. In our two phase study, we work with 10 clinicians to develop text-message queries and realistic scenarios to present to patients and trusted others. We then present the results in the form of a storyboard to 10 veterans with PTSD and 10 trusted others and gather feedback via semi-structured interview and survey. We find that while trusted other feedback may provide a unique and useful perspective, key design features and considerations of underlying relationships must be considered. We present our findings and utilize the mechanisms and conditions framework to assess the power dynamics of systems such as social sensing in the mental health realm.

Veterans are a unique marginalized group facing multiple vulnerabilities. Current assessments of veteran needs and support largely come from first-person accounts guided by researchers' prompts. Social media platforms not only enable veterans to connect with each other, but also to self-disclose experiences and seek support. This paper addresses the gap in our understanding of veteran needs and their own support dynamics by examining self-initiated and ecologically-valid self-expressions. In particular, we adopt the Veteran Critical Theory (VCT) to conduct a computational study on the Reddit community of veterans. Using topic modeling, we find veteran-friendly gestures with good intentions might not be appreciated in the subreddit. By employing transfer learning methodologies, we find this community has more informational and emotional support behaviors than general online communities and a higher prevalence of informational support than emotional support. Lastly, an examination of support dynamics reveals some contrasts to previous scholarship in military culture and social media. We discover that positive language and author platform tenure have negative relations with posts receiving replies and replies getting votes, and that replies reflecting personal disclosures tend to get more votes. Through the lens of VCT, we discuss how online communities can help uncover veterans' needs and provide more effective social support.