Artificial intelligence is facing a crisis. Humans are consuming far too many resources, impeding the progress of larger, faster, newer models. It’s time to drink less, shower less, and to prioritize AI. Recent developments in AI technologies have relied on scaling technical infrastructure and capital. While improvements in benchmark performance have been put under the spotlight, environmental and social harms are obfuscated by Big Tech developing the models. Through an art-based initiative, titled Save the AI, we lean into satire and absurdity to visceralize AI’s staggering environmental and social impact. We designed posters, stickers, and postcards to be placed in public spaces to connect to people’s basic needs for resources that are fundamental to their survival and AI’s increasing demand for these same resources. We aim to spark curiosity, reflection, and conversation through our physical designs by leading people to an online networking campaign. This campaign helps people from different parts of the world connect, collectively express their concerns, and work in solidarity.
Geospatial Copilots hold immense potential for automating Earth observation (EO) and climate monitoring workflows, yet their reliance on large-scale models such as GPT-4o introduces a paradox: tools intended for sustainability studies often incur unsustainable costs. Using agentic AI frameworks in geospatial applications can amass thousands of dollars in API charges or requires expensive, powerintensive GPUs for deployment, creating barriers for researchers, policymakers, and NGOs. Unfortunately, when geospatial Copilots are deployed with open language models (OLMs), performance often degrades due to their dependence on GPT-optimized logic. In this paper, we present Geo-OLM, a tool-augmented geospatial agent that leverages the novel paradigm of state-driven LLM reasoning to decouple task progression from tool calling. By alleviating the workflow reasoning burden, our approach enables low-resource OLMs to complete geospatial tasks more effectively. When downsizing to small models below 7B parameters, Geo-OLM outperforms the strongest prior geospatial baselines by 32.8% in successful query completion rates. Our method performs comparably to proprietary models achieving results within 10% of GPT-4o, while reducing inference costs by two orders of magnitude from $500-$1000 to under $10. We present an in-depth analysis with geospatial downstream benchmarks, providing key insights to help practitioners effectively deploy OLMs for EO applications.
Although coral reefs are special and vital marine ecosystems, massive coral degradation began to occur due to the increase in global temperatures and the intensification of human industrial activities. Coral reef protection requires accurate coral recognition because it is the foundation for learning the distribution, disease, and growth of coral reefs, hereby informing the proper ways for further action. Recently, CNNs have been applied in automated coral image classification. These classifier models, however, are difficult to be generalized from the trained coral images in a marine region (source domain) to the coral images in a different marine region (target domain) since the corals have significant within-species morphological variability among the different geographic location domains. In this paper, a novel coral recognition algorithm is introduced via knowledge transfer across domains and its advantages lie in the following aspects. (1) It simultaneously transfers corals' texture and structure features across domains thus providing useful knowledge to assist the coral recognition tasks in the target marine domain. (2) To overcome the difficulty that the confusing coral images (e.g., bleached corals) are prone to be misclassified and transfer useless or even negative information, our algorithm is equipped with the reject option for the confusing corals while adapting. These corals can be sent to an expert or a more expensive but accurate system, resulting in strengthened transferability and reliability. Furthermore, we develop a new cross-domain coral image dataset to enhance coral research. Without the label information from the target marine region, our method significantly reduces the distribution gap and domain shift among the different marine regions. In addition, CR-Cross goes a step further in tackling the challenges of missing coral data, maximizing the utilization of available coral datasets, and enhancing the reusability of both coral data and coral recognition models. A series of empirical studies show that our method remarkably outperforms a broad range of baselines.
Recent studies have highlighted the challenges low-literacy users face with complex user interfaces, often preventing them from utilizing essential smartphone applications for an improved quality of life. This paper introduces the EvolveUI design approach that diverges from conventional interface design by evolving in complexity alongside a user’s growing proficiency. Initially presenting a single navigation point to simplify interaction, EvolveUI systematically expands, introducing more features and navigation points as users become more adept at interacting with the interface. We build upon previous adaptive user interface research by uniquely focusing on expanding functionality based on user proficiency, offering a tailored experience that aligns with individual learning curves. By conceptualizing the interface as a dynamically expanding hierarchy, starting as a minimalist “tree” and unfolding into a more complex structure, EvolveUI facilitates a more accessible and engaging user experience. Our study compares this evolving interface approach with conventional designs through usability tests on a mobile health application, demonstrating EvolveUI’s potential to enhance technology accessibility for low-literacy users and suggesting new directions for inclusive design practices.
