From Social Media to Smart Advisory: Responsible AI, Data Justice, and the Governance of Farmer-Generated Knowledge in Smallholder Agriculture

From Social Media to Smart Advisory: Responsible AI, Data Justice, and the Governance of Farmer-Generated Knowledge in Smallholder Agriculture | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article From Social Media to Smart Advisory: Responsible AI, Data Justice, and the Governance of Farmer-Generated Knowledge in Smallholder Agriculture Tushar David This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8986304/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Smallholder farmers in developing countries generate rich, structured agricultural knowledge through social media platforms every day — crop diagnostic descriptions, peer-validated price information, technique queries, and market intelligence — without knowing they are doing so, without owning what they produce, and without governance frameworks that protect their rights over this data or ensure they benefit from its potential use in artificial intelligence systems. This paper examines farmer-generated social media data from an original survey of 720 smallholder farmers across all 20 administrative blocks of Prayagraj district, Uttar Pradesh, India, and applies the Responsible AI and Data Justice frameworks to evaluate the ethical conditions under which AI agricultural advisory systems could legitimately build on this data. We find that the data farmers generate constitutes a structurally valuable resource for training localised AI advisory systems — including labelled diagnostic image-text pairs, real-time hyperlocal price signals, and agricultural information demand data — but that current governance conditions are systematically inadequate: genuine informed consent is absent, community data ownership does not exist, language representation is ignored, and algorithmic accountability mechanisms are non-existent. We argue that the choice for agricultural AI development in the Global South is not between using farmer-generated data or not, but between using it extractively and using it responsibly. Realising responsible AI agricultural advisory requires not technical solutions but governance ones: community data trusts, participatory consent mechanisms, language justice requirements, algorithmic bias auditing, and public institutional architecture that aligns AI development with farming community interests rather than technology corporate extraction. Responsible AI Data Justice smallholder farmers agricultural advisory systems ICT4D India 1. Introduction "I describe my crop problem in the WhatsApp group and upload a photograph. Ten people respond within the hour. The problem gets solved. I never thought about where that description goes afterward." — Male farmer, 34, Block Meja, Prayagraj district, India Every day, millions of smallholder farmers across the Global South are generating something they do not realise has profound economic and political significance: structured agricultural data. When a farmer in Prayagraj photographs a yellowing wheat leaf and posts it in a WhatsApp group with a question about the cause, she is producing a labelled image-text pair of diagnostic agricultural data. When sixty farmers in the same group respond with diagnoses, treatments, and outcome reports over the following days, they are producing a crowd-sourced, community-validated agricultural knowledge corpus. When farmers query YouTube for pest identification videos, search WhatsApp groups for mandi prices, or discuss input effectiveness on Facebook, they generate continuous, hyperlocal signals about the agricultural information needs and knowledge stocks of smallholder communities. This data — distributed across commercial platforms, unarchived, unowned, and largely unexamined in the academic literature — constitutes one of the richest repositories of localised smallholder agricultural knowledge that exists anywhere. It is also profoundly at risk of appropriation. Artificial intelligence agricultural advisory systems — increasingly promoted across international development organisations, national governments, and technology corporations as the technological future of agricultural extension in the Global South — require exactly the kind of hyperlocal, crop-specific, climate-sensitive, dialect-language training data that smallholder farmers are generating through social media. India's Digital Agriculture Mission (2024–2028), with a ₹2,817 crore investment envelope, explicitly aims to build AI advisory systems at national scale. The question of whether, how, and under what ethical conditions farmer-generated social media data could or should inform such systems is among the most important and least examined governance questions in the emerging digital agriculture literature. This paper addresses that question. Drawing on an original large-scale survey of 720 smallholder farmers in Prayagraj district, it first characterises the structure, volume, and content of farmer-generated agricultural social media data. It then applies the Responsible AI framework (Jobin et al. 2019 ) and the Data Justice framework (Couldry and Mejias 2019 ; Taylor 2017 ) to evaluate current governance conditions and identify the specific ethical requirements for responsible AI use of this data. The paper argues that the potential of farmer-generated data for AI agricultural advisory systems is substantial but that the trajectory of agricultural AI development — driven by commercial technology corporations and national government missions without adequate governance design — is currently extractive rather than responsible. The paper makes three contributions. First , it provides original empirical characterisation of farmer-generated agricultural social media data as a potential AI training resource — the first such characterisation grounded in a large-scale survey of 720 farming households. Second , it extends the Responsible AI and Data Justice frameworks to the specific context of smallholder agricultural AI in a developing country, identifying governance requirements not addressed in existing frameworks designed for Global North contexts. Third , it proposes a concrete governance architecture — community data trusts, participatory consent, language justice requirements, and public institutional oversight — that would enable responsible rather than extractive AI agricultural development. 2. Theoretical Foundations: Responsible AI and Data Justice 2.1 Responsible AI The responsible AI framework emerges from the convergence of AI ethics scholarship and technology governance practice. A systematic review of 84 AI ethics frameworks from governments, corporations, and civil society organisations by Jobin et al. ( 2019 ) identifies five principles with the highest convergence: transparency, justice and fairness, non-maleficence, beneficence, and privacy. Transparency requires that AI systems be explainable — that their decision processes be understandable to the people affected by them. Justice and fairness require that AI systems not perpetuate or amplify existing inequalities. Non-maleficence requires that systems not cause harm. Beneficence requires that they actively improve human welfare. Privacy requires respect for the data rights of those whose information is processed. Jobin et al. ( 2019 ) note significant divergence in how these principles are operationalised across different contexts, and a near-total absence of frameworks addressing AI governance for smallholder agricultural contexts in developing countries. The application of responsible AI principles to advisory systems for farmers with low digital literacy, operating in minority languages, in institutional environments without regulatory infrastructure, requires context-specific governance design that existing frameworks do not provide. This contextualisation is a central task of the present paper. 2.2 Data Justice The Data Justice framework provides a complementary lens focused specifically on power relations in data generation, ownership, and use. Couldry and Mejias ( 2019 ) argue that the contemporary data economy represents a new form of colonial extraction: the systematic appropriation of value from communities — disproportionately in the Global South — through their digital activities, transferred without compensation to technology corporations. They term this "data colonialism": the naturalisation of a social order in which human life and activity are converted into raw material for commercial data processing. Taylor ( 2017 ) extends this to three operational dimensions of data justice: the ability to remain unidentified in data systems, the ability to be represented in data systems, and the ability to use data systems as tools for advocacy and self-determination. Applied to smallholder agricultural social media data, the Data Justice framework generates a specific set of empirically grounded questions. Who owns the agricultural knowledge encoded in farmer WhatsApp conversations — the farmers who generated it, the platforms that host it, or the AI developers who might use it? What forms of consent are meaningful for farming communities with limited digital literacy and no familiarity with AI systems? How can communities exercise data governance rights in environments without regulatory infrastructure? These questions structure the analysis in Sections 4 and 5 . 