The Hardness of Achieving Impact in AI for Social Impact Research: A Ground-Level View of Challenges & Opportunities
Aditya Majumdar, Wenbo Zhang, Kashvi Prawal, Amulya Yadav · 2025
Read The Hardness of Achieving Impact in AI for Social Impact Research: A Ground-Level View of Challenges & Opportunities by Aditya Majumdar, Wenbo Zhang, Kashvi Prawal, Amulya Yadav (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-14829.txt.
Introduction
Majumdar and three co-authors ask why AI for good causes stalls short of real use in this twenty twenty-five paper. The field pairs AI craft with social science to hit world goals on health and air. Yet most works stall at the proof stage and never reach real streets.
Inside, talks with more than two dozen workers in the field feed the team's map. Their themes span team shape, funds, talk, trust, and day-to-day run snags. Finds stress that a keen lab mate who will co-build and field a tool is rare. Best acts drawn from the talks are cast as a field guide for new teams. The team notes its view leans to north labs and says so plain.
Catalog record
| CREATOR | Aditya Majumdar, Wenbo Zhang, Kashvi Prawal, Amulya Yadav |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 75 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-14829
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The Hardness of Achieving Impact in AI for Social Impact Research: A Ground-Level View of Challenges & Opportunities Aditya Majumdar, Wenbo Zhang, Kashvi Prawal, Amulya Yadav Abstract In an attempt to tackle the United Nations Sustainable Development Goals (UN SDGs), AI for Social Impact (AI4SI) projects focus on harnessing AI to address societal issues in areas such as healthcare, social justice, etc. Unfortunately, despite growing interest in AI4SI, achieving tangible, on-the-ground impact remains a significant challenge. For example, identifying and engaging motivated collaborators who are willing to co-design and deploy AI based solutions in real-world settings is often difficult. Even when such partnerships are established, many AI4SI projects “fail” to progress beyond the proof-of-concept stage, and hence, are unable to transition to at-scale production-level solutions. Furthermore, the unique challenges faced by AI4SI researchers are not always fully recognized within the broader AI community, where such work is sometimes viewed as primarily applied and not aligning with the traditional criteria for novelty emphasized in core AI venues. This paper attempts to shine a light on the diverse challenges faced in AI4SI research by diagnosing a multitude of factors that prevent AI4SI partnerships from achieving real-world impact on the ground. Drawing on semi-structured interviews with six leading AI4SI researchers - complemented by the authors’ own lived experiences in conducting AI4SI research - this paper attempts to understand the day-to-day difficulties faced in developing and deploying socially impactful AI solutions. Through thematic analysis, we identify structural and organizational, communication, collaboration, and operational challenges as key barriers to deployment. While there are no easy fixes, we synthesize best practices and actionable strategies drawn from these interviews and our own work in this space. In doing so, we hope this paper serves as a practical reference guide for AI4SI researchers and partner organizations seeking to engage more effectively in socially impactful AI collaborations. 1 Introduction The number of Artificial Intelligence for Social Impact (AI4SI) 1 1 1 AI4SI is also commonly referred to as AI for Social Good (AI4SG). Though these terms are used interchangeably, we will consistently use AI4SI to refer to this research field in the rest of this paper. projects has risen rapidly in recent years. Between 2018 and 2023, the number tripled from 170 to over 600 projects ( Bankhwal et al. 2024 ) . Most projects in this field are guided by the 17 UN SDGs, which focus on advancing equity in healthcare ( Verma et al. 2023 ) , public safety ( Pita et al. 2008 ) , education ( Jiang and Pardos 2021 ) , wildlife conservation ( Gholami et al. 2017 ) , social justice ( Vajiac et al. 2024a ) , and other areas of global concern. Motivated by a philosophy of “ helping the helpers ”, these AI4SI projects are often carried out by AI researchers in close collaboration with partner organizations (POs) such as non-profits, hospitals, government agencies, and other community stakeholders ( Matias, Hassidim, and Nelson 2025 ) . Thus, they require deep collaboration across disciplinary, organizational, and institutional boundaries. Unlike traditional artificial intelligence (AI) and machine learning (ML) research, these projects are rooted in complex social ecosystems, and their results must align with the practical constraints, workflows, and values of the PO, and the low-resource community they serve. Therefore, expectations for an AI4SI project to be considered a “success” are stringent. Such projects must go beyond theoretical or empirical performance gains to demonstrably drive meaningful social impact within the low-resource communities they aim to serve. Unfortunately, achieving “successes” in the AI4SI space often proves challenging for a variety of reasons. Despite growing interest in the field, researchers often struggle to move AI4SI solutions beyond proof-of-concept or short-term pilots to long-term, at-scale deployment for their POs. A study conducted by Bankhwal et al. 2024 revealed that 72% of AI4SI projects never make it out of the research and development phase. Although technical issues can contribute to this challenge ( Zou et al. 2020 ) , they are rarely the sole reason why AI4SI solutions fail to transition into real-world deployment. Recognizing this, several frameworks ( Kshirsagar, Patel, and Rodriguez 2021 ; Tomašev et al. 2020 ; Floridi et al. 2020 ) have been proposed to aid researchers in designing responsible AI4SI technologies and sustain fruitful collaborations (see section 2 ). Although these frameworks offer insightful guidance, a key limitation in the current literature is that they abstract away from the lived realities of field deployment, making it nebulous how such practices translate when interdisciplinary teams attempt to build sustainable solutions in resource-constrained environments. Hence, this paper focuses on answering the following research question: What factors prevent AI for Social Impact projects from being deployed in real-world settings? To answer this question, we conduct a series of semi-structured interviews with active experienced researchers with successful collaborations in the field of AI4SI. Through these interviews, we identify the structural and organizational, communication, collaboration, and operational hurdles faced by researchers in the field that hinder deployment, supplemented by the authors’ own experiences in conducting AI4SI research. As a result, this paper describes the challenges encountered in developing and deploying AI-powered applications to help PO-based stakeholders, who serve low-resource and underserved communities, operate more efficiently with limited financial and human resources they possess.While there are no easy fixes to the challenges described in this paper, we outline best practices from interviews and our experiences that may help reduce their impact on AI4SI collaboration outcomes. In turn, we hope this paper serves as a helpful reference for both new and experienced AI4SI and Responsible AI practitioners, as well as community partners seeking to integrate AI into their operations. By highlighting commonly encountered challenges, we aim to help stakeholders engage more realistically and meaningfully in such projects. Further, we hope this discussion fosters a sense of awareness and appreciation about AI4SI research (and researchers) among the broader AI community, which is largely fixated on traditional notions of novelty and intellectual merit, and tends to devalue AI4SI research as applied or lacking technical novelty (such as hardness proofs, approximation guarantees, etc). 2 2 2 Our paper title reflects the authors’ experiences with AI4SI papers rejected from traditional AI venues for lacking discussion of the computational “hardness” of the formalized problems. 2 Related Work Conceptual Foundations & Design Principles for AI4SG : In their 2020 chapter, Floridi et al. 2020 look across twenty-seven AI‐for‐Good projects, from disaster‐response chatbots to health‐monitoring tools, and extract seven guiding principles. They emphasize starting with a problem that truly matters, bringing stakeholders into the design from day one, and building in ethical checks at every turn. Each principle is grounded in concrete practices, such as community co‐design workshops and automated fairness audits, to help teams translate ideals into real-world impact. Around the same time, Tomašev et al. 2020 surveyed efforts aligned with the UN’s 17 SDGs. By examining case studies in agriculture, public health and beyond, they distilled ten practical recommendations, everything from setting clear success metrics to embedding an ethics review board. Their work highlights how deep, ongoing collaboration with domain experts keeps projects on track and avoids the “tech for tech’s sake” trap. Building on these high‐level ideas, Kshirsagar, Patel, and Rodriguez 2021 take