Millions of users across the globe turn to AI chatbots for their creative needs, inviting widespread interest in understanding how they represent diverse cultures. However, evaluating cultural representations in open-ended tasks remains challenging and underexplored. In this work, we present TALES, an evaluation of cultural misrepresentations in LLM-generated stories for diverse Indian cultural identities. First, we develop TALES-Tax, a taxonomy of cultural misrepresentations by collating insights from participants with lived experiences in India through focus groups (N=9) and individual surveys (N=15). Using TALES-Tax, we evaluate 6 models through a large-scale annotation study spanning 2,925 annotations from 108 annotators with lived experience and native language proficiency from across 71 regions in India and 14 languages. Concerningly, we find that 88% of the generated stories contain misrepresentations, and such errors are more prevalent in mid- and low-resourced languages and stories based in peri-urban regions in India. We also transform the annotations into TALES-QA, a standalone question bank to evaluate the cultural knowledge of models.

AI technologies are increasingly deployed in high-stakes domains such as education, healthcare, law, and agriculture to address complex challenges in non-Western contexts. This paper examines eight real-world deployments spanning seven countries and 18 languages, combining 17 interviews with AI developers and domain experts with secondary research. Our findings identify six cross-cutting factors — Language, Institution, Safety, Task, End-User Demography, and Domain — that structured how systems were designed and deployed. These factors were shaped by Sociocultural (diversity, practices), Institutional (resources, policies), and Technological (capabilities, limits) influences. We find that building effective AI systems required extensive collaboration between AI developers and domain experts, with human resources proving more critical to achieving safe and effective outcomes in high-stakes domains than technological expertise alone. Additionally, we present 12 guidelines synthesizing these dynamics for designing AI for social good systems that are culturally grounded, equitable, and responsive to the needs of non-Western contexts.

Large language models (LLMs) are increasingly used to promote prosocial and constructive discourse online. Yet little is known about how these models negotiate and shape underlying values when reframing people’s arguments on value-laden topics. We conducted experiments with 465 participants from India and the United States, who wrote comments on homophobic and Islamophobic threads, and reviewed human-written and LLM-rewritten constructive versions of these comments. Our analysis shows that LLM systematically diminishes Conservative values while elevating prosocial values such as Benevolence and Universalism. When these comments were read by others, participants opposing same-sex marriage or Islam found human-written comments more aligned with their values, whereas those supportive of these communities found LLM-rewritten versions more aligned with their values. These findings suggest that value homogenization in LLM-mediated prosocial discourse runs the risk of marginalizing conservative viewpoints on value-laden topics and may inadvertently shape the dynamics of online discourse.

This paper presents a qualitative study with 17 participants that uses video elicitations to investigate how conversational AI agents driven by large language models might support “shared care,” or coordination of home-based care among family caregivers (FCs) and home care workers (HCWs) who care for the same care recipient (CR). Participants saw conversational AI as a promising tool that might help streamline communication, coordinate shift handovers, bridge language gaps, and support onboarding of new or substitute caregivers. That said, caregivers assumed AI agents would inevitably make mistakes and should thus be designed to signal uncertainty and make it easy to report errors. More broadly, participants discussed how AI agents designed for sensitive home care contexts will need to explicitly preserve the human essence of care, minimize extra data work that might distract from caregiving, and always complement—not replace—human judgment.

This study investigates Shiksha Copilot, an AI-assisted lesson planning tool deployed in government schools across Karnataka, India. The system combined LLMs and human expertise through a structured process in which English and Kannada lesson plans were co-created by curators and AI; teachers then further customized these curated plans for their classrooms using their own expertise alongside AI support. Drawing on a large-scale mixed-methods study involving 1,043 teachers and 23 curators, we examine how educators collaborate with AI to generate context-sensitive lesson plans, assess the quality of AI-generated content, and analyze shifts in teaching practices within multilingual, low-resource environments. Our findings show that teachers used Shiksha Copilot both to meet administrative documentation needs and to support their teaching. The tool eased bureaucratic workload, reduced lesson planning time, and lowered teaching-related stress, while promoting a shift toward activity-based pedagogy. However, systemic challenges such as staffing shortages and administrative demands constrained broader pedagogical change. We frame these findings through the lenses of teacher-AI collaboration and communities of practice to examine the effective integration of AI tools in teaching. Finally, we propose design directions for future teacher-centered EdTech, particularly in multilingual and Global South contexts.

Collaboration and community building are crucial for equitable participation in research, especially in disability and accessibility research fields.Engaging with communities of disabled individuals helps understand their unique challenges, expectations, and needs, which should be reflected in assistive technology design.However, a gap exists between research in resource-rich societies and the practices of low-resource settings, where socio-economic and infrastructural constraints are often overlooked.Conversely, the innovation and sustainable solutions in these settings are often

Disabled people on social media often experience ableist hate and microaggressions. Prior work has shown that platform moderation often fails to remove ableist hate leaving disabled users exposed to harmful content. This paper examines how personalized moderation can safeguard users from viewing ableist comments. During interviews and focus groups with 23 disabled social media users, we presented design probes to elicit perceptions on configuring their filters of ableist speech (e.g. intensity of ableism and types of ableism) and customizing the presentation of the ableist speech to mitigate the harm (e.g. AI rephrasing the comment and content warnings). We found that participants preferred configuring their filters through types of ableist speech and favored content warnings. We surface participants distrust in AI-based moderation, skepticism in AI's accuracy, and varied tolerances in viewing ableist hate. Finally we share design recommendations to support users' agency, mitigate harm from hate, and promote safety.

This paper explores opportunities and challenges for data-driven advocacy to support home care workers, an often overlooked group of low-wage, frontline health workers.First, we investigate what data to collect and how to collect it in ways that preserve privacy and avoid burdening workers.Second, we examine how workers and advocates could use collected data to strengthen individual and collective advocacy efforts.Our qualitative study with 11 workers and 15 advocates highlights tensions between workers' desires for individual and immediate benefits and advocates' preferences to prioritize more collective and long-term benefits.We also uncover discrepancies between participants' expectations for how data might transform advocacy and their on-the-ground experiences collecting and using real data.Finally, we discuss future directions for data-driven worker advocacy, including combining different kinds of data to ameliorate challenges, leveraging advocates as data stewards, and accounting for workers' and organizations' heterogeneous goals.