2.3 Analytical Framework Combining these theoretical foundations, this paper applies a five-dimension analytical matrix to evaluate the potential and risks of AI use of farmer-generated social media data: (1) data quality and representativeness; (2) consent and ownership; (3) language and accessibility; (4) algorithmic fairness; and (5) institutional accountability. Each dimension is evaluated against the empirical evidence from the Prayagraj survey, and governance conditions are derived for each. 3. Empirical Context and Data 3.1 Study Setting and Sample The empirical foundation of this paper is an original survey of 720 smallholder farmers conducted across all 20 administrative blocks of Prayagraj district, Uttar Pradesh, India (2023–24). Prayagraj is a predominantly agrarian district in the Gangetic plains with 477,569 cultivated hectares, a mean farm size of 2.8 hectares, and an agricultural economy centred on wheat, rice, and pulses. Rapid smartphone diffusion (45.2%) and 4G network expansion (78% coverage) have enabled social media adoption for agricultural purposes in 58.3% of surveyed farming households (n = 420 of 720). Sample characteristics: mean age 42.3 years (SD = 12.8); 68.4% male, 31.6% female; 72.3% small or marginal landholders below 2 hectares; 23.1% no formal education. Platform-specific adoption among social media users: WhatsApp 76.2%; YouTube 68.4%; Facebook 43.8%. Daily agricultural social media activity: 42.1% of adopters. Qualitative methods — farmer interviews (n = 20) and focus group discussions (n = 8) — supplemented the survey to provide the micro-level accounts of social media content generation cited throughout this paper. 3.2 The Data Farmers Generate: A Taxonomy Analysis of farmer accounts of social media agricultural activity reveals a structured taxonomy of data types with differential AI training value. The most significant category — reported by 73.4% of adopters — is crop problem description: a farmer-generated text and image pair describing symptoms, combined with community responses providing diagnoses, treatment recommendations, and outcome reports. This constitutes a labelled agricultural diagnostic dataset of a kind that is extremely costly to generate through institutional research means. Price information sharing is reported by 71.2% of adopters, generating continuous, hyperlocal price-location-time datasets. Weather and seasonal queries (78.3%), technique video engagement on YouTube (68.4%), and input product reviews (54.2%) constitute further high-value data categories. The cumulative data generation volume is substantial. The modal WhatsApp agricultural group in the survey has 28.4 members and receives an average of 14.7 agricultural messages per day during the crop season. With an average of 3.2 WhatsApp agricultural groups per adopter, the 420 adopters in this survey alone generate approximately 17,000 agricultural social media interactions per day during the growing season. The governance question is not whether this data exists and has value for AI training — it does, clearly. The question is whether AI systems will be designed to benefit the communities that generate it. Table 1 Taxonomy of Farmer-Generated Agricultural Social Media Data and Responsible AI Governance Implications (N = 420 Adopters) Data Category % Adopters Platform AI Training Value Governance Risk Crop diagnostic image-text pairs 73.4% WhatsApp High — labelled diagnostic training data Consent absent; data owned by platform Price information sharing 71.2% WhatsApp/Facebook High — hyperlocal price-location-time signals Value extraction without community benefit Weather and seasonal queries 78.3% WhatsApp Medium — local microclimate indicators Aggregated surveillance risk Technique video engagement 68.4% YouTube High — information demand signals Behavioural data owned by Google/Alphabet Input product queries/reviews 54.2% WhatsApp/Facebook High — input effectiveness feedback Commercial misuse for product targeting Note. Percentages based on 420 social media-adopting farmers. AI Training Value and Governance Risk assessed against the five-dimension analytical framework. 4. Responsible AI Analysis: Five Governance Dimensions 4.1 Consent and Ownership: The Foundational Failure The most fundamental governance failure in the current trajectory of AI agricultural development is the absence of meaningful informed consent. When farmers post crop problems in WhatsApp groups, share prices on Facebook, or query YouTube for agricultural techniques, they are not consenting to the use of that data to train AI systems. Platform terms of service that technically permit such use are written in English, are not read by rural Indian users, and cannot constitute informed consent in any meaningful ethical sense. The survey confirms this: 94.2% of social media adopters in Prayagraj reported no awareness that their social media activity could be used to train AI systems. The Data Justice analysis of this situation is unambiguous. Couldry and Mejias ( 2019 ) describe "data colonialism" as the appropriation of value from communities through their digital activities without their knowledge or consent — precisely the structure in operation here. Farmers generating diagnostic image-text pairs on WhatsApp are producing labelled training data that could generate substantial commercial value for technology corporations if aggregated and used to train AI advisory systems. That value would be created from farmer knowledge, in farmer language, about farmer problems — but would flow entirely to entities outside the farming community. The responsible AI principle of privacy (Jobin et al. 2019 ) requires that data be processed only with genuine, informed consent. The current situation violates this principle categorically. 4.2 Language and Representation: Structural Exclusion Survey data reveals that 67.4% of agricultural WhatsApp communication in Prayagraj occurs primarily in Hindi, 18.2% in a mixture of Hindi and local dialect (predominantly Awadhi and Bhojpuri), and only 14.4% primarily in English. YouTube agricultural content consumption is predominantly in Hindi (71.3%) or regional dialect (12.4%). This linguistic reality has direct implications for AI system fairness. The dominant agricultural AI systems currently available — including IBM's Watson Decision Platform for Agriculture, Microsoft's FarmBeats, and various chatbot advisory platforms promoted through India's Digital Agriculture Mission — are built predominantly on English-language training data and perform significantly worse in Hindi and essentially not at all in regional Indian dialects. Farmer-generated social media data represents a potential corrective: a large Hindi and dialect-language corpus of agricultural knowledge that could train AI systems genuinely calibrated for the farmers who generated it. The governance question is whether this potential will be realised for those farmers, or whether their dialect-language knowledge will improve AI systems that remain inaccessible to them because they produce outputs only in English or standard Hindi. This is a justice and fairness failure in the terms of Jobin et al. ( 2019 ): an AI system that extracts knowledge from a community to improve services for a different community. It also represents a transparency failure — farmers cannot evaluate whether AI advisory outputs are based on training data from their own agricultural context or from research stations in other countries. 4.3 Algorithmic Fairness: The Bias Embedded in Available Data Social media adoption in Prayagraj is not demographically neutral. Male farmers adopt at 63.1%; female farmers at 47.6% (chi-square = 14.82, p < 0.001). Adopters are significantly more educated, more centrally positioned in community networks, and more likely to have regular extension contact than non-adopters. AI systems trained on currently available farmer-generated social media data will therefore learn from a sample that systematically underrepresents the most marginalised farming households: female farmers, low-literacy farmers, and peripheral community members. The consequence is a specific form of algorithmic unfairness: AI advisory systems trained on available farmer-generated data will perform best for farmers who already benefit most from digital agricultural networks — connected, educated, centrally networked male farmers. They will perform worst for the 41.7% of farmers not currently using social media — those with the lowest incomes, lowest literacy, and least institutional support. The responsible AI principle of fairness (Jobin et al. 2019 ) requires that AI systems not amplify existing inequalities. Without active bias correction in data curation and model design, agricultural AI systems built on current social media data will do exactly that. 4.4 Platform Accountability: The Political Economy of Agricultural Data WhatsApp (Meta), YouTube (Google/Alphabet), and Facebook (Meta) are the primary repositories of farmer-generated agricultural social media data. Their terms of service grant broad rights to use content for AI training. They have no legal, contractual, or institutional obligation to share AI advisory systems trained on this data with farming communities, ensure systems are available in local languages, or compensate farming communities for the value their knowledge contributes. The three largest repositories of smallholder agricultural knowledge in India are owned by corporations headquartered in California with no accountability to Indian farming communities. Data Justice analysis Taylor ( 2017 ) argues that data justice requires the ability to use data systems as tools for self-determination. The current structure of platform data governance systematically denies this capacity. Farmers can generate data through commercial platforms but cannot govern it, direct its use, or benefit from its AI applications. This is a political economy problem — a distribution of power over agricultural knowledge that favours technology corporations at the expense of farming communities — and it requires political and institutional solutions, not technical ones. 4.5 Institutional Accountability: The Governance Vacuum The fifth governance failure is institutional: the absence of accountable public institutions with both the mandate and the capacity to govern agricultural AI in the interests of farming communities. India's Digital Agriculture Mission creates a national digital agriculture infrastructure but does not, in its current design, include governance provisions for farmer-generated data, mandatory language requirements for AI advisory systems, independent bias auditing, or farmer representation in AI system governance. National agricultural research institutions (ICAR, SAUs) have institutional legitimacy but lack technical AI capacity. Technology corporations have technical capacity but lack accountability to farming communities. The governance vacuum between these two positions is precisely where data extraction becomes structurally inevitable. 