AI’s Big Promise for Agriculture Is Stuck on Trust, Cost and Control: Here's why

A new systematic review finds that the future of AI-enabled agriculture depends less on technical capability alone and more on trust, data governance and digital access.

The review, published in Frontiers in Artificial Intelligence, examines 38 peer-reviewed studies on artificial intelligence and machine learning in agriculture published between 2018 and August 2025. The authors, Richa Bhattarai, Jennifer Koch, Wolfgang Jentner, Mehreen Habib, Michael C. Wimberly and David Ebert, analyzed the literature through three lenses: the scale of farming operations, farmers' trust in AI and machine learning systems, and the digital divide.

The findings assert that AI and machine learning can support more efficient, productive and sustainable agriculture, but adoption remains constrained by weak trust, unequal digital access, poor explainability, uncertain reliability, high costs and unresolved questions over who controls farm data.

The Promise Is Huge but Adoption Is Not Automatic

The world needs more food, feed and fiber, but farmers must produce it under rising climate stress, water constraints, soil degradation, pest pressure and market volatility. AI and machine learning promise to help manage that complexity by turning large volumes of data into actionable advice.

These technologies can support crop monitoring, disease detection, irrigation scheduling, weed management, yield prediction, input optimization and weather-risk planning. Sensors, drones, satellites, smartphones and farm-management platforms are creating more data than farmers can process manually. AI systems are designed to find patterns in that data and support faster, more precise decisions.

The potential is directly relevant to the Sustainable Development Goals, especially SDG 2 on Zero Hunger, SDG 9 on innovation and infrastructure, SDG 12 on responsible production, and SDG 13 on climate action. Better agricultural intelligence could help farmers reduce input waste, improve resource efficiency and adapt to climate shocks.

The review makes clear that technical promise does not equal farmer adoption. Farmers operate in high-risk environments where bad advice can have real economic consequences. A wrong irrigation recommendation, an inaccurate pest alert or a misleading yield forecast can damage crops and livelihoods. For many farmers, especially smallholders and medium-scale producers, the cost of experimenting with unreliable technology is simply too high.

Farmers are not rejecting innovation by default. The review finds that farmers across scales show interest in AI when tools clearly reduce risk, improve yields or solve immediate operational problems. But they are cautious when systems are opaque, expensive, poorly explained or disconnected from local knowledge.

The Trust Problem Starts With Data

The review identifies data governance as the most frequently discussed barrier to trustworthy agricultural AI. Twenty-one of the 38 studies cite concerns over data privacy, ownership, access, licensing, cybersecurity, benefit sharing and control.

Farm data is not neutral. It can include crop yields, soil conditions, pesticide and fertilizer use, water management practices, geolocation information, financial records and insurance details. In the hands of technology providers, agribusinesses, lenders, insurers or regulators, such data can carry commercial and strategic value.

Farmers worry that data collected from their fields may be used without meaningful consent, sold to third parties, used to strengthen corporate bargaining power or even support surveillance and compliance monitoring. This concern is especially serious where farmers face opaque contracts or lack the legal and technical capacity to understand how data will be used.

The review highlights a structural imbalance: many AI systems are developed by large technology or agribusiness actors, while farmers, particularly smallholders, may have limited bargaining power. This creates a risk that farmers become data suppliers without receiving fair control over, or benefit from, the value created.

For the Global South, the issue is even sharper. Smallholder farmers often face weak digital infrastructure, limited access to legal advice and lower ability to negotiate platform terms. If digital agriculture expands without strong data rights, it could reproduce old inequalities in a new form: farmers provide the data, while platforms capture the value.

A trustworthy AI agenda for agriculture therefore cannot stop at algorithmic transparency. It must include clear rules on data ownership, consent, access, benefit sharing and accountability. Farmers need to know not only what a tool recommends, but also what information it collects and who profits from it.

The Digital Divide Could Decide Who Benefits

The review finds that the digital divide is one of the most persistent barriers to AI adoption in agriculture. Twenty-nine of the 38 papers mention the digital divide, and 25 describe it as high. None describe it as low. This divide is not only about internet access, though connectivity is crucial. It also includes access to devices, sensors, platforms, technical support, digital literacy, data skills and affordable services. Farmers cannot benefit from AI tools if they lack reliable broadband, cannot afford subscriptions, or do not have the training needed to interpret model outputs.

Large farms often generate more data, use more advanced machinery and have better access to capital. AI tools trained on data from large, well-connected farms may perform better in those contexts. Smaller and more diverse farms, by contrast, may be underrepresented in the datasets used to build agricultural AI systems. This creates a feedback loop. Large farms generate more data, receive better tools, adopt faster and gain more benefits. Smaller farms generate less data, receive less suitable tools and fall further behind. If left unaddressed, AI could widen the productivity gap between large commercial farms and smaller producers.

The review also points to intermediate-scale farms as an overlooked opportunity. These farms are neither subsistence smallholdings nor highly capitalized industrial operations. Yet a large share of global agriculture falls into this middle category. If AI tools are designed only for large farms, the technology may miss one of the most important adoption markets in global agriculture.

For development agencies and governments, this finding is critical. Digital agriculture policy should not focus only on frontier technologies. It must also build the foundations: rural connectivity, farmer training, locally relevant advisory systems, affordable financing and public-interest data infrastructure.

Explainability, Reliability and Cost Will Shape the Future

Farmers need AI systems that explain themselves. The review identifies lack of explainability as a major barrier. In agriculture, a recommendation is more likely to be trusted when farmers understand why it was made and how it relates to local conditions. Farmers rely on experience, field observation, seasonal patterns and local ecological knowledge. If an AI system recommends an action that contradicts what a farmer sees on the ground, the system must be able to explain its reasoning in practical language.

Black-box AI is a poor fit for high-stakes agricultural decisions. Farmers need tools that communicate uncertainty, show the data behind recommendations and allow users to compare algorithmic advice with field realities. Reliability is equally important. Agricultural systems are variable by nature. Soil, weather, pests, water availability and crop conditions can change across short distances and from season to season. A tool that works in one region may fail in another. The review finds that trust declines when AI-generated advice conflicts with observable field conditions or lacks local validation.

Cost is another major barrier. AI-enabled agriculture may require sensors, software, subscriptions, maintenance, data storage, hardware upgrades and training. Large farms can spread these costs over more land and production volume. Small and medium-scale farms face tighter margins and greater uncertainty over returns.

On the whole, AI could help farmers manage climate risk, improve yields, reduce input waste and strengthen food security. But the review shows that the benefits will not be distributed automatically. They will depend on who has access to digital infrastructure, who controls agricultural data, whose knowledge is built into AI systems and whether farmers can trust the recommendations they receive.

The paper also outlines a practical roadmap for building trust in agricultural AI:

• Governments should create enabling conditions for trustworthy digital agriculture, which means stronger data governance, rural broadband investment, farmer-centered training, transparent procurement standards and support for tools that work across farm scales.
• Development agencies: The study calls for inclusive AI strategies that prioritize smallholders and intermediate-scale farms, especially in low- and middle-income countries. AI should not become another technology that benefits the already connected while leaving vulnerable producers behind.
• Agritech companies should treat trust as a core design principle. Farmers need tools that are explainable, reliable, affordable and transparent about data use. Building trust is not a public relations exercise; it is essential to adoption.
• More field-based research is needed in the Global South, along with studies that test AI systems across different crops, climates, farm sizes and socio-economic conditions. Future work should also examine cooperative data models, open-source tools, farmer data rights and the labor impacts of digital agriculture.

The future of AI in farming will not be decided by algorithms alone. It will be decided by whether farmers see these tools as partners or threats.