We present a qualitative study that investigates the implications of current and near-future AI deployment for home care workers (HCWs), an overlooked group of frontline healthcare workers. Through interviews with 22 HCWs, care agency staff, and worker advocates, we find that HCWs do not understand how AI works, how their data can be used, or why AI systems might retain their information. HCWs are unaware that AI is already being utilized in their work, primarily via algorithmic shift-matching systems adopted by agencies. Participants detail the risks AI poses in sensitive care settings for HCWs, patients, and agencies, including threats to workers' autonomy and livelihoods, and express concerns that workers will be held accountable for AI mistakes, with the burden of proving AI's decisions incorrect falling on them. Considering these risks, participants advocate for new regulations and democratic governance structures that protect workers and control AI deployment in home care work.

Large Language Models (LLMs) and Text-To-Image (T2I) models have demonstrated the ability to generate compelling text and visual stories. However, their outputs predominantly reflect the sensibilities of Western ideologies, often resulting in an outsider’s gaze on other cultures. As a result, non-Western communities have to put extra effort into generating culturally specific stories. To address this challenge, we developed a visual storytelling tool called Kahani that generates culturally grounded visual stories for non-Western cultures. Our tool leverages off-the-shelf models GPT-4 Turbo and Stable Diffusion XL (SDXL). By using Chain of Thought (CoT) and T2I prompting techniques, we capture the cultural context from user’s prompt and generate vivid descriptions of the characters and scene compositions. To evaluate the effectiveness of Kahani, we conducted a comparative user study with ChatGPT-4 (with DALL-E3) in which participants from different regions of India compared the cultural relevance of stories generated by the two tools. The results of the qualitative and quantitative analysis performed in the user study show that Kahani’s visual stories are more culturally nuanced than those generated by ChatGPT-4. In 27 out of 36 comparisons, Kahani outperformed or was on par with ChatGPT-4, effectively capturing cultural nuances and incorporating more Culturally Specific Items (CSI), validating its ability to generate culturally grounded visual stories.

This paper explores the nature and spread of viral WhatsApp content among everyday users in three diverse countries: India, Indonesia, and Colombia. By analyzing hundreds of viral messages collected with participants’ consent from private WhatsApp groups, we provide one of the first cross-cultural categorizations of viral content on WhatsApp. Despite the differences in cultural and geographic settings, our findings reveal striking similarities in the types of groups users engage with and the viral content they receive, particularly in the prevalence of misinformation. Our comparative analysis shows that viral content often includes political and religious narratives, with misinformation frequently recirculated despite prior debunking by fact-checking organizations. These parallels suggest that closed messaging platforms like WhatsApp facilitate similar patterns of information dissemination across different cultural contexts. This work contributes to the broader understanding of global digital communication ecosystems and provides a foundation for future research on information flow and moderation strategies in private messaging platforms.

WhatsApp is the largest social media platform in the Global South and is a virulent force in global misinformation and political propaganda. Due to end-to-end encryption WhatsApp can barely review any content and mostly rely on volunteer moderation by group admins. Yet, little is known about how WhatsApp group admins manage their groups, what factors and values influence moderation decisions, and what challenges they face while managing their groups. To fill this gap, we interviewed admins of 32 diverse groups and reviewed content from 30 public groups in India and Bangladesh. We observed notable differences in the formation, members' behavior, and moderation of public versus private groups, as well as in how WhatsApp admins operate compared to those on other platforms. We used Baumrind's typology of 'parenting styles' as a lens to examine how admins enact care and control during volunteer moderation. We identified four styles based on how caring and controlling the admins are and discuss design recommendations to help them better manage problematic content in WhatsApp groups.

This paper examines if large language models (LLMs) can help people write constructive comments on divisive social issues due to the difficulty of expressing constructive disagreement online. Through controlled experiments with 600 participants from India and the US, who reviewed and wrote constructive comments on threads related to Islamophobia and homophobia, we observed potential misalignment between how LLMs and humans perceive constructiveness in online comments. While the LLM was more likely to prioritize politeness and balance among contrasting viewpoints when evaluating constructiveness, participants emphasized logic and facts more than the LLM did. Despite these differences, participants rated both LLM-generated and human-AI co-written comments as significantly more constructive than those written independently by humans. Our analysis also revealed that LLM-generated comments integrated significantly more linguistic features of constructiveness compared to human-written comments. When participants used LLMs to refine their comments, the resulting comments were more constructive, more positive, less toxic, and retained the original intent. However, LLMs often distorted people's original views--especially when their stances were on a spectrum instead of being outright polarizing. Based on these findings, we discuss ethical and design considerations in using LLMs to facilitate constructive discourse online.

Home care workers (HCWs) are an important group of frontline workers that deliver essential at-home care services to enable older adults to age in place. Despite their importance in patient care, research has shown that HCWs are an overlooked and undervalued workforce: HCWs work in isolated conditions, are paid low wages, experience high levels of stress and burnout, and more. As a result, despite being motivated to try and be healthy, this essential workforce suffers from poor physical and mental health outcomes. This paper combines data from focus groups, interviews, and a month-long field study with HCWs to investigate the feasibility and utility of using activity tracking devices to provide HCWs with fine-grained awareness and insights into daily activities that affect their health and wellbeing. We explore HCWs' reactions to both their individual and collective data, discussing their efforts towards positive behavior change, but also highlighting systemic and occupational factors that may limit HCWs' agency and control over their own activities. Finally, we discuss the potential for HCWs' collective data to raise awareness about their working conditions and provide data-driven evidence to aid advocacy efforts towards improved policies, better wages, or greater protections for this vital workforce.

