Where does authority reside in AI-enabled agricultural advisory systems? CGIAR’s partnership architecture as the knowledge layer for AI-enabled agricultural advice

Artificial intelligence (AI) is increasingly embedded in agricultural advisory systems and food-system decision making. While most AI-for-agriculture initiatives focus on downstream components, such as models, interfaces, and user engagement, these systems also govern which evidence is accessible, how data uncertainty and context are represented, and which advisory recommendations are permissible. As AI-enabled advisory systems are scaled, questions of authority, accountability, and legitimacy become central, particularly in food-insecure and climate vulnerable contexts. This white paper reframes where authority resides in AI-enabled agricultural advisory systems by identifying the knowledge layer, rather than AI interfaces, as the primary locus of decision-making power.
It further examines whether CGIAR’s partnership-based research architecture is institutionally positioned to steward this layer as public digital infrastructure. The study uses a conceptual and institutional analysis informed by developments in Retrieval-Augmented Generation (RAG), AI governance, and agricultural advisory systems. The analysis shows that in modular, retrieval-based AI systems, advisory outcomes are determined primarily upstream by governance of the knowledge layer, comprising decision-grade evidence, semantic structures, decision logic, and institutional safeguards. Control over this layer, rather than ownership of AI models or interfaces, shapes what AI systems can recommend. CGIAR’s partnership architecture has a comparative and paramount advantage in stewarding this layer as public digital infrastructure. Weakly governed knowledge layers risk delivering scaling advice that is generic, infeasible, or inequitable.
This perspective posits that the highest-leverage investment for AI in food security lies not in creating additional interfaces or models, but in building and governing the shared knowledge-layer infrastructure. Anchoring evidence standards, contextual validity, and accountability upstream will enable public institutions, such as CGIAR, to shape and steward AI-enabled agricultural advisory across platforms, while preserving innovation at the model and interface levels. This perspective advances three calls to action:
1. Researchers must prioritize the generation of datasets and scientific outputs that are explicitly structured, interoperable, and AI-ready, enabling integration into AI-enabled advisory decision support systems.
2. International (e.g., CGIAR) and national agricultural research institutions must reaffirm and strengthen their foundational role in producing rigorously validated evidence. In the emerging AI ecosystem, their comparative advantage lies in generating high-quality, site-specific, and context-aware data and knowledge that underpin reliable digital intelligence.
3. Policy- and decision makers should recognize the strategic importance of investing in governed knowledge infrastructures such as curated data, standards, and validation frameworks that form the knowledge layer of AI systems rather than focusing primarily on downstream artifacts such as chatbots, APIs, or other application interfaces.
This will have profound implications for AI governance, advisory sovereignty, and public investment priorities in agricultural systems worldwide.