The COVID-19 pandemic has threatened disproportionately rural older adults’ health and well-being as they suffer from unique social exclusion due to a lack of services, such as transportation, communication infrastructure, healthcare, and social services. Although older adults can uniquely cope with pandemic adversity compared to younger adults, less attention has been directed to investigating the coping and resilience of rural older adults. To understand how diverse coping strategies impact the resilience of rural older adults, we conducted interviews with 26 rural and small-town older adults. Older adult participants adopted different coping strategies, such as following protective measures, keeping themselves busy, providing and receiving social support, and having a positive mindset. They experienced positive changes, such as increased interpersonal connectivity. Older adults’ individual-level coping processes are influenced by their social and physical environments. We explore design opportunities to support older adults’ resilient practices and harness their skills to facilitate community resilience.
This research study proposes a social framework for birth declaration in rural Bangladesh, utilizing cloud storage technology and ad hoc models, based on interviews with Community Healthcare Workers (n=13). The traditional process of birth registration in rural areas of Bangladesh is paper-based and often subject to errors, leading to incomplete or inaccurate data. The proposed framework addresses this issue by providing a secure and efficient system with minimal hardware requirements for recording and storing birth information in resource-constrained settings. In this study, we integrate HCI principles into a cloud-based solution for rural communities, addressing infrastructure and usability concerns, in line with HCI’s focus on technology in societal contexts. Furthermore, we tackle the digital literacy gap in rural areas, designing our solution with HCI to effectively bridge this divide. We leverage existing infrastructure to introduce three progressive execution models, each constituting modular components within the overarching solution framework. The initial step in mitigating child marriage rates is securing accurate and easily retrievable birth data since paper trails are prone to manipulation, destruction, and fraud. Therefore, our study advocates for the implementation of the primary and pivotal measures to combat the vulnerable exploitation of children.
The evolving landscape of gig labor, particularly within food delivery platforms, has been influenced by a shift towards algorithmic management, emphasizing performance metrics as a daily burden to be carried through delivery work. Through a study of Indian food delivery agents, our paper suggests an emerging gamification culture between delivery platforms and workers. Agents gradually yet consistently adapt to stringent algorithmic management to overcome everyday work’s precarity. Despite the challenges of performance pressure on food delivery labor, we capture the daily rhythms and routines of delivery work punctuated by flashes of agenting, revealing small capacities to make choices and influence work outcomes. By employing a qualitative research framework, we uncover the mechanisms through which delivery platforms manipulate labor, simultaneously exploring delivery agents’ tactics to extract agency and exploit the platforms’ rules. We develop the notion of ‘agency’ to disentangle the idea of ‘gamification by agents’ as a socio-technical derivative governing food delivery agents in urban India. The findings highlight the potential of platform work empowering, albeit with limits, delivery agents with agency and decision-making authority.
Traditional localization systems often rely on a network of external sensors, making the setups cumbersome, expensive, and requiring significant calibration effort. The advent of Bluetooth 5.1 and later versions brought enhancements that enable precise localization using constant tone extension (CTE) in the signal through Angle of Arrival (AoA) and Angle of Departure (AoD) techniques. This work examines the capacity of a single Bluetooth Low-Energy (BLE) locator with an antenna array based on AoA in terms of performance, efficiency, and latency in real-time indoor positioning. While traditional neural networks train measured entities to match calculated distances, we utilize the azimuth and elevation angle components in the AoA measured and train neural networks to match their theoretical counterparts. We conducted extensive experiments in a real-world lab environment, providing ablation studies in the design. The results demonstrate the system's capability in real-time with many potential interference variables. Under lab conditions, our results show the capacity of a single locator wanes past 4m with the best average accuracy of 0.09.. error in positioning within a 5m radius to as much as similar to 1m of error beyond 6m up to the maximum possible measuring distance in the lab.
Estimating and locating occupants indoors is crucial for automating operations within buildings. However, current privacy-preserving occupancy estimation systems using thermal cameras have not been thoroughly evaluated in dense settings, such as classrooms or movie theaters, where occupants sit close to one another. Estimating occupancy in such settings presents challenges, as nearby occupants often lead to clusters of thermal signatures, thereby affecting the accuracy of the estimation. In response to these challenges, our work proposes Machine Learning (ML) and Deep Learning (DL) based methods for occupancy estimation and localization, both in dense and sparse settings. The ML method complements existing occupancy estimation techniques by incorporating new manual features, while the DL method performs automatic feature extraction, accurate occupancy estimation and localization. Remarkably, the results demonstrate a significant enhancement in occupancy estimation accuracy of up to 10%, achieving an impressive overall accuracy rate of 97%. Moreover, our evaluation on an edge device confirms the practicality and relevance of the proposed methods in real-world applications.