Content creators (e.g., gamers, activists, vloggers) with marginalized identities are at-risk of experiencing hate and harassment. This paper examines the ableist hate and harassment that disabled content creators experience on social media. Through surveys (N=50) and interviews (N=20) with disabled creators, we developed a taxonomy of 11 types of ableist hate and harassment (e.g., eugenics-related speech, denial and stigmatization of accessibility) and outlined how ableism harms creators’ well-being and content creation practices. Using statistical modeling, we investigated differences in ableist experiences given creators’ intersecting identities such as race and sexuality. We found that LGBTQ disabled creators face significantly more ableist hate compared to non-LGBTQ disabled creators. Lastly, we discuss our findings through an infrastructure lens to highlight how disabled creators experience platform-enabled ableism, undergo labor to cope with hate, and develop strategies to safeguard against future hate.

People with disabilities experience high levels of social discrimination worldwide. But, these harms are more pronounced in the Global South due to the intense stigma around disability and its intersections with structural embeddings of patriarchy. The massive growth of social media in the Global South provides people with disabilities a unique opportunity to advocate for disability rights and challenge regressive ableist norms. Yet, little is known about the challenges they face in doing their advocacy work on social media. Through interviews with 20 disability advocates in India with diverse gender identities and abilities, we found that disability advocates routinely face ableist hate and harassment, patronizing and invalidating comments, and lack of visibility and support, which forces them to self-censor as a form of self-protection, leading to low advocacy outcomes. We draw on these findings to illuminate the role of social media in the invisibilization of people with disabilities in the online sphere.

AI technologies are increasingly deployed to support community health workers (CHWs) in high-stakes healthcare settings, from malnutrition diagnosis to diabetic retinopathy. Yet, little is known about how such technologies are understood by CHWs with low digital literacy and what can be done to make AI more understandable for them. This paper examines the potential of explorable explanations in improving AI understanding for CHWs in rural India. Explorable explanations integrate visual heuristics and written explanations to promote active learning. We conducted semi-structured interviews with CHWs who interacted with a design probe in which AI predictions of child malnutrition were accompanied by explorable explanations. Our findings show that explorable explanations shift CHWs’ AI-related folk theories, help develop a more nuanced understanding of AI, augment CHWs’ learning and occupational capabilities, and enhance their ability to contest AI decisions. We also uncover the effects of CHWs’ sociopolitical environments on AI understanding and argue for a more holistic conception of AI explainability that goes beyond cognition and literacy.

This paper presents Saharaline, an intervention designed to provide collective social support for teachers in low-income schools. Implemented as a WhatsApp-based helpline, Saharaline enables teachers to reach out for personalized, long-term assistance with a wide range of problems and stressors, including pedagogical, emotional, and technological challenges. Depending on the support needed, teachers’ requests are routed to appropriate domain experts— staff employed by educational non-profit organizations who understand teachers’ on-the-ground realities—who offer localized and contextualized assistance. Via a three-month exploratory deployment with 28 teachers in India, we show how Saharaline’s design enabled a collective of diverse education experts to craft and deliver localized solutions that teachers could incorporate into their practice. We conclude by reflecting on the efficacy of our intervention in low-resource work contexts and provide recommendations to enhance collective social support interventions similar to Saharaline.

WhatsApp groups have become a hotbed for the propagation of harmful content including misinformation, hate speech, polarizing content, and rumors, especially in Global South countries. Given the platform's end-to-end encryption, moderation responsibilities lie on group admins and members, who rarely contest such content. Another approach is fact-checking, which is unscalable, and can only contest factual content (e.g., misinformation) but not subjective content (e.g., hate speech). Drawing on recent literature, we explore deliberation---open and inclusive discussion---as an alternative. We investigate the role of a conversational agent in facilitating deliberation on harmful content in WhatsApp groups. We conducted semi-structured interviews with 21 Indian WhatsApp users, employing a design probe to showcase an example agent. Participants expressed the need for anonymity and recommended AI assistance to reduce the effort required in deliberation. They appreciated the agent's neutrality but pointed out the futility of deliberation in echo chamber groups. Our findings highlight design tensions for such an agent, including privacy versus group dynamics and freedom of speech in private spaces. We discuss the efficacy of deliberation using deliberative theory as a lens, compare deliberation with moderation and fact-checking, and provide design recommendations for future such systems. Ultimately, this work advances CSCW by offering insights into designing deliberative systems for combating harmful content in private group chats on social media.

Home care workers (HCWs) are professionals who provide care to older adults and people with disabilities at home. However, HCWs are vulnerable and especially susceptible to wage theft, or not being paid their legally-entitled wages in full by their employers. Prior work has examined other low-wage work settings to show how technology is designed and deployed has the potential to both cause and address wage theft. We extend this work by examining the relationship between technology and wage theft in the home care context. We collaborated closely with a local grassroots organization to conduct interviews with workers and labor, legal, and payroll experts. We uncovered how the complex, volatile, and diverse nature of home care complicates the errors in time-tracking systems. Through design provocations and focus groups with workers and experts, we also investigated the potential of technology as a part of broader efforts to curb wage theft through educating and empowering isolated HCWs. While we found that approachable design could reduce errors in existing systems, make employer processes more transparent, and help workers exchange knowledge to build collective power, we also discuss concerns around burden, privacy, and accountability when designing technologies for HCWs and other low-wage workers.

AI-driven tools are increasingly deployed to support low-skilled community health workers (CHWs) in hard-to-reach communities in the Global South. This paper examines how CHWs in rural India engage with and perceive AI explanations and how we might design explainable AI (XAI) interfaces that are more understandable to them. We conducted semi-structured interviews with CHWs who interacted with a design probe to predict neonatal jaundice in which AI recommendations are accompanied by explanations. We (1) identify how CHWs interpreted AI predictions and the associated explanations, (2) unpack the benefits and pitfalls they perceived of the explanations, and (3) detail how different design elements of the explanations impacted their AI understanding. Our findings demonstrate that while CHWs struggled to understand the AI explanations, they nevertheless expressed a strong preference for the explanations to be integrated into AI-driven tools and perceived several benefits of the explanations, such as helping CHWs learn new skills and improved patient trust in AI tools and in CHWs. We conclude by discussing what elements of AI need to be made explainable to novice AI users like CHWs and outline concrete design recommendations to improve the utility of XAI for novice AI users in non-Western contexts.

