AI in Agriculture: Opportunities and Challenges

Blog » AI in Agriculture: Opportunities and Challenges

Share On

Blog Posts

A colorful and exciting illustration showing the integration of Large Language Models (LLMs) with agriculture. The image features vibrant elements such as a farming field, a tractor, a farmer, and symbols representing AI and LLM technology. There are also images of crops, an apple, and a barn, all set against a backdrop of bright, abstract patterns.

Embracing Generative AI in Agriculture

Access Old Satellite Images with Our Interactive Tool

3 ways to Assess and Mitigate Wildfires with Satellite and AI

According to Forbes Information is the new Gold. Agricultural practices generate tremendous amounts of Information. So why does agriculture, where margins are thin, leave gold on the table? Why is there rather limited use of artificial intelligence in agribusiness? What factors drive adopting data driven agricultural practices? How will AI affect agriculture? Can artificial intelligence provide solutions to the challenges in the agriculture industry? This post attempts to answer these questions and to shed light on fundamental challenges and opportunities around widespread adoption of AI in agriculture.

Is AI Applicable to agricultural practices?

AI and AI-driven solutions are the buzz words of the 20s. Touted as the panacea of our age, AI has been thrown at every challenge in industry and society, and not all of those initiatives are fruitful. Technically, there are some areas where reaping the benefits of application of AI is more pragmatic and realistic. Big data and machine learning shine in areas where extensive data is available and the patterns in the data are complex. Agriculture generates big data, and that data is complex. Besides variables that farmers can control such as variety selection, seeding date, and fertigation rate, there are many variables that are beyond farmers’ control such as environmental factors. The multifactorial nature of these variables makes finding correlations between an input (e.g. interval time between application of fertilizers) and an output (e.g. expected yield) extremely difficult. Such scenarios are exactly why AI in agriculture could shine. AI can deal with multidimensional problems with high variance where randomness is also a determining factor.

Plenty of Room for Improvement of AI in agriculture

Tracking applications of artificial intelligence in different industries and compare it to Ag-sector proves that agriculture is not fully reaping the benefits of the latest technologies in AI. If you Google search for an AI driven solution for any given problem in finance, you will find hundreds of hits on content and relevant quality sample codes. In contrast a search for application of AI in agriculture to improve the yield of wheat, rice or corn brings a rather limited number of resources, considering that these three crops feed the majority of the world. It is a fact from business point of view that wherever there is a gap between where an industry is and where it could potentially be, there is opportunity for growth, profit to take and bottom line to improve. Early adopters and agile thinkers are those who gain big in pivotal moments of the digital age.

In spite of the bottlenecks, transformation of Ag-sector to a data driven sector is not a matter of ‘if’, it’s a matter of ‘how’ and ‘how fast’.

What Are the Current Applications of AI in Agriculture?

We live in an interesting era where almost every week there are news around new applications of AI that could help businesses and consumers. Farming, horticulture, floriculture, and orchard industries are not exceptions. Although not with the same speed as some other industries, agriculture embraces innovations brought by AI. Here is non-exhaustive list of some applications of artificial intelligence in upstream agriculture:

Smart sprayers: Why would you spray herbicides on both plants and weeds if you could target just the weed? Computer vision could help reduce the cost of herbicide consumption while avoiding plant products’ contamination with chemicals if it is not needed.
Yield Prediction: AI brings predictability to Ag space on a scale and precision that is not achievable otherwise. This could help farmers to take proactive rather reactive measures in their practices.
Smart Greenhouses: Smart Greenhouse Market is projected to reach USD 2.1 Billion by 2025 from USD 1.4 billion in 2020. The controlled and semi-controlled environment in greenhouses makes grower to use sensors and collect data more extensively than farmers. Data collected from various sensors and actuators, if integrated while streaming, brings new opportunity through AI algorithms to provide insights for growers.
Integrated pest management using computer vision: Lack of enough experienced scouting personnel is a challenge for grower these days more than ever. What if a computer program connected to fixed or moving cameras doing that for growers? This means real time vision of a greenhouse (or a farm) even when nobody is there to evaluate plants one by one.
Plant Disease Prediction: The occurance of diseases can be predicted based on historical data and the upcoming weather situation. It will save a lot of money for farmers and help them to apply pesticides more effectively.
Large Language Models (LLMs): Utilizing advanced AI, LLMs can process vast amounts of agricultural data to provide insights on crop management, streamline administrative tasks, and enhance decision-making processes by synthesizing information from diverse sources. This application not only improves efficiency but also supports more informed strategies on the farm.

LLMs in agriculture

To learn about challenges and opportunities of generative AI models like GPT in farming read our blog post: Embracing Generative AI in Agriculture

Opportunities for Applying AI in Agriculture

In spite of the challenges that hindered a widespread adoption of artificial intelligence in the agriculture sector, the sector has recently welcomed AI driven initiatives with a speeding pace. Following trends in scientific publication, momentum in emergence of new Ag-tech startups, and pivot of agricultural corporations toward AI approaches, all point to the upcoming seismic changes in Ag-sector. The projection for expansion of global agriculture analytics market size is estimated to expand by 75% from 2020 to 2025.