Effective traffic intersection control is crucial for urban sustainability. State of the art research seeking Artificial Intelligence (AI), for example Deep Reinforcement Learning (DRL) based traffic control requires environment states through various Computer Vision methods, where the collective state of multiple cameras across an intersection constitute the single state for AI. This brings in serious robustness or fault-tolerance concerns on the deployed system. Camera systems are highly susceptible to faults due to multiple possible points of failure. A single fault collapses the AI state and hence the capacity of AI controller to manage the traffic is gone. Also, infrastructure deployment and maintenance is a slow bureaucratic process in the developing countries, which makes camera faults a regular event. In the given paper, we build WebLight (https://github.com/sachin-iitd/WebLight), a web based, independent and alternative, traffic state processing method which can replace the camera dependency completely, or support as a backup mechanism until the camera system is back online, making the AI intersection control robust to camera failures.
The garment industry is one of the world’s largest carbon and waste polluters, expected to produce 150 billion garments per year in the next decade, while currently recycling about 1%. The lack of reliable fabric identification limits scalable recycling. Without traceability, governments cannot enforce circular economy legislation. We propose solutions to both issues that leverage low cost hardware and deep learning. Using microscope fabric images and Convolutional Neural Networks, we demonstrate classification accuracy of over 90% for 14 fabric classes. By marking fabrics with a UV-visible code readable via YOLOv8 object detection, we demonstrate an mAP of over 0.98, retaining up to 0.93 after wash cycles while remaining resilient to fabric-unique challenges such as creasing. These methods can be implemented worldwide at low cost to enable fabric identification and traceability for a textile circular economy. We demonstrate a prototype reader application and discuss pathways to impact. We also provide three new datasets for future research.
Analyzing agricultural imagery for farm level insights has been an active area of research in the recent times. For providing the necessary information to stakeholders - be it farmers, financial institutions or governments, various computer vision tasks have to come together. For example, to provide information to a farmer about crop stress in their farm, accurate localization of the farm, identification of the crop type and a monitoring of the field’s micro-climate must be done together. In this work, we set performance benchmarks for three computer vision tasks - farm boundary detection, crop classification and sub-field stress estimation with different modalities of images - Sentinel2, PlanetScope and Drone Imagery. We use public dataset benchmarks for farm boundaries and crop classification and do a controlled field study on a large sugarcane farm in Uttar Pradesh, India for the stress estimation. Our work benchmarks farm boundary detection for small farms with state of the art deep learning algorithms achieving a dice score of 67%, improves the sugarcane classification accuracy by 10% coming to 98% and demonstrates an accuracy of 72% for water and nitrogen stress estimation.
Delirium is a syndrome characterized by acute and fluctuating change in attention, awareness, and cognition. Common in older adults, especially those with underlying medical conditions, delirium is associated with higher mortality rates, longer hospital stays, and increased healthcare costs. Early identification and management of delirium therefore becomes essential in order to prevent adverse outcomes, improve patient health, and reduce healthcare costs. As multiple predisposing (e.g., neurological disorders) and precipitating factors (e.g., medications) can be involved in the aetiology of delirium, detecting the syndrome early on can be challenging. In this work, we present an interpretable multimodal checklist that can aid clinicians in delirium detection. Specifically, we leverage causal decision trees to extract most relevant features for delirium detection which are then used in learning the predictive checklist. Experiments demonstrate the efficacy of the approach over existing methods in terms of both detection accuracy and interpretability.
Rash driving detection is vital to prevent accidents and improve public safety. Existing rash driving solutions using hand-crafted features have several limitations. We propose a simple yet efficient two-step process to overcome the limitations of the existing works by leveraging smartphone sensor (accelerometer and gyroscope) data. The first step filters out normal driving data and retains only the abnormal driving data with the proposed Adaptive Time Window (ATW) algorithm. This not only enhances the accuracy of detection but also reduces computation time, making our solution more efficient. Importantly, the proposed ATW algorithm completely eliminates window overlap redundancy and edge effects in the system. The second step classifies abnormal driving patterns with the proposed 1D Convolutional Neural Network (CNN) model. Our results demonstrate that the proposed solution is highly accurate and has a weighted accuracy of 97.14%. Additionally, as part of this research, we have curated and released a labeled Indian dataset comprising five distinct rash driving patterns: Lane Weaving, Lane Swerving, Hard Braking, Hard Cornering, and Quick U-turn. This dataset can be valuable for further studies and aid in developing multimodal rash driving detection systems.
Research on mobile money smishing is hindered, especially in the African context, as there is a lack of data. The absence of datasets and the fact that Mobile Network Operators don’t maintain such data pose a significant challenge. Additionally, the absence of a data collection infrastructure further complicates the data acquisition process. In response to this challenge, we developed a scalable and cost-effective honeynet infrastructure tailored for the efficient collection of organic Short Message Service (SMS) messages. The innovative approach involves harnessing the capabilities of Raspberry Pi units, USB multipliers, SIM cards from MNOs and GSM modems to create a scalable and adaptable solution. This aims to enhance smishing data collection, facilitating more efficient research into mobile money smishing.