Social media platforms have witnessed an unprecedented growth in users from rural communities in India. Many of these users are new to online information environments and are highly susceptible to misinformation. Fact-checking has the potential to reduce the proliferation and impact of misinformation; however, little is known about how fact-checking organizations in India serve rural users. To fill this gap, we conducted interviews with 12 prominent fact-checking organizations in India to understand their current practices and challenges in providing their services to rural users and the associated human and technological infrastructure they use. We discovered several measures that fact-checking organizations take to increase the reach, awareness, and relevance of fact-checked content for rural users, such as engaging with stringer networks and utilizing vernacular languages. However, fact-checking organizations also face severe challenges that limit both the scale of their work and engagement from rural users. Drawing on these findings, we provide design and policy recommendations to improve the reach, awareness, and relevance of fact-checked content for social media users in rural areas.

Social media platforms use content moderation to reduce and remove problematic content. However, much of the discourse on the benefits and pitfalls of moderation has so far focused on users in the West. Little is known about how users in the Global South interact with the humans and algorithms behind opaque moderation systems. To fill this gap, we conducted interviews with 19 Bangladeshi social media users who received restrictions for violating community standards on Facebook. We found that the users perceived the underlying human-AI infrastructure to imbibe coloniality in the form of amplifying power relations, centering Western norms, and perpetuating historical injustices and erasure of minoritized expressions. Based on the findings, we establish that the current moderation systems often propagate historical power relations and patterns of oppression, and discuss ways to rethink moderation in a fundamentally decolonial way.

Home health aides are paid professionals who provide long-term care to an expanding population of adults who need it. However, aides' work is often unrecognized by the broader caregiving team despite being in demand and crucial to care---an invisibility reinforced by ill-suited technological tools. In order to understand the invisible work aides perform and its relationship to technology design, we interviewed 13 aides employed by home care agencies in New York City. These aides shared examples that demonstrated the intertwined nature of both types of invisible work (i.e., emotions- and systems-based) and expanded the sociological mechanisms of invisibility (i.e., sociocultural, sociolegal, sociospatial) to include the sociotechnical. Through these findings, we investigate the opportunities, tensions, and challenges that could inform the design of tools created for these important, but often overlooked, frontline caregivers.

Ableist microaggressions are subtle forms of discrimination that disabled people experience daily, perpetuating inequalities and maintaining their ongoing marginalization. Despite the importance of understanding such harms, little work has been done to examine how disabled people are discriminated against online. We address this gap by investigating how disabled people experience ableist microaggressions on social media and how they respond to and cope with these experiences. By conducting interviews with 20 participants with various disabilities, we uncover 12 archetypes of ableist microaggressions on social media, reveal participants’ coping mechanisms, and describe the long-term impact on their wellbeing and social media use. Lastly, we present design recommendations, re-evaluating how social media platforms can mitigate and prevent these harmful experiences.

We present a qualitative study with 36 diverse social media users in India to critically examine how low-resource communities engage with fake videos, including cheapfakes and AI-generated deepfakes. We find that most users are unaware of digitally manipulated fake videos and perceive videos to be fake only when they present inaccurate information. Few users who know about doctored videos expect them to be of poor quality and know nothing about sophisticated deepfakes. Moreover, most users lack the skills and willingness to spot fake videos and some were oblivious to the risks and harms of fake videos. Even when users know a video to be fake, they prefer to take no action and sometimes willingly share fake videos that favor their worldview. Drawing on our findings, we discuss design recommendations for social media platforms to curb the spread of fake videos.

Social media has witnessed an unprecedented growth in users based in low-income communities in the Global South. However, much remains unknown about the drivers of misinformation in such communities. To fill this gap, we conducted an interview-based study to examine how rural and urban communities in India engage with misinformation on WhatsApp. We found that misinformation led to bitterness and conflict – rural users who had higher social status heavily influenced the perceptions and engagement of marginalized members. While urban users relied on the expertise of gatekeepers for verification, rural users engaged in collective deliberations in offline spaces. Both rural and urban users knowingly forwarded misinformation. However, rural users propagated hyperlocal misinformation, whereas urban users forwarded misinformation to reduce their efforts to assess information credibility. Using a public sphere lens, we propose that the reactions to misinformation provide a view of Indian society and its schisms around class, urbanity, and social interactions.

Women’s economic empowerment is central to gender equality. However, work opportunities available to low-income women in patriarchal societies are infrequent. While crowd work has the potential to increase labor participation of such women, much remains unknown about their engagement with crowd work and the resultant opportunities and tensions. To fill this gap, we critically examined the adoption and use of a crowd work platform by low-income women in India. Through a qualitative study, we found that women faced tremendous challenges, for example, in seeking permission from family members to do crowd work, lack of family support and encouragement, and often working in unfavorable environments where they had to hide their work lives. While crowd work took a toll on their physical and emotional wellbeing, it also led to increased confidence, agency, and autonomy. We discuss ways to reduce frictions and tensions in participation of low-income women on crowd work platforms.

As community-driven organizations sought to support their constituents through the COVID-19 crisis, many drew on digital volunteers to expand their capacity and reach. However, coordinating the efforts of virtual volunteers is a challenging task with few empirical studies of the associated risks and best practices. In this paper, we report on the activities of CGNet Swara, a citizen journalism platform that published 401 distress calls from vulnerable communities stranded in India due to the imposition of a nationwide lockdown. CGNet mobilized 11 digital volunteers to help these contributors over a period of nearly 2 months. We found that a lack of proper guidance to digital volunteers and outdated organizational policies resulted in demonstrable harms to vulnerable communities. We discuss risks that are inherent in collaborations between organizations extending themselves to crisis response and emergent groups of digital volunteers, and how they can be mitigated by real-time monitoring and development of standard operating procedures relating to impact metrics, verification standards and disclosure policies.