5. Towards Responsible Agricultural AI: Governance Conditions 5.1 Community Data Trusts The foundational governance requirement is a legal institutional arrangement under which farmer-generated agricultural social media data is held collectively on behalf of the farming communities that generate it. Community data trusts — legal entities that hold data, negotiate the terms of its use, and ensure value flows back to member communities — have been proposed in various forms in the data governance literature (Hardinges et al. 2019 ). For smallholder agricultural data in India, such trusts could be constituted at the Farmer Producer Organisation (FPO) level, with oversight by state agricultural departments, and with mandatory representation of female farmers and marginal landholders in governance structures. India's proposed National Data Governance Framework provides a potential legislative foundation, though its current architecture does not specifically address community agricultural data. 5.2 Participatory Consent Mechanisms Meaningful informed consent for AI use of farmer-generated agricultural data cannot be achieved through platform terms of service. It requires community-level consent processes mediated by trusted institutional actors. Agricultural extension officers — identified in the survey as the single most powerful predictor of digital platform adoption, and as the primary quality-assurance mechanism for digital agricultural information — are the most feasible mechanism for community-level consent mediation. Extension officers who understand and can explain data governance arrangements in local languages are the practical bridge between abstract responsible AI principles and the operational consent of farming communities. 5.3 Language Justice Requirements Any AI agricultural advisory system claiming to serve smallholder farmers in northern India must operate in Hindi and in the major regional dialects — Awadhi, Bhojpuri, Maithili — that are the primary agricultural communication languages of the region. This requires deliberate training data curation ensuring dialect-language content is adequately represented, investment in natural language processing for low-resource Indian languages, and quality evaluation of AI advisory outputs conducted in target languages by representative farmer users. Output accessibility requires voice-based interfaces and visual formats for users with low literacy — the 23.1% of survey respondents with no formal education cannot access text-based AI advisory outputs regardless of language. 5.4 Algorithmic Bias Auditing and Accountability AI agricultural advisory systems must be subject to mandatory regular bias auditing — systematic evaluation of whether recommendations perform differently across the demographic subgroups the system is intended to serve, with specific attention to gender and literacy performance differentials. Results should be publicly reported and should trigger mandatory system corrections where performance gaps exceed defined thresholds. Independent bias auditing should be conducted by institutions without financial interest in the AI system — national agricultural universities are a natural candidate in the Indian context. 5.5 Public Institutional Architecture The governance conditions above imply a specific institutional architecture. Commercial technology corporations cannot be the primary developers and operators of AI agricultural advisory systems in the smallholder context — their incentive structures are misaligned with community data ownership, language justice, and bias accountability. The most viable model is a public-private partnership governed by farmer representation: AI systems developed by technical partners under contractual governance arrangements that mandate community ownership, language requirements, bias auditing, and data benefit sharing. India's Digital Agriculture Mission, with its ₹2,817 crore investment, has the resources to establish this architecture — if governance design is prioritised alongside infrastructure development. 6. Discussion 6.1 From Data Extraction to Knowledge Amplification The central argument of this paper is that the choice facing agricultural AI development in the Global South is not between using farmer-generated data and not — that decision has effectively been made by the commercial platforms that host and legally own it. The choice is between extractive use and responsible use: between AI systems that appropriate smallholder knowledge to generate value for technology corporations, and AI systems that amplify smallholder knowledge to generate value for farming communities. The Prayagraj survey documents a farming community that has built, without institutional support, a functioning digital agricultural knowledge system: peer diagnostic networks, distributed price discovery mechanisms, community-validated information flows. That system generates data daily that could train AI advisory capabilities superior to anything currently available to these farmers. An AI diagnostic system trained on farmer-generated WhatsApp image-text pairs from Prayagraj would be substantially better calibrated for Prayagraj farming conditions than any model trained on research-station data. A real-time price advisory system trained on the hyperlocal price-sharing data generated in farmer WhatsApp groups would provide information quality that formal price information services cannot match. The governance question is whether agricultural AI development will be designed to serve the communities that generate this knowledge, or to extract it from them. The current trajectory — commercial platform ownership of data, public investment without governance design, AI systems built for institutional rather than farmer users — points toward extraction. The responsible AI and data justice frameworks provide the tools to design a different trajectory. 6.2 Implications for India's Digital Agriculture Mission India's Digital Agriculture Mission represents an opportunity to establish the governance architecture for responsible agricultural AI at national scale — and a risk of institutionalising extractive AI development through inadequate governance design. The Mission's current architecture creates the infrastructure conditions for AI agricultural advisory systems but does not include the governance provisions identified in this paper: farmer data ownership frameworks, language justice requirements, independent bias auditing, or farmer representation in AI governance. Without these provisions, the Mission risks producing AI advisory systems that amplify existing agricultural inequalities rather than correcting them — performing best for the already-connected, already-educated, already-networked farmers who need them least, while remaining inaccessible and poorly calibrated for the most marginalised farming communities. The governance architecture proposed in this paper — community data trusts, participatory consent, language requirements, bias auditing — is not an add-on to be designed after AI systems are built. It is the precondition for those systems having a legitimate claim to serve development rather than extraction. 6.3 Limitations Four limitations merit acknowledgement. First, the characterisation of farmer-generated data types relies on survey self-reports and qualitative accounts rather than direct analysis of social media content — platform access restrictions prevent systematic content analysis, a methodological constraint affecting all research in this area. Second, the governance conditions proposed in Section 5 are normative derivations from the Responsible AI and Data Justice frameworks and require validation through participatory design processes with farming communities. Third, the analysis is geographically bounded to Prayagraj district; governance requirements in other South Asian or sub-Saharan African smallholder contexts may differ. Fourth, the rapidly evolving landscape of AI regulation in India means specific institutional recommendations may require updating as the National Data Governance Framework and Digital Agriculture Mission architecture develop. 7. Conclusion The farmers of Prayagraj are generating an agricultural knowledge commons. Without knowing it, without owning it, and without governance frameworks that protect it, they are daily producing the diagnostic data, price intelligence, and community knowledge that could train AI advisory systems more useful and more locally calibrated than any agricultural AI system currently available. The question this paper has asked is not whether that knowledge has value — it clearly does. The question is whether the AI systems that could be trained on it will be built to serve the communities that generated it. The Responsible AI and Data Justice frameworks converge on a clear verdict about the current trajectory: systematic extraction is more likely than responsible amplification, because the governance conditions for responsible use do not exist. Consent is absent. Community ownership is absent. Language justice is absent. Bias accountability is absent. Institutional architecture that aligns AI development with farming community interests rather than technology corporate extraction does not yet exist. Building these governance conditions is not a peripheral task to be completed after AI systems are deployed. It is the ethical precondition on which any claim that agricultural AI serves development — rather than replicating the colonial extraction of knowledge from the Global South for the benefit of the Global North — ultimately depends. India's Digital Agriculture Mission, and the broader international community of agricultural AI development, faces a governance choice. This paper has tried to make its terms clear. Declarations This study was conducted as part of the author’s academic dissertation at Sam Higginbottom University of Agriculture, Technology and Sciences (SHUATS), Prayagraj, India, and was reviewed and approved by the university’s academic evaluation committee as part of the degree requirements. The research was carried out under the supervision of designated faculty