For a farmer waiting for rain, uncertainty is never an abstraction. Seeds may already have been bought, the soil prepared, and a loan repayment date fixed, even as the seasons become harder to read. An AI service might help with a forecast or a planting decision. Yet if its advice is wrong, the loss will be felt in the household long before it appears in any assessment of model performance.
That tension followed me through a program held on 26–30 October 2026 in Bandung on climate-resilient smallholder livelihoods. I am affiliated with the Non-Aligned Movement Centre for South-South Technical Cooperation (NAM CSSTC), which organized the program with the Centre on Integrated Rural Development for Asia and the Pacific (CIRDAP) and Indonesian partners. Participants approached rural challenges through different institutional experiences, often returning to the everyday limits imposed by water, finance, and access to markets. What stayed with me was the distance between proposing an intervention and building the institutional capacity to sustain it. Although technology was only one part of the discussions, the experience raised a broader question for agricultural AI: who ensures that farming communities can question the systems introduced to support them?
The question has become more urgent amid overlapping shocks. Summarizing the Mid-Year Update of the Global Report on Food Crises 2026, the European Commission’s Joint Research Centre warned in September that El Niño, the Middle East conflict, and reduced humanitarian funding could worsen acute food insecurity into 2027. An October 2026 OECD policy brief showed how volatility in energy-intensive fertilizer supply chains affects farm profitability and production decisions. Such pressures make better agricultural information more valuable while raising the consequences of relying on systems that cannot be adequately scrutinized.
Beyond Access to Agricultural AI
Stay ahead of the geopolitical week.
MD Briefing delivers expert analysis across five global fronts — the Indo-Pacific, energy, geoeconomics, European security, and the Middle East — every Monday morning. Free.
Agricultural AI is often presented as a last-mile challenge: bring advice within reach and measure how many farmers use it. Access matters enormously. The International Telecommunication Union estimates that 17 per cent of rural residents in low-income countries used the internet in 2026, compared with 41 per cent in urban areas. These modelled figures reveal a substantial divide even before language, device costs, and trust enter the picture. But a message delivered to a phone still has to make sense in a particular field, under conditions the model may barely understand.
Digital Green’s FarmerChat offers a glimpse of what more accessible advice can look like. Farmers can ask questions through voice or text and share photographs, opening possibilities for those less comfortable with conventional digital interfaces. A peer-reviewed case study found expanded access to personalized agricultural information, although exchanges remained predominantly individualized and one-way. The study included a Digital Green co-author and drew partly on implementation data; it did not establish causal improvements in crop yields. A June 2026 analysis by the Center for Global Development and Digital Green reported that, in one telecom partnership, waiving FarmerChat data charges was associated with 31 percent more queries; rigorous impact trials were still under way.
These findings are encouraging, but they invite a more demanding measure of inclusion. Farmers need advice informed by changing conditions and a practical way to say when it does not fit what they see. The quality of that relationship matters as much as the number of questions answered, particularly where an extension worker may be the only trusted bridge between scientific guidance and local experience.
Who Pays When Advice Fails?
For smallholders, a bad season can leave traces that last for years. The IPCC documents distress sales of productive assets after extreme events, which can weaken a household’s ability to recover. Crop losses may also add pressure to food prices and public budgets. In research based on international data from 1990–2011, Marc Bellemare found that higher food-price levels contributed to greater social unrest, although escalation depends on political and institutional conditions. These are risks to be managed, not inevitable outcomes of a failed harvest.
AI can improve decisions without removing the uncertainty of farming. It can also reshape who bears that uncertainty. One farmer may have the connectivity to act on an early warning while a neighbor does not; an inaccurate digital assessment could have wider consequences if it informs credit or insurance. When something goes wrong, a farmer may have no clear route to appeal, a public official may lack the technical records needed to investigate, and a provider may have met the formal requirements of its contract. Responsibility can become difficult to locate precisely when it matters most.
