Artificial intelligence in agriculture: what works today
AI in farming, clearly assessed: disease ID apps, spot sprayers, yield estimation, weather models, chatbots, livestock sensors, accuracy and limits.
On this page
Artificial intelligence (AI) is promoted as the next revolution in farming, with promises of robots that weed, drones that diagnose, chatbots that advise and models that predict harvests. Some of this is already working on millions of farms. Some is promising but unproven, and some is mostly marketing. For students, extension workers and farmers, the useful question is not "Is AI the future?" but "Which AI tools work today, how well, and for whom?"
In agriculture, AI mostly means machine learning: computer models that learn patterns from data, such as images, satellite readings, sensor streams or text, instead of following fixed rules. Deep learning with neural networks has made image and speech recognition practical. Large language models now power conversational advisory tools.
Applications that work at scale
Pest and disease identification
Smartphone apps such as Plantix and PlantVillage Nuru identify crop diseases, pests and nutrient deficiencies from photographs. They are trained on millions of labelled images, and are used by millions of farmers in India, Africa and elsewhere. Accuracy is high for common, well-photographed problems, and lower for rare problems, early symptoms, mixed infections or poor photographs.
Weed-detecting spot sprayers
Camera-equipped boom sprayers, such as "See & Spray" systems, recognise weeds in real time and spray only them. Manufacturers and trials report herbicide savings of roughly 50–75% or more in some fallow and row-crop situations. These systems are used on large farms in the USA, Australia and Brazil. Laser and mechanical robotic weeders are used in some vegetable crops.
Livestock monitoring
Neck collars, ear tags and pedometers use accelerometers and machine learning to detect oestrus (heat), rumination changes and early illness in dairy cattle. Cameras detect lameness or body condition. These systems are widely used on commercial dairy farms, and they are now being adapted for smaller Indian dairies.
Mapping from satellites
Machine learning classifies crop type, sown area, crop condition, burned fields and flood damage from satellite images. India uses such methods in crop area estimation, drought monitoring and crop residue burning surveillance.
Proven and scaling up
- Yield estimation: models combine satellite indices, weather and crop-cutting data to estimate yields at village or district level. Under PMFBY, the YES-TECH system (2023) gives technology-based yield estimates a weight in calculating insurance claims for rice and wheat.
- Weather prediction: AI weather models trained on decades of reanalysis data, such as Google DeepMind's GraphCast and ECMWF's AIFS, now match or beat traditional numerical models on many medium-range measures, at a fraction of the computing cost. National weather services are testing and adopting them.
- Advisory chatbots: large language model systems answer farmers' questions by voice or text in local languages, drawing on vetted extension content. India's Kisan e-Mitra handles PM-KISAN queries, and other pilots provide crop advice. Early evaluations are promising but highlight the risk of confident wrong answers.
- Breeding: machine learning supports genomic prediction, high-throughput phenotyping from drone images, and gene editing target design.
Still early or experimental
- Robotic fruit harvesting: picking soft fruits such as strawberries and apples without damage remains slow and costly, though prototypes are improving.
- Fully autonomous tractors: these are available in limited forms, but regulation, safety and cost restrict use.
- Price forecasting for farmers: markets are affected by policy and shocks that models handle poorly.
- Whole-farm autonomous decision-making: AI systems that plan an entire season are far from reliable.
How to judge an AI tool
A model's headline "accuracy" can hide important weaknesses. For a disease detector, for example, a few measures matter:
A disease app in the field
An app is tested on 1,000 leaf photos, of which 100 truly show blast. The app flags 120 photos as blast, and 90 of these are correct. Its precision is 90 ⁄ 120 = 75%: one in four alerts is a false alarm, which could lead to unnecessary spraying. Its recall is 90 ⁄ 100 = 90%: it misses 10 diseased plants. Of the 900 healthy photos, 870 are correctly cleared, so overall accuracy is (90 + 870) ⁄ 1,000 = 96%, which sounds excellent but hides the false alarms. That is why tools should be tested on local crops and conditions, and why advice should include confirmation steps.
Other questions to ask:
- Was it tested on data like mine? Models trained in one region or season often perform worse elsewhere. This is called dataset shift.
- What happens when it is wrong? A wrong pest identification that triggers an unnecessary spray costs money and harms the environment.
- Who owns the data? Farm data has value, so farmers need clear consent and benefit sharing.
- Does it work offline, in local languages, and on low-cost phones?
Risks and limits
- Bias and exclusion: tools built for large, connected farms may not serve smallholders, women farmers or those without smartphones.
- Hallucination: language models can give fluent but wrong or unsafe advice, such as incorrect pesticide doses. Grounding in vetted content and human review matter.
- Over-reliance: AI should support, not replace, field observation and extension workers.
- Energy and cost: large models need computing power, though most farm-level tools are light.
- Accountability: who is responsible when automated advice fails?
Key points
- AI in agriculture mainly means machine learning on images, satellite data, sensor streams and text.
- Working at scale: disease and pest identification apps, weed-detecting spot sprayers, livestock sensors and satellite crop mapping.
- Scaling up: yield estimation (YES-TECH under PMFBY), AI weather models, advisory chatbots and genomic prediction.
- Still early: robotic harvesting, fully autonomous machines and reliable price forecasts.
- Judge tools by precision and recall on local data, not headline accuracy alone.
- Risks include dataset shift, hallucination, exclusion, data ownership and accountability.
What the exams ask
- Give examples of AI applications in crops and livestock.
- Explain machine learning, training data and dataset shift in simple terms.
- Calculate precision and recall from given data.
- Discuss the benefits and limitations of AI for smallholders.
- Describe government initiatives that use AI or technology-based yield estimation.
Described in general terms: the kind of examination, not a past paper. Laisha never reproduces exam questions.
Questions people ask
Will AI replace extension workers?
Unlikely. AI can handle routine queries at scale, translate content and flag problems early, which frees extension staff for field visits, demonstrations and complex problems. Trusted human advisers remain essential, especially when advice involves risk or new practices.
Is AI advice safe for pesticide decisions?
Only if it is grounded in approved label recommendations and local thresholds, and if farmers confirm the diagnosis. Wrong identification or dose advice can cause crop damage, resistance or health risks. Treat AI suggestions as a second opinion, and check them against official recommendations.
Do I need big data to use AI on a farm?
Not necessarily. Many tools are already trained, and they just need a photograph or location. Building new models does need large, well-labelled datasets. Shared public datasets and farmer data cooperatives are helping make that possible.
Sources
- Liakos, K. G., Busato, P., Moshou, D., Pearson, S. and Bochtis, D. "Machine learning in agriculture: a review." Sensors 18, 2674 (2018).
- Lam, R. et al. "Learning skillful medium-range global weather forecasting." Science 382, 1416–1421 (2023).
- Mohanty, S. P., Hughes, D. P. and Salathé, M. "Using deep learning for image-based plant disease detection." Frontiers in Plant Science 7, 1419 (2016).
- FAO. The State of Food and Agriculture 2022: Leveraging Automation in Agriculture for Transforming Agrifood Systems. FAO, Rome, 2022.
More in latest in agriculture
Discussion
Questions, corrections and notes from readers on this article. Every comment is checked before it appears: keep it on the topic, and do not share phone numbers, links or contact details.
Last reviewed