members in accordance with institutional academic and ethical guidelines for research involving human participants. All participants were informed about the purpose of the study, and their voluntary informed consent was obtained prior to participation. Data were collected anonymously, and confidentiality of respondents was strictly maintained throughout the research process. Author Contribution T.D. conceptualised the study, designed the research methodology, collected and analysed the data, developed the theoretical framework, and wrote the manuscript in its entirety. The author read and approved the final manuscript. Acknowledgement The author thanks the 720 farming households of Prayagraj district who participated in this research, and the block development officers of Uttar Pradesh who facilitated fieldwork. The author used Claude (Anthropic, claude-sonnet-4-6, February 2026) for language refinement and formatting assistance only. All research design, data collection, analysis, and writing are the author's sole responsibility. Data Availability The survey data that support the findings of this study are available from the corresponding author upon reasonable request. Data are not publicly deposited due to informed consent restrictions protecting participant confidentiality. References Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research 81:1–15. Couldry, N., & Mejias, U. A. (2019). The costs of connection: How data is colonizing human life and appropriating it for capitalism . Stanford University Press. Duncombe, R. (2016). Mobile phones for agricultural and rural development: A literature review and suggestions for future research. European Journal of Development Research , 28 (2), 213–235. https://doi.org/10.1057/ejdr.2015.29 FAO. (2022). Artificial intelligence in the agrifood sector: Applications, risks and governance . Food and Agriculture Organisation of the United Nations. Floridi, L., Cowls, J., Beltrametti, M., et al. (2018). An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations. Minds and Machines , 28 (4), 689–707. https://doi.org/10.1007/s11023-018-9482-5 Hardinges, J., Tennison, J., Shadbolt, N., et al. (2019). Data trusts: From theory to practice . Open Data Institute. Heeks, R. (2020). ICT4D 3.0? Part 1 — The components of an emerging fourth-generation paradigm. Electronic Journal of Information Systems in Developing Countries , 86 (3), e12124. https://doi.org/10.1002/isd2.12124 Jensen, R. (2007). The digital provide: Information (technology), market performance, and welfare in the south Indian fisheries sector. Quarterly Journal of Economics , 122 (3), 879–924. https://doi.org/10.1162/qjec.122.3.879 Jobin, A., Ienca, M., & Vayena, E. (2019). The global landscape of AI ethics guidelines. Nature Machine Intelligence , 1 (9), 389–399. https://doi.org/10.1038/s42256-019-0088-2 Lioutas, E. D., & Charatsari, C. (2020). Smart farming and short food supply chains: Are they compatible? Land Use Policy , 94 , 104541. https://doi.org/10.1016/j.landusepol.2020.104541 Milan, S., & Treré, E. (2019). Big data from the South(s): Beyond data universalism. Television & New Media , 20 (4), 319–335. https://doi.org/10.1177/1527476419837739 Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016). The ethics of algorithms: Mapping the debate. Big Data & Society , 3 (2), 1–21. https://doi.org/10.1177/2053951716679679 Mooney, P. (2018). Blocking the chain: Industrial food chain concentration, big data platforms and food sovereignty solutions . ETC Group. Pringle, R. (2019). Five ways AI could improve the world. Nature , 570 (7761), 151. https://doi.org/10.1038/d41586-019-01831-7 Rahwan, I. (2018). Society-in-the-loop: Programming the algorithmic social contract. Ethics and Information Technology , 20 (1), 5–14. https://doi.org/10.1007/s10676-017-9430-8 Sharma, R., Kamble, S. S., Gunasekaran, A., Kumar, V., & Kumar, A. (2020). A systematic literature review on machine learning applications for sustainable agriculture supply chain performance. Computers & Operations Research , 119 , 104926. https://doi.org/10.1016/j.cor.2020.104926 Taylor, L. (2017). What is data justice? The case for connecting digital rights and freedoms globally. Big Data & Society , 4 (2), 1–14. https://doi.org/10.1177/2053951717736335 Vinuesa, R., Azizpour, H., Leite, I., et al. (2020). The role of artificial intelligence in achieving the Sustainable Development Goals. Nature Communications , 11 (1), 1–10. https://doi.org/10.1038/s41467-019-14108-y World Bank. (2022). Future of food: Harnessing digital technologies to improve food system outcomes . World Bank. Zuboff, S. (2019). The age of surveillance capitalism: The fight for a human future at the new frontier of power . PublicAffairs. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8986304","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":619828173,"identity":"1c110fce-3f32-45e2-b4cc-a6cfce4fa686","order_by":0,"name":"Tushar David","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABD0lEQVRIie3RsWoCMRjA8e84yC1f2zWSU1/hRLAtCr6K0uEWXZ3vEOKiuGaqr9DuHSKB3CLOQjucy00OjgpX6F0p0oKR0qlD/oRAQn4kEACb7X9GysmJikH3IPG0SQzn8TupiJKQ3xMAhiDhREx1aZilxxeozsVD/NrJ3/ybxSSF/UjBNYvO30IHt41pBk2x6Y/bQ54h1Rg4Yq2A+NJECEUJ/Wi15GwYKQSN4F7xgtCegYRZJS/IoiR3ucK69lL3/SLptVh5y1MScwZEYaAhcJ1LZLVrMV/S5nMSj+9nXGFDD4LldB2iiXiT4mE72ak+Km+7OeSqW1PJNj2M2rW6OE++oj+XEj5/zGaz2Wx/7QNlxlyizegaJAAAAABJRU5ErkJggg==","orcid":"","institution":"Sam Higginbottom Institute of Agriculture","correspondingAuthor":true,"prefix":"","firstName":"Tushar","middleName":"","lastName":"David","suffix":""}],"badges":[],"createdAt":"2026-02-27 09:43:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8986304/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8986304/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107706045,"identity":"7e1a1906-1318-486b-af46-13499c85b9da","added_by":"auto","created_at":"2026-04-24 09:17:14","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":223731,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8986304/v1/504f5b38-e856-4934-acde-fd76a6894221.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"From Social Media to Smart Advisory: Responsible AI, Data Justice, and the Governance of Farmer-Generated Knowledge in Smallholder Agriculture","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e \u003cem\u003e\"I describe my crop problem in the WhatsApp group and upload a photograph. Ten people respond within the hour. The problem gets solved. I never thought about where that description goes afterward.\" \u0026mdash; Male farmer, 34, Block Meja, Prayagraj district, India\u003c/em\u003e \u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003eEvery day, millions of smallholder farmers across the Global South are generating something they do not realise has profound economic and political significance: structured agricultural data. When a farmer in Prayagraj photographs a yellowing wheat leaf and posts it in a WhatsApp group with a question about the cause, she is producing a labelled image-text pair of diagnostic agricultural data. When sixty farmers in the same group respond with diagnoses, treatments, and outcome reports over the following days, they are producing a crowd-sourced, community-validated agricultural knowledge corpus. When farmers query YouTube for pest identification videos, search WhatsApp groups for mandi prices, or discuss input effectiveness on Facebook, they generate continuous, hyperlocal signals about the agricultural information needs and knowledge stocks of smallholder communities. This data \u0026mdash; distributed across commercial platforms, unarchived, unowned, and largely unexamined in the academic literature \u0026mdash; constitutes one of the richest repositories of localised smallholder agricultural knowledge that exists anywhere.\u003c/p\u003e \u003cp\u003eIt is also profoundly at risk of appropriation. Artificial intelligence agricultural advisory systems \u0026mdash; increasingly promoted across international development organisations, national governments, and technology corporations as the technological future of agricultural extension in the Global South \u0026mdash; require exactly the kind of hyperlocal, crop-specific, climate-sensitive, dialect-language training data that smallholder farmers are generating through social media. India's Digital Agriculture Mission (2024\u0026ndash;2028), with a ₹2,817 crore investment envelope, explicitly aims to build AI advisory systems at national scale. The question of whether, how, and under what ethical conditions farmer-generated social media data could or should inform such systems is among the most important and least examined governance questions in the emerging digital agriculture literature.\u003c/p\u003e \u003cp\u003eThis paper addresses that question. Drawing on an original large-scale survey of 720 smallholder farmers in Prayagraj district, it first characterises the structure, volume, and content of farmer-generated agricultural social media data. It then applies the Responsible AI framework (Jobin et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) and the Data Justice framework (Couldry and Mejias \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Taylor \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) to evaluate current governance conditions and identify the specific ethical requirements for responsible AI use of this data. The paper argues that the potential of farmer-generated data for AI agricultural advisory systems is substantial but that the trajectory of agricultural AI development \u0026mdash; driven by commercial technology corporations and national government missions without adequate governance design \u0026mdash; is currently extractive rather than responsible.