Frontline health workers provide essential services for their communities, but much of their work remains invisible—undervalued and underappreciated. Examining this invisible work ensures new technologies do not amplify or reinforce inequitable power structures, especially as governments and organizations push to digitize health work processes. We build on a burgeoning conversation by studying how invisible work manifests and how this invisibility can be challenged in two contexts of frontline health: home health aides in New York City, USA and Accredited Social Health Activists (ASHAs) in Uttar Pradesh, India. We highlight three shared manifestations of invisible work: (1) work done outside of the workers’ boundaries (2) work done to gain and share knowledge and (3) work done to manage relationships. These common categories are experienced differently in the two contexts, raising nuances to consider when designing technology for frontline health workers. We discuss these nuances and other tensions through concrete examples of how workers can escalate feedback and conflicts, quantify implicit expertise about patients, or build more awareness of their situation. Our paper guides the creation of technologies that take into account a more comprehensive understanding of the frontline health workers’ processes and highlight more of their contributions.

Artificial intelligence (AI) and machine learning (ML) are quickly becoming pervasive in ways that impact the lives of all humans across the globe. In an effort to make otherwise ”black box” AI/ML systems more understandable, the field of Explainable AI (XAI) has arisen with the goal of developing algorithms, toolkits, frameworks, and other techniques that enable people to comprehend, trust, and manage AI systems. However, although XAI is a rapidly growing area of research, most of the work has focused on contexts in the Global North, and little is known about if or how XAI techniques have been designed, deployed, or tested with communities in the Global South. This gap is concerning, especially in light of rapidly growing enthusiasm from governments, companies, and academics to use AI/ML to “solve” problems in the Global South. Our paper contributes the first systematic review of XAI research in the Global South, providing an early look at emerging work in the space. We identified 16 papers from 15 different venues that targeted a wide range of application domains. All of the papers were published in the last three years. Of the 16 papers, 13 focused on applying a technical XAI method, all of which involved the use of (at least some) data that was local to the context. However, only three papers engaged with or involved humans in the work, and only one attempted to deploy their XAI system with target users. We close by reflecting on the current state of XAI research in the Global South, discussing data and model considerations for building and deploying XAI systems in these regions, and highlighting the need for human-centered approaches to XAI in the Global South.

The COVID-19 pandemic has necessitated rapid top-down dissemination of reliable and actionable information. This presents unique challenges in engaging low-literate communities that live in poverty and lack access to the Internet. We describe the design and deployment of a voice-based social media platform, accessible over simple phones, for actively engaging such communities in Pakistan with reliable COVID information. We developed three strategies to overcome users’ hesitation, mistrust, and skepticism in engaging with COVID content. Users were: (1) encouraged to listen to reliable COVID advisory, (2) incentivized to share authentic content with others, and (3) prompted to critically think about COVID-related information behaviors. Using a mixed-methods evaluation, we show that users approached with all three strategies had a significantly higher engagement with COVID content compared to others. We discuss how new designs of social media can enable users to engage with and propagate authentic information.

This paper presents a between-subjects design experiment with 478 people in India to investigate how rural and urban social media users perceive credible and fake posts, and how different types of sources impact their perceptions of information credibility and sharing behaviors. Our findings reveal that: (1) rural social media users were less adept in differentiating between credible and fake posts than their urban counterparts, and (2) source effects on trust and sharing intent manifested differently for urban and rural users. For example, fake posts from family members garnered greater trust among urban users but were trusted the least by rural users. In case of sharing Facebook posts, urban users were more willing to share fake posts from family, whereas, rural users were more inclined to share fake posts from journalists. Drawing on these findings, we propose design interventions to counteract fake news in low-resource environments of the Global South.

Response bias has been framed as the tendency of a participant's response to be skewed by a variety of factors, including study design and participant-researcher dynamics. Response bias is a concern for all researchers who conduct studies with people — especially those working with participants with disabilities. This is because these participants’ diverse needs require methodological adjustments and differences in disability identity between the researcher and participant influence power dynamics. Despite its relevance, there is little literature that connects response bias to accessibility. We conducted semi-structured interviews with 27 accessibility researchers on how response bias manifested in their research and how they mitigated it. We present unique instances of response bias and how it is handled in accessibility research; insights into how response bias interacts with other biases like researcher or sampling bias; and philosophies and tensions around response bias such as whether to accept or address it. We conclude with guidelines on thinking about response bias in accessibility research.

Most children who are blind live in low-resource settings and attend schools that have poor technical infrastructure, overburdened teachers, and outdated curriculum. Our work explores the role digital games can play to develop digital skills of such children in the early grades. Recognizing the critical role of teachers in introducing children to technology, we conducted a mixed-methods study to examine which attributes of digital games teachers find useful for children and what challenges they perceive in integrating digital games in schools for the blind. Our findings indicate that teachers prefer games that align well with curriculum objectives, promote learning, improve soft skills, and increase engagement with computers. Despite being overburdened and lacking technological support, teachers expressed strong enthusiasm to integrate these games in school curriculum and schedule. We conclude by discussing design implications for designers of accessible games in low-resource settings.

Recent advances in Artificial Intelligence (AI) suggest that AI applications could transform healthcare delivery in the Global South. However, as researchers and technology companies rush to develop AI applications that aid the health of marginalized communities, it is critical to consider the needs and perceptions of the community health workers (CHWs) who will have to integrate these AI applications into the essential healthcare services they provide to rural communities. We describe a qualitative study examining CHWs’ perceptions of an AI application for automated disease diagnosis. Drawing on data from 21 interviews with CHWs in rural India, we characterize (1) CHWs’ knowledge, perceptions, and understandings of AI; and (2) the benefits and challenges that CHWs anticipate as AI applications are integrated into their workflows, including their opinions on automation of their work, possible misdiagnosis and errors, data access and surveillance issues, security and privacy challenges, and questions concerning trust. We conclude by discussing the implications of our work for HCI and AI research in low-resource environments.

Individuals in distress adopt varied pathways in pursuit of care that aligns with their individual needs. Prior work has established that the first resource an individual leverages can influence later care and recovery, but less is understood about how the design of a point of care might interact with subsequent pathways to care. We investigate how the design of the Indian mental health helpline system interacts with complex sociocultural factors to marginalize caller needs. We draw on interviews with 18 helpline stakeholders, including individuals who have engaged with helplines in the past, shedding light on how they navigate both technological and structural barriers in pursuit of relief. Finally, we use a design justice framework rooted in Amartya Sen’s conceptualization of realization-focused justice to discuss implications and present recommendations towards the design of technology-mediated points of mental health support.

