Disinformation is among the top risks of generative artificial intelligence (AI) misuse.Global adoption of generative AI necessitates redteaming evaluations (i.e., systematic adversarial probing) that are robust across diverse languages and cultures, but red-teaming datasets are commonly US-and English-centric.To address this gap, we propose "anecdoctoring", a novel red-teaming approach that automatically generates adversarial prompts across languages and cultures.We collect misinformation claims from fact-checking websites in three languages (English, Spanish, and Hindi) and two geographies (US and India).We then cluster individual claims into broader narratives and characterize the resulting clusters with knowledge graphs, with which we augment an attacker LLM.Our method produces higher attack success rates and offers interpretability benefits relative to few-shot prompting.Results underscore the need for disinformation mitigations that scale globally and are grounded in realworld adversarial misuse.

The digital divide refers to disparities in access to and use of digital tooling across social and economic groups. This divide can reinforce marginalization both at the individual level and at the level of places, because persistent economic advantages accrue to places where new technologies are adopted early. To what extent are emerging generative artificial intelligence (AI) tools subject to these social and spatial divides? We leverage a large-scale search query database to characterize U.S. residents' knowledge of a novel generative AI tool, ChatGPT, during its first six months of release. We identify hotspots of higher-than-expected search volumes for ChatGPT in coastal metropolitan areas, while coldspots are evident in the American South, Appalachia, and the Midwest. Nationwide, counties with the highest rates of search have proportionally more educated and more economically advantaged populations, as well as proportionally more technology and finance-sector jobs in comparison with other counties or with the national average. Observed associations with race/ethnicity and urbanicity are attenuated in fully adjusted hierarchical models, but education emerges as the strongest positive predictor of generative AI awareness. In the absence of intervention, early differences in uptake show a potential to reinforce existing spatial and socioeconomic divides.

Chatbots have shown promise as tools to scale qualitative data collection. Recent advances in Large Language Models (LLMs) could accelerate this process by allowing researchers to easily deploy sophisticated interviewing chatbots. We test this assumption by conducting a large-scale user study (n=399) evaluating 3 different chatbots, two of which are LLM-based and a baseline which employs hard-coded questions. We evaluate the results with respect to participant engagement and experience, established metrics of chatbot quality grounded in theories of effective communication, and a novel scale evaluating ''richness'' or the extent to which responses capture the complexity and specificity of the social context under study. We find that, while the chatbots were able to elicit high-quality responses based on established evaluation metrics, the responses rarely capture participants' specific motives or personalized examples, and thus perform poorly with respect to richness. We further find low inter-rater reliability between LLMs and humans in the assessment of both quality and richness metrics. Our study offers a cautionary tale for scaling and evaluating qualitative research with LLMs.

Urban environmental monitoring campaigns depend on expertise from city agencies, residents, and researchers. Deployment efforts rarely include all three stakeholders, typically leading to initiatives that struggle to produce credible, actionable data. We describe the implementation of a large-scale, long-term air quality sensing network in Chicago Illinois; detail stakeholder interviews and meetings; and present three interfaces—–a website accessible via in-situ QR codes, APIs, and a mobile, mixed-media experience. We show how a collaborative approach created a more equitable sensor distribution compared to crowdsourced or regulatory designs. We highlight shared goals of education, engagement, and empowerment despite the diversity of tool and analytics needs across stakeholder groups. Reflecting on our work, we develop a “three-legged stool” framework representing the criticality of balanced participation from three key stakeholder groups—city, community, and research—in deploying novel urban technologies. This approach can help HCI researchers facilitate more democratic technology deployments in urban spaces.

This paper presents Eclipse, a platform for low-cost urban environmental sensing using solar-powered and cellular-connected devices. Dense sensor networks promise to monitor pollution at fine spatial and temporal resolutions, yet few cities have actually implemented such networks due to high costs and limited accuracy. We address these barriers by developing an end-to-end framework for urban air quality sensing with minimal infrastructure requirements. We designed an unobtrusive device that collects data on fine particulate matter (PM2.5), temperature, relative humidity, and barometric pres-sure. A modular design further includes four low-cost gas sensors - Ozone (03), Nitrogen Dioxide (NO2), Sulfur Dioxide (SO2), and Carbon Monoxide (CO) - selected based on local priorities. We deployed 115 devices across Chicago, reliably collecting data for over 90% of expected sensor-hours from July 2 - September 30, 2021. We further developed a calibration strategy that reduced errors by 41.2 – 98.8%, improving accuracy to levels recommended for hotspot detection (PM2.5and 03) or education (NO2and SO2). Through this work, we offer insights on the real-world deployment of a replicable, large-scale, end-to-end platform for hyperlocal urban environmental sensing.