\u003c/p\u003e \u003cp\u003eThe paper makes three contributions. \u003cem\u003eFirst\u003c/em\u003e, it provides original empirical characterisation of farmer-generated agricultural social media data as a potential AI training resource \u0026mdash; the first such characterisation grounded in a large-scale survey of 720 farming households. \u003cem\u003eSecond\u003c/em\u003e, it extends the Responsible AI and Data Justice frameworks to the specific context of smallholder agricultural AI in a developing country, identifying governance requirements not addressed in existing frameworks designed for Global North contexts. \u003cem\u003eThird\u003c/em\u003e, it proposes a concrete governance architecture \u0026mdash; community data trusts, participatory consent, language justice requirements, and public institutional oversight \u0026mdash; that would enable responsible rather than extractive AI agricultural development.\u003c/p\u003e"},{"header":"2. Theoretical Foundations: Responsible AI and Data Justice","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Responsible AI\u003c/h2\u003e \u003cp\u003eThe responsible AI framework emerges from the convergence of AI ethics scholarship and technology governance practice. A systematic review of 84 AI ethics frameworks from governments, corporations, and civil society organisations by Jobin et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) identifies five principles with the highest convergence: transparency, justice and fairness, non-maleficence, beneficence, and privacy. Transparency requires that AI systems be explainable \u0026mdash; that their decision processes be understandable to the people affected by them. Justice and fairness require that AI systems not perpetuate or amplify existing inequalities. Non-maleficence requires that systems not cause harm. Beneficence requires that they actively improve human welfare. Privacy requires respect for the data rights of those whose information is processed.\u003c/p\u003e \u003cp\u003eJobin et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) note significant divergence in how these principles are operationalised across different contexts, and a near-total absence of frameworks addressing AI governance for smallholder agricultural contexts in developing countries. The application of responsible AI principles to advisory systems for farmers with low digital literacy, operating in minority languages, in institutional environments without regulatory infrastructure, requires context-specific governance design that existing frameworks do not provide. This contextualisation is a central task of the present paper.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data Justice\u003c/h2\u003e \u003cp\u003eThe Data Justice framework provides a complementary lens focused specifically on power relations in data generation, ownership, and use. Couldry and Mejias (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) argue that the contemporary data economy represents a new form of colonial extraction: the systematic appropriation of value from communities \u0026mdash; disproportionately in the Global South \u0026mdash; through their digital activities, transferred without compensation to technology corporations. They term this \"data colonialism\": the naturalisation of a social order in which human life and activity are converted into raw material for commercial data processing. Taylor (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) extends this to three operational dimensions of data justice: the ability to remain unidentified in data systems, the ability to be represented in data systems, and the ability to use data systems as tools for advocacy and self-determination.\u003c/p\u003e \u003cp\u003eApplied to smallholder agricultural social media data, the Data Justice framework generates a specific set of empirically grounded questions. Who owns the agricultural knowledge encoded in farmer WhatsApp conversations \u0026mdash; the farmers who generated it, the platforms that host it, or the AI developers who might use it? What forms of consent are meaningful for farming communities with limited digital literacy and no familiarity with AI systems? How can communities exercise data governance rights in environments without regulatory infrastructure? These questions structure the analysis in Sections \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Sec15\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Analytical Framework\u003c/h2\u003e \u003cp\u003eCombining these theoretical foundations, this paper applies a five-dimension analytical matrix to evaluate the potential and risks of AI use of farmer-generated social media data: (1) data quality and representativeness; (2) consent and ownership; (3) language and accessibility; (4) algorithmic fairness; and (5) institutional accountability. Each dimension is evaluated against the empirical evidence from the Prayagraj survey, and governance conditions are derived for each.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Empirical Context and Data","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Study Setting and Sample\u003c/h2\u003e \u003cp\u003eThe empirical foundation of this paper is an original survey of 720 smallholder farmers conducted across all 20 administrative blocks of Prayagraj district, Uttar Pradesh, India (2023\u0026ndash;24). Prayagraj is a predominantly agrarian district in the Gangetic plains with 477,569 cultivated hectares, a mean farm size of 2.8 hectares, and an agricultural economy centred on wheat, rice, and pulses. Rapid smartphone diffusion (45.2%) and 4G network expansion (78% coverage) have enabled social media adoption for agricultural purposes in 58.3% of surveyed farming households (n\u0026thinsp;=\u0026thinsp;420 of 720). Sample characteristics: mean age 42.3 years (SD\u0026thinsp;=\u0026thinsp;12.8); 68.4% male, 31.6% female; 72.3% small or marginal landholders below 2 hectares; 23.1% no formal education.\u003c/p\u003e \u003cp\u003ePlatform-specific adoption among social media users: WhatsApp 76.2%; YouTube 68.4%; Facebook 43.8%. Daily agricultural social media activity: 42.1% of adopters. Qualitative methods \u0026mdash; farmer interviews (n\u0026thinsp;=\u0026thinsp;20) and focus group discussions (n\u0026thinsp;=\u0026thinsp;8) \u0026mdash; supplemented the survey to provide the micro-level accounts of social media content generation cited throughout this paper.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.2 The Data Farmers Generate: A Taxonomy\u003c/h2\u003e \u003cp\u003eAnalysis of farmer accounts of social media agricultural activity reveals a structured taxonomy of data types with differential AI training value. The most significant category \u0026mdash; reported by 73.4% of adopters \u0026mdash; is crop problem description: a farmer-generated text and image pair describing symptoms, combined with community responses providing diagnoses, treatment recommendations, and outcome reports. This constitutes a labelled agricultural diagnostic dataset of a kind that is extremely costly to generate through institutional research means. Price information sharing is reported by 71.2% of adopters, generating continuous, hyperlocal price-location-time datasets. Weather and seasonal queries (78.3%), technique video engagement on YouTube (68.4%), and input product reviews (54.2%) constitute further high-value data categories.\u003c/p\u003e \u003cp\u003eThe cumulative data generation volume is substantial. The modal WhatsApp agricultural group in the survey has 28.4 members and receives an average of 14.7 agricultural messages per day during the crop season. With an average of 3.2 WhatsApp agricultural groups per adopter, the 420 adopters in this survey alone generate approximately 17,000 agricultural social media interactions per day during the growing season. The governance question is not whether this data exists and has value for AI training \u0026mdash; it does, clearly. The question is whether AI systems will be designed to benefit the communities that generate it.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cem\u003eTaxonomy of Farmer-Generated Agricultural Social Media Data and Responsible AI Governance Implications (N\u0026thinsp;=\u0026thinsp;420 Adopters)\u003c/em\u003e\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003e Data Category\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e% Adopters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePlatform\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAI Training Value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGovernance Risk\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCrop diagnostic image-text pairs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e73.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhatsApp\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHigh \u0026mdash; labelled diagnostic training data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eConsent absent; data owned by platform\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrice information sharing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e71.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhatsApp/Facebook\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHigh \u0026mdash; hyperlocal price-location-time signals\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eValue extraction without community benefit\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeather and seasonal queries\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e78.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhatsApp\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMedium \u0026mdash; local microclimate indicators\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAggregated surveillance risk\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTechnique video engagement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e68.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eYouTube\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHigh \u0026mdash; information demand signals\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBehavioural data owned by Google/Alphabet\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInput product queries/reviews\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e54.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWhatsApp/Facebook\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHigh \u0026mdash; input effectiveness feedback\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCommercial misuse for product targeting\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eNote. Percentages based on 420 social media-adopting farmers. AI Training Value and Governance Risk assessed against the five-dimension analytical framework.