Although WhatsApp-based communication is playing an increasingly large role in the professional lives of teachers in low-income schools, the nature of the interactions that occur and how these interactions enable cooperative work are not well understood. We contribute a qualitative analysis of 26 existing WhatsApp groups (35,567 messages) that examines (1) the strategies used to encourage interaction within teacher-focused WhatsApp groups, and (2) how these interactions are sustained by teachers, management, and organizations over a period of time. We use teacher networks and activity awareness model to make sense of WhatsApp-based interactions and show how WhatsApp narrows the gap between management and teachers, leading to additional work and stress for teachers. WhatsApp was also used to circulate polarizing and malicious information, leading to a variety of interesting content moderation strategies. Our findings expand the scope of research on teacher networks to low-income contexts and will inform future interventions that enable cooperative work among teachers.

Smartphones play an increasingly large role in the professional lives of teachers in low-income contexts, creating an urgent need to better understand the role of technology-related stress (technostress) in teachers' smartphone use for work. We contribute a mixed methods study analyzing the impact of smartphone use on teachers' work lives in low-income Indian schools. Findings from 70 interviews and 1,361 survey responses suggest that although smartphones aid teaching and administrative functions, smartphone use also significantly predicts burnout among teachers, with technostress providing a major explanation for this relationship. We reveal how teachers' work is constantly surveilled and monitored via technology and how teachers' personal smartphones were controlled and repurposed through socio-technical structures by the higher management to serve management's goals, substantially increasing the work teachers were required to perform outside of work hours. Our work extends technostress research to HCI4D contexts and highlights the need to develop better support structures for teachers and rethink how smartphones are used in their work.

Current ICTD best practices call on practitioner organizations that deploy technology interventions with marginalized communities to commit to long-term engagements and provide continuous support. This paper describes a qualitative study that examines the challenges and issues that arise for organizations that have tried to answer this call: a set of education-focused non-profit organizations in India that have invested heavily in building long-term relationships with low-income schools and that are deeply committed to providing ongoing support, both in-person and via technology. Interviews and observations with 71 participants (51 from eight organizations, and 15 teachers and five principals from 12 schools) reveal (1) the challenges and issues that arise as organizations integrate smartphones into teacher support programs, and (2) the strategies that teachers and organizations use to ease smartphone adoption in teacher support programs. Our findings uncover the effects of organizations' smartphone-oriented support programs on teachers' workloads, efficacy, and stress, and highlight opportunities for organizations to improve their programs and support structures.

Social media platforms are widely used by people to report, access, and share information during outbreaks and epidemics. Although government agencies and healthcare institutions in developed regions are increasingly relying on social media to develop epidemic forecasts and outbreak response, there is a limited understanding of how people in developing regions interact on social media during outbreaks and what useful insights this dataset could offer during public health crises. In this work, we examined 28,688 tweets to identify public health issues during dengue epidemic in Bangladesh and found several insights, such as irregularities in dengue diagnosis and treatment, shortage of blood supply for Rh negative blood groups, and high local transmission of dengue during Eid-ul-Adha, that impact disease preparedness and outbreak response. We discuss the opportunities and challenges in analyzing tweets and outline how government agencies and healthcare institutions can use social media health data to inform policy making during public health crises.

Although voice forums are widely used to enable marginalized communities to produce, consume, and share information, their financial sustainability is a key concern among HCI4D researchers and practitioners. We present ReCall, a crowdsourcing marketplace accessible via phone calls where low-income rural residents vocally transcribe audio files to gain free airtime to participate in voice forums as well as to earn money. We conducted a series of experimental and usability evaluations with 28 low-income people in rural India to examine the effect of phone types, channel types, and review modes on speech transcription performance. We then deployed ReCall for two weeks to 24 low-income rural residents who placed 5,879 phone calls, completed 29,000 micro tasks to yield transcriptions with 85% accuracy, and earned INR 20,500. Our mixed-methods analysis indicates that each minute of crowd work on ReCall gives users eight minutes of free airtime on another voice forum, and thus illustrates a way to address the financial sustainability of voice forums.

HCI4D researchers and practitioners have leveraged voice forums to enable people with literacy, socioeconomic, and connectivity barriers to access, report, and share information. Although voice forums have received impassioned usage from low-income, low-literate, rural, tribal, and disabled communities in diverse HCI4D contexts, the participation of women in these services is almost non-existent. In this paper, we investigate the reasons for the low participation of women in social media voice forums by examining the use of Sangeet Swara in India and Baang in Pakistan by marginalized women and men. Our mixed-methods approach spanning content analysis of audio posts, quantitative analysis of interactions between users, and qualitative interviews with users indicate gender inequity due to deep-rooted patriarchal values. We found that women on these forums faced systemic discrimination and encountered abusive content, flirts, threats, and harassment. We discuss design recommendations to create social media voice forums that foster gender equity in use of these services.

The developing world has increasingly relied on data driven policies. Numerous development agencies have pushed for on-ground data collection to support the development work they pursue. Many governments have launched their own efforts for frequent information gathering. Overall, the amount of data collected is tremendous, yet there are significant issues in doing useful analysis. Most of these barriers manifest in data cleaning and merging, and require a data engineer to support some parts of the analysis. In this paper, we investigate the challenges of cleaning development data through an interview based study. We conducted face to face interviews of 13 stakeholders, eight from international development organizations and five government workers from Pakistan, including both managers and data analysts. From analysis of the interviews we identified common challenges faced in processing development data including correcting open text fields, merging hierarchical data, and extracting data from textual formats such as PDF. We construct a basic taxonomy of data cleaning challenges, and identify areas where support tools can improve the process. Ultimately, the objective is to empower regular data users to easily do the necessary data cleaning and scrubbing for analysis.