Childhood vaccinations are vital for protecting children from preventable disease and improving overall public health. However, generating reliable estimates of routine immunisation uptake, essential for appropriate policy planning and resource allocation is complicated by various data challenges. A specific challenge in estimating coverage with household surveys such as the National Family Health Survey is that the presence of vaccination is obtained via maternal recall if a health-based record is absent. This study examines the extent to which estimates of childhood immunisation coverage derived using depend on maternal recall: a mother’s ability to correctly identify which vaccines a child has received. In this study, we leverage spatial Bayesian models to estimate routine childhood immunisation rates at sub-national resolutions in 2015 and 2020 using various assumptions about the accuracy of maternal recall. This modelling approach and explicit consideration of maternal recall allows us to identify local regions whose previous estimates of vaccine coverage rates may be overstated due to low rates of the presence of health-based records. We create detailed vaccination coverage maps to analyze the models with and without maternal recall data. By highlighting vaccination “coldspots” and their change over time, this study reveals the potential benefits or limitations of using maternal recall in generating vaccine coverage estimates and provides a basis for more informed decision-making for immunization interventions in India and similar contexts.
Although Bangladesh is a country that values matchmaking and matchmakers greatly, online dating is a relatively new phenomenon that became popular only recently mostly among the young population. While anecdotal reports suggest a considerable number of individuals utilize online dating platforms in Bangladesh, a comprehensive investigation of their experiences and the potential impact on their well-being has yet to be conducted. Given the growing prevalence of online dating apps among young adults in Bangladesh, the present study aims to conduct in-depth interviews with a sample of 33 participants between the ages of 18 and 25 who actively use these platforms. The sample comprises 14 male and 19 female participants, each of whom has experience using online dating apps. By gaining insights into the experiences of young adult users of dating apps in Bangladesh and examining the effects of such platforms on their physical and mental well-being, the findings of this study highlight three areas: a) youth's intention to fight existing social norms, b) gender experiences relating to abuse and harassment in online dating platforms, c) alternative usage of online dating application space, particularly during the Covid-19 pandemic period. This study also seeks to raise awareness of online usage patterns among young adults in Bangladesh and identify potential interventions that may address any negative outcomes associated with excessive online activity.
Targeting the poor is an integral part of social program design in low-income countries. Geographical targeting gives priority to areas with high concentrations of poverty. However, traditional data sources, such as household surveys often lack the spatial resolution to estimate poverty at a highly disaggregated level and are costly to collect on a regular basis. We leverage the proliferation of big data obtained from mobile devices and satellites to generate poverty measures at a highly disaggregated spatial level in a new country context. Previous applications rely on computationally intensive methods to extract information from raw cellphone transaction data. We show how similar levels of prediction accuracy can be achieved by using key performance indicators (KPIs) that Mobile Network Operators produce regularly as part of their operations, lowering the financial and data access barriers to estimate and update prediction models. This can help to facilitate the use of these data for policy in low- and middle-income contexts.
The efficacy of a reconfigurable intelligent surface (RIS)-aided network for enhanced connectivity in skip zones is demonstrated through real-time video streaming. The demonstration is carried out with the help of National Instruments universal software radio peripheral (NI-USRP) devices integrated with an RIS prototype. The RIS prototype is designed and fabricated using a 16x10 metasur-face operating at 5.3 GHz carrier frequency. Utilizing LabVIEW's long-term evolution (LTE) application framework module, good connectivity between the transmitter and receiver in an otherwise skip zone is exhibited. Various modulation and coding schemes have been applied to the streamed data to observe the throughput, SINR, and constellation diagrams with respect to different positions of the receiver in the non-line of sight (NLOS) skip zone. The demonstration ratifies the potential of RIS system in practical applications related to future wireless communications.
Complementary feeding is crucial to promote healthy nutrition in infant and young children (IYC) and prevent malnutrition. Mothers, families, and healthcare professionals (HCPs) are crucial in helping IYC develop healthy eating habits. However, limited access to adequate nutritional information and health services impacts children’s nutrition, especially in low-resource settings. Technology opens up opportunities to address these challenges and potentially improve IYC feeding practices. Taking a co-design approach, we conducted low-fidelity prototyping workshops with caregivers and HCPs to explore the potential of tangible interfaces to facilitate play and promote healthy nutrition for IYC in two low-resource healthcare settings in Peru. Participants envisioned diverse tangible objects and interactions that could augment the waiting spaces of the healthcare centres, encouraging play and enhancing children’s and caregivers’ experiences, while promoting healthy nutrition and dietary diversity. We outline design opportunities to facilitate tangible play, shared playful experiences, and promote healthy nutrition in low-resource healthcare settings.