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Responsible AI Analysis: Five Governance Dimensions","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Consent and Ownership: The Foundational Failure\u003c/h2\u003e \u003cp\u003eThe most fundamental governance failure in the current trajectory of AI agricultural development is the absence of meaningful informed consent. When farmers post crop problems in WhatsApp groups, share prices on Facebook, or query YouTube for agricultural techniques, they are not consenting to the use of that data to train AI systems. Platform terms of service that technically permit such use are written in English, are not read by rural Indian users, and cannot constitute informed consent in any meaningful ethical sense. The survey confirms this: 94.2% of social media adopters in Prayagraj reported no awareness that their social media activity could be used to train AI systems.\u003c/p\u003e \u003cp\u003eThe Data Justice analysis of this situation is unambiguous. Couldry and Mejias (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) describe \"data colonialism\" as the appropriation of value from communities through their digital activities without their knowledge or consent \u0026mdash; precisely the structure in operation here. Farmers generating diagnostic image-text pairs on WhatsApp are producing labelled training data that could generate substantial commercial value for technology corporations if aggregated and used to train AI advisory systems. That value would be created from farmer knowledge, in farmer language, about farmer problems \u0026mdash; but would flow entirely to entities outside the farming community. The responsible AI principle of privacy (Jobin et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) requires that data be processed only with genuine, informed consent. The current situation violates this principle categorically.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Language and Representation: Structural Exclusion\u003c/h2\u003e \u003cp\u003eSurvey data reveals that 67.4% of agricultural WhatsApp communication in Prayagraj occurs primarily in Hindi, 18.2% in a mixture of Hindi and local dialect (predominantly Awadhi and Bhojpuri), and only 14.4% primarily in English. YouTube agricultural content consumption is predominantly in Hindi (71.3%) or regional dialect (12.4%). This linguistic reality has direct implications for AI system fairness.\u003c/p\u003e \u003cp\u003eThe dominant agricultural AI systems currently available \u0026mdash; including IBM's Watson Decision Platform for Agriculture, Microsoft's FarmBeats, and various chatbot advisory platforms promoted through India's Digital Agriculture Mission \u0026mdash; are built predominantly on English-language training data and perform significantly worse in Hindi and essentially not at all in regional Indian dialects. Farmer-generated social media data represents a potential corrective: a large Hindi and dialect-language corpus of agricultural knowledge that could train AI systems genuinely calibrated for the farmers who generated it. The governance question is whether this potential will be realised for those farmers, or whether their dialect-language knowledge will improve AI systems that remain inaccessible to them because they produce outputs only in English or standard Hindi.\u003c/p\u003e \u003cp\u003eThis is a justice and fairness failure in the terms of Jobin et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e): an AI system that extracts knowledge from a community to improve services for a different community. It also represents a transparency failure \u0026mdash; farmers cannot evaluate whether AI advisory outputs are based on training data from their own agricultural context or from research stations in other countries.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Algorithmic Fairness: The Bias Embedded in Available Data\u003c/h2\u003e \u003cp\u003eSocial media adoption in Prayagraj is not demographically neutral. Male farmers adopt at 63.1%; female farmers at 47.6% (chi-square\u0026thinsp;=\u0026thinsp;14.82, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Adopters are significantly more educated, more centrally positioned in community networks, and more likely to have regular extension contact than non-adopters. AI systems trained on currently available farmer-generated social media data will therefore learn from a sample that systematically underrepresents the most marginalised farming households: female farmers, low-literacy farmers, and peripheral community members.\u003c/p\u003e \u003cp\u003eThe consequence is a specific form of algorithmic unfairness: AI advisory systems trained on available farmer-generated data will perform best for farmers who already benefit most from digital agricultural networks \u0026mdash; connected, educated, centrally networked male farmers. They will perform worst for the 41.7% of farmers not currently using social media \u0026mdash; those with the lowest incomes, lowest literacy, and least institutional support. The responsible AI principle of fairness (Jobin et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) requires that AI systems not amplify existing inequalities. Without active bias correction in data curation and model design, agricultural AI systems built on current social media data will do exactly that.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Platform Accountability: The Political Economy of Agricultural Data\u003c/h2\u003e \u003cp\u003eWhatsApp (Meta), YouTube (Google/Alphabet), and Facebook (Meta) are the primary repositories of farmer-generated agricultural social media data. Their terms of service grant broad rights to use content for AI training. They have no legal, contractual, or institutional obligation to share AI advisory systems trained on this data with farming communities, ensure systems are available in local languages, or compensate farming communities for the value their knowledge contributes. The three largest repositories of smallholder agricultural knowledge in India are owned by corporations headquartered in California with no accountability to Indian farming communities.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eData Justice analysis\u003c/strong\u003e \u003cp\u003eTaylor (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) argues that data justice requires the ability to use data systems as tools for self-determination. The current structure of platform data governance systematically denies this capacity. Farmers can generate data through commercial platforms but cannot govern it, direct its use, or benefit from its AI applications. This is a political economy problem \u0026mdash; a distribution of power over agricultural knowledge that favours technology corporations at the expense of farming communities \u0026mdash; and it requires political and institutional solutions, not technical ones.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.5 Institutional Accountability: The Governance Vacuum\u003c/h2\u003e \u003cp\u003eThe fifth governance failure is institutional: the absence of accountable public institutions with both the mandate and the capacity to govern agricultural AI in the interests of farming communities. India's Digital Agriculture Mission creates a national digital agriculture infrastructure but does not, in its current design, include governance provisions for farmer-generated data, mandatory language requirements for AI advisory systems, independent bias auditing, or farmer representation in AI system governance. National agricultural research institutions (ICAR, SAUs) have institutional legitimacy but lack technical AI capacity. Technology corporations have technical capacity but lack accountability to farming communities. The governance vacuum between these two positions is precisely where data extraction becomes structurally inevitable.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Towards Responsible Agricultural AI: Governance Conditions","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Community Data Trusts\u003c/h2\u003e \u003cp\u003eThe foundational governance requirement is a legal institutional arrangement under which farmer-generated agricultural social media data is held collectively on behalf of the farming communities that generate it. Community data trusts \u0026mdash; legal entities that hold data, negotiate the terms of its use, and ensure value flows back to member communities \u0026mdash; have been proposed in various forms in the data governance literature (Hardinges et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). For smallholder agricultural data in India, such trusts could be constituted at the Farmer Producer Organisation (FPO) level, with oversight by state agricultural departments, and with mandatory representation of female farmers and marginal landholders in governance structures. India's proposed National Data Governance Framework provides a potential legislative foundation, though its current architecture does not specifically address community agricultural data.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Participatory Consent Mechanisms\u003c/h2\u003e \u003cp\u003eMeaningful informed consent for AI use of farmer-generated agricultural data cannot be achieved through platform terms of service. It requires community-level consent processes mediated by trusted institutional actors. Agricultural extension officers \u0026mdash; identified in the survey as the single most powerful predictor of digital platform adoption, and as the primary quality-assurance mechanism for digital agricultural information \u0026mdash; are the most feasible mechanism for community-level consent mediation. Extension officers who understand and can explain data governance arrangements in local languages are the practical bridge between abstract responsible AI principles and the operational consent of farming communities.