Many countries across the globe are engaged in efforts to promote a cashless society. In India, there is a strong top-down push by the government and the private sector for mobile payments. In this work, we examine the benefits and pitfalls people perceive in using mobile payments for customer-merchant transactions. Through interviews with 19 customers and 15 merchants across rural, periurban, and urban India, we explore people's awareness about mobile payment systems, perceived trade-offs among different payment methods, and the security, privacy, and utility barriers in its adoption and use. Overall, we found that customers were interested in adopting mobile payments for referral rewards and sign-up incentives, but were hesitant to regularly use them, whereas merchants saw mobile payments as an unnecessary burden for their business. We discuss the nuanced views of customers and merchants, and offer recommendations to address the barriers in the use of mobile payments for customer-to-merchant transactions.

Agricultural market information (MI) services provide smallholder farmers with convenient access to price information and have demonstrated potential to improve their incomes. Some recent evaluations of MI systems, however, have shown disappointing results and brought forth many complicating factors. Cautious of the mixed literature, we investigate the potential effectiveness and likely limitations of an MI service for improving livelihoods of smallholder farmers in the context of Loop, a shared transport-to-market-service for farmers. We conducted interviews with 17 farmers and 3 commission agents in Buxar, Bihar (India). Consistent with "information scarcity" and "information asymmetry" theories, we report how many farmers in this area regularly use mobile phones to check prices for choosing markets and negotiating transactions. Participants reported increases in the numbers of traders and price stabilization since the arrival of mobile phones. However, we found many other diverse factors that often outweigh the importance of market prices and inhibit market access, including market capacity, time, unfamiliarity with new markets, personal relationships, attitudes towards risk, credit relationships, and physical danger. Finally, to probe which of these additional factors might be addressable using an MI service, we present exploratory findings from preliminary user-interface studies.

We contribute to the intersection of multilingualism and human-computer interaction (HCI) with our investigation of language preferences in the context of the interface design of interactive systems. Through interview data collected from avid smartphone users located across distinct user groups in India, none of whom were native English speakers, we examine the factors that shape language choice and use on their mobile devices. Our findings indicate that these users frequently engage in English communication proactively and enthusiastically, despite their lack of English fluency, and we detail their motivations for doing so. We then discuss how language in technology use can be a way of putting forth mobility as an aspect of one's identity, making the case for an intersectional approach to studying language in HCI.

Evaluations of technological artifacts in HCI4D contexts are known to suffer from high levels of participant response bias---where participants only provide positive feedback that they think will please the researcher. This paper describes a practical, low-cost intervention that uses the concept of social proof to influence participant response bias and successfully elicit critical feedback from study participants. We subtly exposed participants to feedback that they perceived to be provided by people 'like them', and experimentally controlled the tone and content of the feedback to provide either positive, negative, or no social proof. We then measured how participants' quantitative and qualitative evaluations of an HCI artifact changed based on the feedback to which they were exposed. We conducted two controlled experiments: an online experiment with 245 MTurk workers and a field experiment with 63 women in rural India. Our findings reveal significant differences between participants in the positive, negative, and no social proof conditions, both online and in the field. Participants in the negative condition provided lower ratings and a greater amount of critical feedback, while participants in the positive condition provided higher ratings and a greater amount of positive feedback. Taken together, our findings demonstrate that social proof is a practical and generalizable technique that could be used by HCI researchers to influence participant response bias in a wide range of contexts and domains.

BSpeak is an accessible crowdsourcing marketplace that enables blind people in developing regions to earn money by transcribing audio files through speech. We examine accessibility and usability barriers that 15 first-time users, who are low-income and blind, experienced while completing transcription tasks on BSpeak and Mechanical Turk (MTurk). Our mixed-methods analysis revealed severe accessibility barriers in MTurk due to the absence of landmarks, unlabeled UI elements, and improper use of HTML headings. Compared to MTurk, participants found BSpeak significantly more accessible and usable, and completed tasks with higher accuracy in lesser time due to its voice-based implementation. In a two-week field deployment of BSpeak in India, 24 low-income blind users earned rupee 7,310 by completing over 16,000 transcription tasks to yield transcriptions with 87% accuracy. Through our analysis of BSpeak's strengths and weaknesses, we provide recommendations for designing crowdsourcing marketplaces for low-income blind people in resource-constrained settings.

Prior research suggests that security and privacy needs of users in developing regions are different than those in developed regions. To better understand the underlying differentiating factors, we conducted a systematic review of Human-Computer Interaction for Development and Security & Privacy publications in 15 proceedings, such as CHI, SOUPS, ICTD, and DEV, from the past ten years. Through an in-depth analysis of 114 publications that discuss security and privacy needs of people in developing regions, we identified five key factors---culture, knowledge gaps, unintended technology use, context, and usability and cost considerations---that shape security and privacy preferences of people in developing regions. We discuss how these factors influence their security and privacy considerations using case studies on phone sharing and surveillance. We then present a set of design recommendations and research directions for addressing security and privacy needs of people in resource-constrained settings.

We present a mixed-methods study that compares and contrasts two distinct localization approaches for imparting video-based health education to rural Indian populations. Prior research efforts in the Human-Computer Interaction for Development (HCI4D) community have emphasized the value of localization of educational content. Our research uses a combination of focus groups, interviews, and knowledge retention tests to deconstruct what localization means and how it might be incorporated. Our findings highlight how characteristics of localized videos shape the participants' viewing experiences. For development-focused organizations that leverage video-based learning, our study provides insights regarding the costs and benefits of distinct localization approaches, and offers design recommendations for producing videos that yield greater traction and higher information absorption.

Speech transcription is an expensive service with high turnaround time for audio files containing languages spoken in developing countries and regional accents of well-represented languages. We present Respeak - a voice-based, crowd-powered system that capitalizes on the strengths of crowdsourcing and automatic speech recognition (instead of typing) to transcribe such audio files. We created Respeak and optimized its design through a series of cognitive experiments. We deployed it with 25 university students in India who completed 5464 micro-transcription tasks, transcribing 55 minutes of widely-varied audio content, and collectively earning USD 46 as mobile airtime. The Respeak engine aligned the transcript generated by five randomly selected users to transcribe Hindi and Indian English audio files with a word error rate (WER) of 8.6% and 15.2%, respectively. The cost of speech transcription was USD 0.83 per minute with a turnaround time of 39.8 hours, substantially less than industry standards. Using a mixed-methods analysis of cognitive experiments, system performance and qualitative interviews, we evaluate Respeak's design, user experience, strengths, and weaknesses. Our findings suggest that Respeak improves the quality of speech transcription while enhancing the earning potential of low-income populations in resource-constrained settings.

