Smallholder farms — often less than two hectares — produce a large share of the food in many low- and middle-income countries, yet their farmers often have limited access to agronomists, credit and market information. Smartphones are far more widespread than tractors in these communities, making mobile AI one of the most promising ways to reach them.

Crop disease diagnosis

A 2016 study by researchers including Sharada Mohanty, David Hughes and Marcel Salathé trained deep learning models on a public dataset of tens of thousands of images of healthy and diseased plant leaves from the PlantVillage project, achieving high accuracy in identifying crop–disease combinations under test conditions. The work showed that smartphone-based diagnosis was feasible.

Since then, apps have been developed to let farmers photograph a leaf and receive a likely diagnosis and treatment advice — some working offline on the phone itself. Real-world accuracy is lower than in controlled tests, because field photos vary in lighting, angle and background, so good tools combine AI with extension workers and local expertise.

Advice in local languages

Language models and speech technology make it possible to deliver agricultural advice by voice and in local languages, through apps, chat services or phone lines. This matters where literacy is limited or where agricultural information exists only in a few dominant languages. Projects combine AI with verified agronomic content from research institutes to avoid incorrect advice.

Weather and climate information

Localized forecasts and alerts about rainfall, heat and pests help farmers decide when to plant, irrigate, spray and harvest. AI-based weather models can improve forecasts in regions with few weather stations.

Markets and finance

  • price information for crops in nearby markets;
  • connecting farmers with buyers;
  • alternative credit scoring using mobile and farm data, enabling loans for farmers without formal credit history;
  • crop insurance triggered by satellite-observed weather events.

The risks

  • Wrong advice can destroy a harvest; tools must be tested locally and validated by agronomists.
  • Data exploitation: farmers’ data may be used by companies without fair benefit.
  • Exclusion: women farmers, older farmers and those without smartphones or connectivity may be left out.
  • Credit risks: algorithmic lending can lead to over-indebtedness if not carefully designed.
  • Dependency on platforms that may change terms or disappear.

What works

Experience from development programs suggests successful projects:

  1. start from farmers’ actual problems, not technology;
  2. work through trusted local organizations and extension services;
  3. support local languages and voice;
  4. function with poor connectivity;
  5. measure real outcomes such as yields, income and losses — not app downloads.

A day with a digital advisory service

Consider a smallholder maize farmer who notices yellowing leaves. With a phone-based service, she might:

  1. photograph affected leaves in a voice-guided app that works in her language;
  2. receive a likely diagnosis — for example a nutrient deficiency or a specific disease — with a confidence level;
  3. get advice on low-cost treatment options suitable for her farm, drawn from validated agronomic content;
  4. receive an alert that heavy rain is forecast, suggesting she delay applying fertilizer;
  5. check prices in nearby markets before selling her harvest.

If the diagnosis is uncertain, the service can connect her with an extension worker, who receives the photo and the app’s assessment. The combination of AI and human expertise is more reliable than either alone.

Designing for low connectivity

Effective tools compress models to run on inexpensive phones, sync data when connectivity is available, support SMS or voice calls for basic phones, and use little mobile data, which can be expensive for farmers.

What to watch next

  • Voice-based agricultural assistants in more local languages.
  • Independent evaluations of AI advisory services on farm incomes.
  • AI weather forecasting for regions with sparse observation networks.
  • Data protection rules for farmers in digital agriculture.

Key takeaways

  • Smartphones make mobile AI a practical way to reach smallholder farmers.
  • A 2016 PlantVillage study showed deep learning could identify crop diseases from leaf images.
  • Voice and local-language advice can widen access to agronomic knowledge.
  • Weather, market and credit services complement diagnosis tools.
  • Local validation, data fairness and inclusion determine real impact.

Frequently asked questions

Do these apps work without internet?

Some run AI models directly on the phone and work offline; others need connectivity for full features.

How accurate is phone-based disease diagnosis?

Accuracy in the field is generally lower than in lab tests. It works best combined with expert advice.

Who builds these tools?

Universities, nonprofits, governments, agricultural companies and startups, often in partnership with extension services.