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e5.3 Language Justice Requirements\u003c/h2\u003e \u003cp\u003eAny AI agricultural advisory system claiming to serve smallholder farmers in northern India must operate in Hindi and in the major regional dialects \u0026mdash; Awadhi, Bhojpuri, Maithili \u0026mdash; that are the primary agricultural communication languages of the region. This requires deliberate training data curation ensuring dialect-language content is adequately represented, investment in natural language processing for low-resource Indian languages, and quality evaluation of AI advisory outputs conducted in target languages by representative farmer users. Output accessibility requires voice-based interfaces and visual formats for users with low literacy \u0026mdash; the 23.1% of survey respondents with no formal education cannot access text-based AI advisory outputs regardless of language.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e5.4 Algorithmic Bias Auditing and Accountability\u003c/h2\u003e \u003cp\u003eAI agricultural advisory systems must be subject to mandatory regular bias auditing \u0026mdash; systematic evaluation of whether recommendations perform differently across the demographic subgroups the system is intended to serve, with specific attention to gender and literacy performance differentials. Results should be publicly reported and should trigger mandatory system corrections where performance gaps exceed defined thresholds. Independent bias auditing should be conducted by institutions without financial interest in the AI system \u0026mdash; national agricultural universities are a natural candidate in the Indian context.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e5.5 Public Institutional Architecture\u003c/h2\u003e \u003cp\u003eThe governance conditions above imply a specific institutional architecture. Commercial technology corporations cannot be the primary developers and operators of AI agricultural advisory systems in the smallholder context \u0026mdash; their incentive structures are misaligned with community data ownership, language justice, and bias accountability. The most viable model is a public-private partnership governed by farmer representation: AI systems developed by technical partners under contractual governance arrangements that mandate community ownership, language requirements, bias auditing, and data benefit sharing. India's Digital Agriculture Mission, with its ₹2,817 crore investment, has the resources to establish this architecture \u0026mdash; if governance design is prioritised alongside infrastructure development.\u003c/p\u003e \u003c/div\u003e"},{"header":"6. Discussion","content":"\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e6.1 From Data Extraction to Knowledge Amplification\u003c/h2\u003e \u003cp\u003eThe central argument of this paper is that the choice facing agricultural AI development in the Global South is not between using farmer-generated data and not \u0026mdash; that decision has effectively been made by the commercial platforms that host and legally own it. The choice is between extractive use and responsible use: between AI systems that appropriate smallholder knowledge to generate value for technology corporations, and AI systems that amplify smallholder knowledge to generate value for farming communities.\u003c/p\u003e \u003cp\u003eThe Prayagraj survey documents a farming community that has built, without institutional support, a functioning digital agricultural knowledge system: peer diagnostic networks, distributed price discovery mechanisms, community-validated information flows. That system generates data daily that could train AI advisory capabilities superior to anything currently available to these farmers. An AI diagnostic system trained on farmer-generated WhatsApp image-text pairs from Prayagraj would be substantially better calibrated for Prayagraj farming conditions than any model trained on research-station data. A real-time price advisory system trained on the hyperlocal price-sharing data generated in farmer WhatsApp groups would provide information quality that formal price information services cannot match.\u003c/p\u003e \u003cp\u003eThe governance question is whether agricultural AI development will be designed to serve the communities that generate this knowledge, or to extract it from them. The current trajectory \u0026mdash; commercial platform ownership of data, public investment without governance design, AI systems built for institutional rather than farmer users \u0026mdash; points toward extraction. The responsible AI and data justice frameworks provide the tools to design a different trajectory.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e6.2 Implications for India's Digital Agriculture Mission\u003c/h2\u003e \u003cp\u003eIndia's Digital Agriculture Mission represents an opportunity to establish the governance architecture for responsible agricultural AI at national scale \u0026mdash; and a risk of institutionalising extractive AI development through inadequate governance design. The Mission's current architecture creates the infrastructure conditions for AI agricultural advisory systems but does not include the governance provisions identified in this paper: farmer data ownership frameworks, language justice requirements, independent bias auditing, or farmer representation in AI governance.\u003c/p\u003e \u003cp\u003eWithout these provisions, the Mission risks producing AI advisory systems that amplify existing agricultural inequalities rather than correcting them \u0026mdash; performing best for the already-connected, already-educated, already-networked farmers who need them least, while remaining inaccessible and poorly calibrated for the most marginalised farming communities. The governance architecture proposed in this paper \u0026mdash; community data trusts, participatory consent, language requirements, bias auditing \u0026mdash; is not an add-on to be designed after AI systems are built. It is the precondition for those systems having a legitimate claim to serve development rather than extraction.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e6.3 Limitations\u003c/h2\u003e \u003cp\u003eFour limitations merit acknowledgement. First, the characterisation of farmer-generated data types relies on survey self-reports and qualitative accounts rather than direct analysis of social media content \u0026mdash; platform access restrictions prevent systematic content analysis, a methodological constraint affecting all research in this area. Second, the governance conditions proposed in Section \u003cspan refid=\"Sec15\" class=\"InternalRef\"\u003e5\u003c/span\u003e are normative derivations from the Responsible AI and Data Justice frameworks and require validation through participatory design processes with farming communities. Third, the analysis is geographically bounded to Prayagraj district; governance requirements in other South Asian or sub-Saharan African smallholder contexts may differ. Fourth, the rapidly evolving landscape of AI regulation in India means specific institutional recommendations may require updating as the National Data Governance Framework and Digital Agriculture Mission architecture develop.\u003c/p\u003e \u003c/div\u003e"},{"header":"7. Conclusion","content":"\u003cp\u003eThe farmers of Prayagraj are generating an agricultural knowledge commons. Without knowing it, without owning it, and without governance frameworks that protect it, they are daily producing the diagnostic data, price intelligence, and community knowledge that could train AI advisory systems more useful and more locally calibrated than any agricultural AI system currently available. The question this paper has asked is not whether that knowledge has value \u0026mdash; it clearly does. The question is whether the AI systems that could be trained on it will be built to serve the communities that generated it.\u003c/p\u003e \u003cp\u003eThe Responsible AI and Data Justice frameworks converge on a clear verdict about the current trajectory: systematic extraction is more likely than responsible amplification, because the governance conditions for responsible use do not exist. Consent is absent. Community ownership is absent. Language justice is absent. Bias accountability is absent. Institutional architecture that aligns AI development with farming community interests rather than technology corporate extraction does not yet exist.\u003c/p\u003e \u003cp\u003eBuilding these governance conditions is not a peripheral task to be completed after AI systems are deployed. It is the ethical precondition on which any claim that agricultural AI serves development \u0026mdash; rather than replicating the colonial extraction of knowledge from the Global South for the benefit of the Global North \u0026mdash; ultimately depends. India's Digital Agriculture Mission, and the broader international community of agricultural AI development, faces a governance choice. This paper has tried to make its terms clear.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eThis study was conducted as part of the author’s academic dissertation at Sam Higginbottom University of Agriculture, Technology and Sciences (SHUATS), Prayagraj, India, and was reviewed and approved by the university’s academic evaluation committee as part of the degree requirements. The research was carried out under the supervision of designated faculty members in accordance with institutional academic and ethical guidelines for research involving human participants. All participants were informed about the purpose of the study, and their voluntary informed consent was obtained prior to participation. Data were collected anonymously, and confidentiality of respondents was strictly maintained throughout the research process.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eT.D. conceptualised the study, designed the research methodology, collected and analysed the data, developed the theoretical framework, and wrote the manuscript in its entirety. The author read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe author thanks the 720 farming households of Prayagraj district who participated in this research, and the block development officers of Uttar Pradesh who facilitated fieldwork. The author used Claude (Anthropic, claude-sonnet-4-6, February 2026) for language refinement and formatting assistance only. All research design, data collection, analysis, and writing are the author's sole responsibility.