We examine the dissemination of mobile phone videos in the context of Projecting Health - a community health project in rural India. Our research objective was to identify the most effective means of promoting the distribution of health videos on a largely offline network of mobile phones in a resource-constrained environment. We compared three different distribution channels: mobile shop owners, laptop owners, and community health workers in a fourteen-week intervention that relied on data collected via missed calls from viewers and callbacks made to them. We present the design of our experiment, describe the challenges in deploying this experiment, and discuss overall findings. All three distribution channels were successful in targeting the community; mobile shops had access to most community members but the community health workers were most successful in getting videos out to those who were most interested in viewing them. Many participants were motivated to distribute the videos for the benefit of community. However, the number of missed calls received decreased over time, suggesting the exploration of alternative mechanisms to extrinsically motivate intermediaries and viewers for broader video distribution.

We present the first analysis of the use and non-use of social media platforms by low-income blind users in rural and peri-urban India. Using a mixed-methods approach of semi-structured interviews and observations, we examine the benefits received by low-income blind people from Facebook, Twitter and WhatsApp and investigate constraints that impede their social media participation. We also present a detailed analysis of how low-income blind people used a voice-based social media platform deployed in India that received significant traction from low-income people in rural and peri-urban areas. In eleven-weeks of deployment, fifty-three blind participants in our sample collectively placed 4784 voice calls, contributed 1312 voice messages, cast 33,909 votes and listened to the messages 46,090 times. Using a mixed-methods analysis of call logs, qualitative interviews, and phone surveys, we evaluate the strengths and weaknesses of the platform and benefits it offered to low-income blind people.

Interactive voice forums have emerged as a promising platform for people in developing regions to record and share audio messages using low-end mobile phones. However, one of the barriers to the scalability of voice forums is the process of screening and categorizing content, often done by a dedicated team of moderators. We present Sangeet Swara, a voice forum for songs and cultural content that relies on the community of callers to curate high-quality posts that are prioritized for playback to others. An 11-week deployment of Sangeet Swara found broad and impassioned usage, especially among visually impaired users. We also conducted a follow-up experiment, called Talent Hunt, that sought to reduce reliance on toll-free telephone lines. Together, our deployments span about 53,000 calls from 13,000 callers, who submitted 6,000 posts and 150,000 judgments of other content. Using a mixed-methods analysis of call logs, audio content, comparison with outside judges, and 204 automated phone surveys, we evaluate the user experience, the strengths and weaknesses of community moderation, financial sustainability, and the implications for future systems.

Though many researchers have studied how to incentivize people to respond to surveys, little is known about how these incentives impact respondents' willingness to recruit others to participate as well. In this paper, we show that the incentives offered for individual survey responses can have a dramatic impact on the overall reach of a survey through a network of peers. In a field experiment in India, we made a survey accessible via mobile phones and offered respondents either a fixed incentive (guaranteed payment of about $0.17) or a lottery incentive (1% chance of winning $17). When asked to choose, a significant fraction of respondents preferred the lottery incentive. However, when encouraged to spread the survey, the fixed incentive spread over 100 times further, reaching about 800 people in a day. We interpret this surprising result and discuss the implications for HCI.

In many developing countries, it is challenging for remote communities to access reliable and accurate sources of news. This paper examines the case of Somaliland, where there is a fragmented media landscape that is subject to tribal biases and manipulations. Often, exaggeration or misrepresentation of facts prompts conflicts and hostility towards authorities, especially in remote areas. To foster more transparent communications between governmental institutions and the public, we built and deployed an interactive voice forum, accessible to anyone with a basic mobile handset. Called Ila Dhageyso, this forum enables callers to listen and respond to official recordings from the Presidential Ministry, as well as to the responses of other callers. Audio messages are also threaded and posted on YouTube to involve Somaliland people in diaspora. From December 2012 to April 2013, Ila Dhageyso logged 4,300 posts from callers. This paper reviews the media climate and geopolitical circumstances in Somaliland that motivated the formation of Ila Dhageyso. It also describes the design, implementation, and experiences gained in the first exploratory deployment of the service.

With the increasing convergence of network, services, and devices, more number of end user devices are accessing digital media content which were hitherto accessible mainly from computers and television sets. This makes the digital media more prone to illegal copying and distribution. In this paper, we propose a robust digital video watermarking scheme for copyright protection using the principles of visual cryptography, scene change detection in videos, and conversion of each scene into an image by averaging the frames in the scene. In the proposed method, data is extracted and stored as one share of the watermark pattern instead of embedding data into the original video as in the conventional watermarking schemes. Since no data is embedded, the method is robust against watermark attacks aimed at detecting and destroying or distorting data embedded into videos. Simulation results show that the method is robust even in the case of 50% frame drop. All the watermarked scenes, not frames, need to be dropped for an effective attack against the watermark. Dropping of scenes makes the video less attractive for illegal copying and distribution.

Copyright protection of digital media content is becoming increasingly important with increasing potential for content monetization which is attributable to digital convergence. High volumes of uploaded content in social networking and media sharing sites makes it increasingly difficult to monitor all the uploaded content for copyright infringement, piracy and illegal distribution. Currently, digital watermarking is predominantly used for copyright protection of images and videos. Watermarking involves embedding data into an image or a video in an invisible or visible manner. This traditional method is prone to many known watermark attacks aimed at detecting and distorting or destroying the embedded data. In this paper, we present the No Mark Method, a robust, reliable, and efficient scheme for copyright protection of digital video content without embedding any data in to it. The method is based on visual cryptography and can be used to watermark individual scenes in a video and hence protects not only the copyright of complete video but also of each individual scene in the video. The method is robust against video fragmentation attack, unauthorized insertion of scenes from a protected video into another video and various other video watermarking attacks. There is no involvement of any notary agency to establish copyright ownership of a digital video.