Should farmers embrace AI?

The writer is the chief economist of the UN Food and Agriculture Organization in Rome

Pope Leo XIV is from Chicago, but he might as well be from rural Iowa. He made a prescient observation in his recent encyclical on AI: digital technology should be assessed according to whether it ensures fair access to opportunities, speaking directly to farmers everywhere without ever mentioning agriculture.

The common concern is that AI will produce goods and services with less labour, leaving us all with fewer jobs. But agriculture has different challenges, including ageing farmers, skills gaps and low labour productivity.

AI could boost industry productivity; the question is which farms it would do this for. It could make large, highly mechanised farms more efficient. But smaller operations that cannot afford expensive machinery, high-speed internet or subscriptions to AI services may fall even further behind.

This matters because smallholders (with less than five acres) make up 85 per cent of the world’s farms. In poorer countries, they are crucial for food security, producing 60 per cent of food crops.

Agriculture has some of the worst information gaps of any sector. Farmers often have to make big, long-term decisions about everything from seeds to market prices with scant information. This is compounded by climate and geopolitical shocks that can rapidly disrupt growing conditions and affect fuel and fertiliser prices. The gap affects smallholders the most.

In some places, “extension workers” — agricultural advisers who help farmers adopt new practices — face the impossible task of reaching millions of producers across vast areas.

AI may prove a godsend in this case. It could enable advisers to support more farmers and narrow the information gap by translating information into local languages, integrating weather, soil and market data, identifying crop diseases and improving price transparency.

A 2026 study of 240 agricultural advisers in Benin, for example, found that those using generative AI gave better advice. In India, an AI tool that can tell cotton farmers, in eight Indian languages, when and how much pesticide to apply has been used by more than 30,000 of them, with revenue gains of up to 24 per cent. In a trial in China, an AI system using weather and soil data and computer vision to guide irrigation and pesticide use increased wheat and maize yields by 7 and 4 per cent respectively.

But, alongside this promise, the Pope rightly sounds a warning: technology can create new forms of exclusion.

An AI lending model may deny smallholders a loan or charge higher interest because they lack financial records used to assess risk. A subsidy system may exclude farmers without formal land titles. AI systems trained on large farms may not suit small plots.

There is also evidence AI tools can be biased. For instance, some systems were found to systematically favour commercial products for protecting crops over sustainable biodiversity-based alternatives, reinforcing farmers’ dependence on products they must buy rather than practices they can adopt for free and are more beneficial to nature.

So how can farmers of all sizes use AI to become more productive?

By sticking to a simple principle: the higher the potential for harm, the greater the need for caution. An AI tool that identifies a crop disease presents relatively limited risks and should be easy to test and deploy. The stakes are higher for systems that help determine if a farmer receives a loan or insurance, or makes decisions affecting harvests or food safety. Those applications should have strict requirements for testing, human oversight and transparency. And farmers should be able to challenge decisions and seek redress.

At the highest risk levels, safeguards should be backed by binding legal agreements. This matters as much for farmers as for global AI safety and could establish common obligations for the largest AI companies, including independent safety audits and safeguards against discrimination or abuse of market power.

Regulation alone won’t get AI on to a farmer’s phone in Benin. Governments will still need to invest in connectivity, digital skills, data and advisory services. But it can help prevent a technology with enormous potential from widening the gap between those with access and those without.

Some 44 per cent of young people depend on agrifood systems for their livelihoods. To thrive, much less survive, they’ll need the help of AI. The challenge is to ensure it doesn’t create a lost generation of excluded people.