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe survey data that support the findings of this study are available from the corresponding author upon reasonable request. Data are not publicly deposited due to informed consent restrictions protecting participant confidentiality.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBuolamwini, J., \u0026amp; Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research 81:1\u0026ndash;15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCouldry, N., \u0026amp; Mejias, U. A. (2019). \u003cem\u003eThe costs of connection: How data is colonizing human life and appropriating it for capitalism\u003c/em\u003e. Stanford University Press.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuncombe, R. (2016). Mobile phones for agricultural and rural development: A literature review and suggestions for future research. \u003cem\u003eEuropean Journal of Development Research\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e(2), 213\u0026ndash;235. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/ejdr.2015.29\u003c/span\u003e\u003cspan address=\"10.1057/ejdr.2015.29\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFAO. (2022). \u003cem\u003eArtificial intelligence in the agrifood sector: Applications, risks and governance\u003c/em\u003e. Food and Agriculture Organisation of the United Nations.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFloridi, L., Cowls, J., Beltrametti, M., et al. (2018). An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations. \u003cem\u003eMinds and Machines\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e(4), 689\u0026ndash;707. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11023-018-9482-5\u003c/span\u003e\u003cspan address=\"10.1007/s11023-018-9482-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHardinges, J., Tennison, J., Shadbolt, N., et al. (2019). \u003cem\u003eData trusts: From theory to practice\u003c/em\u003e. Open Data Institute.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeeks, R. (2020). ICT4D 3.0? Part 1 \u0026mdash; The components of an emerging fourth-generation paradigm. \u003cem\u003eElectronic Journal of Information Systems in Developing Countries\u003c/em\u003e, \u003cem\u003e86\u003c/em\u003e(3), e12124. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/isd2.12124\u003c/span\u003e\u003cspan address=\"10.1002/isd2.12124\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJensen, R. (2007). The digital provide: Information (technology), market performance, and welfare in the south Indian fisheries sector. \u003cem\u003eQuarterly Journal of Economics\u003c/em\u003e, \u003cem\u003e122\u003c/em\u003e(3), 879\u0026ndash;924. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1162/qjec.122.3.879\u003c/span\u003e\u003cspan address=\"10.1162/qjec.122.3.879\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJobin, A., Ienca, M., \u0026amp; Vayena, E. (2019). The global landscape of AI ethics guidelines. \u003cem\u003eNature Machine Intelligence\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e(9), 389\u0026ndash;399. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s42256-019-0088-2\u003c/span\u003e\u003cspan address=\"10.1038/s42256-019-0088-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLioutas, E. D., \u0026amp; Charatsari, C. (2020). Smart farming and short food supply chains: Are they compatible? \u003cem\u003eLand Use Policy\u003c/em\u003e, \u003cem\u003e94\u003c/em\u003e, 104541. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.landusepol.2020.104541\u003c/span\u003e\u003cspan address=\"10.1016/j.landusepol.2020.104541\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMilan, S., \u0026amp; Trer\u0026eacute;, E. (2019). Big data from the South(s): Beyond data universalism. \u003cem\u003eTelevision \u0026amp; New Media\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(4), 319\u0026ndash;335. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/1527476419837739\u003c/span\u003e\u003cspan address=\"10.1177/1527476419837739\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., \u0026amp; Floridi, L. (2016). The ethics of algorithms: Mapping the debate. \u003cem\u003eBig Data \u0026amp; Society\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e(2), 1\u0026ndash;21. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/2053951716679679\u003c/span\u003e\u003cspan address=\"10.1177/2053951716679679\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMooney, P. (2018). \u003cem\u003eBlocking the chain: Industrial food chain concentration, big data platforms and food sovereignty solutions\u003c/em\u003e. ETC Group.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePringle, R. (2019). Five ways AI could improve the world. \u003cem\u003eNature\u003c/em\u003e, \u003cem\u003e570\u003c/em\u003e(7761), 151. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/d41586-019-01831-7\u003c/span\u003e\u003cspan address=\"10.1038/d41586-019-01831-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahwan, I. (2018). Society-in-the-loop: Programming the algorithmic social contract. \u003cem\u003eEthics and Information Technology\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(1), 5\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10676-017-9430-8\u003c/span\u003e\u003cspan address=\"10.1007/s10676-017-9430-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSharma, R., Kamble, S. S., Gunasekaran, A., Kumar, V., \u0026amp; Kumar, A. (2020). A systematic literature review on machine learning applications for sustainable agriculture supply chain performance. \u003cem\u003eComputers \u0026amp; Operations Research\u003c/em\u003e, \u003cem\u003e119\u003c/em\u003e, 104926. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cor.2020.104926\u003c/span\u003e\u003cspan address=\"10.1016/j.cor.2020.104926\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaylor, L. (2017). What is data justice? The case for connecting digital rights and freedoms globally. \u003cem\u003eBig Data \u0026amp; Society\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e(2), 1\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/2053951717736335\u003c/span\u003e\u003cspan address=\"10.1177/2053951717736335\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVinuesa, R., Azizpour, H., Leite, I., et al. (2020). The role of artificial intelligence in achieving the Sustainable Development Goals. \u003cem\u003eNature Communications\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(1), 1\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41467-019-14108-y\u003c/span\u003e\u003cspan address=\"10.1038/s41467-019-14108-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Bank. (2022). \u003cem\u003eFuture of food: Harnessing digital technologies to improve food system outcomes\u003c/em\u003e. World Bank.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZuboff, S. (2019). \u003cem\u003eThe age of surveillance capitalism: The fight for a human future at the new frontier of power\u003c/em\u003e. PublicAffairs.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Responsible AI, Data Justice, smallholder farmers, agricultural advisory systems, ICT4D, India","lastPublishedDoi":"10.21203/rs.3.rs-8986304/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8986304/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eSmallholder farmers in developing countries generate rich, structured agricultural knowledge through social media platforms every day \u0026mdash; crop diagnostic descriptions, peer-validated price information, technique queries, and market intelligence \u0026mdash; without knowing they are doing so, without owning what they produce, and without governance frameworks that protect their rights over this data or ensure they benefit from its potential use in artificial intelligence systems. This paper examines farmer-generated social media data from an original survey of 720 smallholder farmers across all 20 administrative blocks of Prayagraj district, Uttar Pradesh, India, and applies the Responsible AI and Data Justice frameworks to evaluate the ethical conditions under which AI agricultural advisory systems could legitimately build on this data. We find that the data farmers generate constitutes a structurally valuable resource for training localised AI advisory systems \u0026mdash; including labelled diagnostic image-text pairs, real-time hyperlocal price signals, and agricultural information demand data \u0026mdash; but that current governance conditions are systematically inadequate: genuine informed consent is absent, community data ownership does not exist, language representation is ignored, and algorithmic accountability mechanisms are non-existent. We argue that the choice for agricultural AI development in the Global South is not between using farmer-generated data or not, but between using it extractively and using it responsibly. Realising responsible AI agricultural advisory requires not technical solutions but governance ones: community data trusts, participatory consent mechanisms, language justice requirements, algorithmic bias auditing, and public institutional architecture that aligns AI development with farming community interests rather than technology corporate extraction.\u003c/p\u003e","manuscriptTitle":"From Social Media to Smart Advisory: Responsible AI, Data Justice, and the Governance of Farmer-Generated Knowledge in Smallholder Agriculture","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-22 10:44:58","doi":"10.21203/rs.3.rs-8986304/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"65b0e58c-5f45-4548-b0c0-915110713e2a","owner":[],"postedDate":"April 22nd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-22T10:44:58+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-22 10:44:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8986304","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8986304","identity":"rs-8986304","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}