<aside> ℹ️ This presentation features at PAIRS 2026 in New Delhi on 18th February 2026 at 11:00 IST
🎙️ Thread on PAIRS Discussion Server (Discord) (register first)
</aside>
Participatory and human-centered approaches have become central to the discourse on responsible AI, premised on the idea that involving end-users will better align technological outputs with local needs and contexts. Yet in the agriculture and development (AgDev) sector, there is a widening disconnect between the limited benefits accruing to farmers and the rhetoric used by corporate and development actors to justify broader investment in AI. The language of inclusion and empowerment is often coopted to legitimise the rapid scaling of systems promoted as solutions to climate stress, food insecurity, and smallholder vulnerability. In practice, however, participation often remains tokenistic—amounting to a form of “participation washing.” Farmers’ fields are repurposed as experimental sites for collecting training data and stress-testing new technologies, while narratives of “digital divides” dominate design agendas, perpetuating a deficit model that overlooks existing local practices and capacities. Rather than shifting power or epistemic authority, such initiatives reproduce extractive dynamics, instrumentalising participation in service of commercial ends.
This paper presents a systematic scoping review of literature examining how participatory and human-centered AI approaches in the AgDev sector are framed—the design principles they articulate, the purposes they serve, and the conditions under which they are imagined to operate. Drawing on the authors’ experience with agricultural innovation and digital extension in the Global South, the paper develops a reflective synthesis of participatory AI along two dimensions. First, the quality of participation is assessed using Arnstein’s ladder of civic participation to distinguish token consultation from shared decision-making. Second, the scope of participation is evaluated through Gloria Miller’s four-factor test—power, legitimacy, urgency, and harm—to assess whose interests and capacities shape engagement across the AI lifecycle.
A preliminary review (see Appendix) reveals that participation is diversely conceptualized and inconsistently realized in AgDev literature. Forms of engagement range from individual user feedback to social learning systems, institutionalised multi-stakeholder collaboration, and participatory assurance frameworks involving farmers and policymakers. Ethical orientations likewise vary—from procedural inclusion to transformative learning, co-governance, and emerging notions of accountability and explainability. Across these cases, infrastructure and design choices are decisive: participation depends less on digital access than on how systems accommodate diversity, trust, and local capacity. All four studies position “feedback” as a central design principle—whether between users and systems, farmers and institutions, or developers and policymakers—yet few evolve into genuine co-design or shared oversight. The limited integration of participatory assurance and governance mechanisms underscores the need for frameworks that more effectively connect technical validation with social accountability in AgDev AI.
The discussion explores what forms of participation are feasible and desirable within the ethical and institutional realities of resource-constrained and power-unequal contexts. Rather than prescribing a fixed model, it asks how participatory AI might move beyond inclusion as input toward shared agency in design and evaluation. The aim is to inform future frameworks for responsible innovation that treat participation as an evolving process of co-production and accountability, attentive to both the epistemic and political dimensions of AI in AgDev.