A 2026 scoping review by Michelle Ragany, Sjaak Wolfert, and Rozita Dara highlights unclear liability for adverse outcomes, including potential crop losses from ineffective AI recommendations. Establishing responsibility requires evidence of what a system advised and what happened on the ground. A disappointing harvest, by itself, cannot prove that an algorithm caused the damage.
The Price of Governing AI
The World Bank’s assessment of agricultural AI emphasizes skills and governance alongside technology. These are demanding investments for agricultural agencies already managing limited budgets and competing needs. Testing advice locally takes trained people. An extension worker must make time to hear a complaint; an agronomist may need to examine the field. Those hours are part of the true cost of a dependable service, and they continue after a donor-funded pilot has ended. Higher fertilizer or fuel costs can make spending on oversight harder to defend politically, even when it protects the value of the investment.
The deeper difficulty is a double asymmetry. Farmers may lack the technical means to assess a model and the influence to alter the terms on which its advice is delivered. Protocol power lies in seemingly mundane choices about data formats, portability, and complaint procedures. Technology developed abroad may strengthen local capability, while a domestic system can create its own form of lock-in. What matters is whether institutions can understand the service well enough to scrutinize it and retain the freedom to change course.
Practical accountability therefore needs a place in the budget. For higher-risk advisory services, public procurement could finance independent agronomic review and accessible complaints through existing extension networks. Providers could be contractually required to investigate substantiated technical failures, with remedies tied to evidence. Oversight should reflect the stakes: general cultivation information and advice affecting irrigation allocations or credit eligibility warrant different levels of scrutiny.
The test is whether errors are discoverable, decisions are contestable, and costs are assignable, whoever is at fault.
A Multiplex Ecosystem That Can Correct Itself
Imagine a drought in an Indonesian rice-growing community. An AI service recommends additional irrigation using soil readings but overlooks agreed rules for sharing a scarce water source. Indonesia’s Ministerial Regulation No. 30/PRT/M/2015 provides for farmer participation in irrigation management through water-user associations, or P3A. Article 4 recognizes that participation, while Article 9 sets out P3A rights and responsibilities concerning tertiary irrigation systems. The AI incident is hypothetical, but the community institution is real. A recommendation can be technically plausible and still unsettle the arrangements through which neighbors share water.
A multiplex digital ecosystem matters when that disagreement arises. Farmers could bring the concern through their association, with extension workers examining what the system missed. The provider would investigate its assumptions, while public authorities would ensure that the complaint leads to a response. Each party sees a different part of the problem. A vendor may seek consistent recommendations at manageable cost; a farming community must live with the consequences when local rules are overlooked. The capacity to listen across those differences and to make a correction stick is part of resilience itself.
Existing farmer organizations and research institutions offer a practical starting point, provided participation is supported rather than treated as unpaid labor. Smaller deployments can reveal where a system is useful and where its safeguards need strengthening. That learning is easier to preserve when duties and costs are agreed before a crisis, instead of negotiated after a family has suffered a loss.
South–South Cooperation as Governance Learning
The conversations in Bandung reminded me that similar ambitions can meet very different realities. A solution that looks affordable in one country may be difficult to maintain in another, even where the need is equally pressing. South–South exchanges can make these differences visible and help institutions recognize the questions they should ask before committing to a technology. There is value in sharing what has proved difficult as well as what has worked.
Cooperation could then take a modest but useful form: common methods for evaluating agricultural AI, or adaptable procedures for recording complaints and assigning responsibilities in technology contracts. Such tools would need to be tested with farming communities and fitted to national institutions. Their worth would lie in strengthening local judgement, rather than creating a uniform model for countries whose agricultural realities differ.
As another uncertain planting season approaches, the farmer waiting for rain still needs advice that works in the field and a way to challenge it when experience suggests otherwise. Climate resilience cannot rest on systems whose benefits are celebrated by institutions while their failures are quietly paid for by the people who grow the food.