Funding to Tech startups in agribusiness grew more than 6 times in 6 years (Source)

There are 4 main driving forces for the large scale adoption of AI in agricultural practices;

1.Global Food Demand

This cause is the elephant in the room. Agricultural practices are and will be shifting toward an efficient data driven industry for an obvious reason; the global food demand is increasing drastically. UN estimate of the world population by 2050 is 9.8 billion, and the world food demand is expected to increase by 59%-98% between 2005 and 2050. This demand is not going to be addressed by tapping more to dwindling water resources, shoving more fertilizers into soil and cutting more trees to expand farms. Indeed, crop yield around the world is estimated to be affected adversely by climate change. The demand for more food is to be addressed by adopting technologies in Ag-business that improve efficiency in sustainable and climate friendly manners. Wide spread adoption of precision agriculture is the most effective approach to improve yield and save on costs with minimal environmental footprints. Data collection and control over farms through data are inherent characteristics of precision agriculture. This ongoing transformation inevitably opens up doors for farmers and growers to adopt AI and data driven practices for an obvious reason: if you already collected data, what holds you back from extracting more insights from it? Like any other transformation, this change is going to take time to settle; nonetheless it already is well started and gaining momentum year by year.

Consumption of food is also growing in parallel with population (Source)

2.Availability and Affordability of Infrastructures for Data Collection

Over the past two decades, the extent of data collection in the agriculture sector exploded by orders of magnitude. Farmers and growers now have their fingers on vast data that is granular and detailed on a level that was not imaginable 20 years ago. Thanks to advances in IoT technologies, data collection systems are becoming more affordable and widely used year by year. This means farmers and growers can have integrated data collection systems at a fraction of cost of that of 15 years ago (considering advancement in the IoT products over past 15 years).

Adoption of IoT in Ag space generates and provides the most important commodity for AI in agriculture: Data!

Americas and EU have the biggest chunks of growing agricultural IoT (Source)

3. Competitiveness and Increasing Cost of Labour and Materials

Data driven agricultural practices reduces the costs and improves the yield. On corporation level, this is a known upcoming change in agriculture business. The news about every single big name in agribusiness (e.g., Monsanto and Bayer) shows their initiatives in AI, and that they already started programs and budget to reap benefit of AI in agriculture. On the grower/farmer level thought, a practical use of artificial intelligence tools is limited, which proves that practical application of artificial intelligence in agriculture sector is at its infancy. History of adoption of technologies, whether it was electricity, internal combustion vehicles, internet, or smart phones, shares a repeatable pattern: it is exponential. It means that the use and extent of technology at early stages is dismal and the price of technology is high, but as it finds the momentum it accelerates in exponential trend. This pattern holds true particularly for small scale farming, where disruptive technologies could shrink the significant yield gap because of lack of proper technologies.

4. International/Governmental Initiatives for AI in Agriculture

A growing number of international organizations, governments, NGOs, and non-profits are attending to data driven agriculture initiatives. Their activities come in form of education, facilities, grants and funding, and most importantly open data and data collection infrastructures. Data collection infrastructures are crucially important as they serve as the backbone for a meaningful progress in AI in agriculture. FAO, GARDIAN, CIMMYT, USDA , and GODAN are among fruitful examples.

Challenges that Slowed Down Adopting of AI in Upstream Agriculture

Any meaningful initiative toward adoption of AI in agriculture needs to evaluate the practical barriers that slowed down transformation of this industry compared to some other sectors. Although the landscape of use of technologies in agriculture is rapidly changing, in order to project the future directions, we need to understand where it came from and what the current status is. There were (and to some extent ‘are’) four fundamental speed bumps in the way of adopting applicable AI-based technologies in upstream Ag-sector;

I. Data Collection

In spite of growth of precision agriculture in past three decades, there are two main problems that have hindered wide-spread applications of machine learning predictions in the sector. First, a significant portion of farmers and growers have not adopted utilities of precision-Ag (and there are many studies and reasons around that). Second, the use of precision-Ag technologies does not mean collecting all critical information using IoT or other mediums for the purpose of prediction.

The latter factor could be a breaking point for usability of AI from a technical point of view. Most agricultural activities have some inherent uncertainty (in AI terminology: irreducible error) because of environmental factors that are not under our control. This means ongoing collaboration around data collection procedures should take place between data scientists with domain knowledge of agriculture on one side, and farmers, growers, agronomists, and consultants on the other side. This shared knowledge and understanding pave the way for developing models with reliable utilities for farmers and growers.

II. Data Ownership and Privacy

Who owns the data in Ag-sector? The answer to this question is not as straightforward as you might expect. Yes, data is owned and accessed by producers, which are farmers, greenhouse managers, breeders, agronomists, etc, but that is not the whole picture. Although owners have access to some critical data, they don’t have immediate access to many other sources of relevant data such as historical environmental data in the area or satellite imagery, unless they pay for the access and get trained for using it. This gap (particularly the lack of access to environmental and satellite data on the side of farmers) could be well covered by data scientists and data engineers who are trained to gather and merge data from disparate sources.

Data ownership is an issue in many sectors, but a confounding issue in Ag-sector

Data Privacy is another confounding issue in sharing farming data. Anonymizing, encrypting, and aggregating are among methods to address issues around intellectual properties while applying AI in agriculture on large scale. There are appropriate non-disclosure agreements (NDAs) and AI techniques to help analyze the data and extract useful insights without dealing with issues around intellectual property of farming data particularly when granularity of data matters or data has spatial aspect.

III. Agricultural Data Has Been Scattered

Even if data is collected thoroughly, and a farmer (or an advisor) could have access to all relevant data from the farm, the data collection among farms is not centralized. That means a collective knowledge/insight derived from AI-based techniques do not fully flourish. To illustrate the sharp contrast, let’s compare agriculture to finance. In finance, you can have access to a great wealth data for all publicly traded companies from around the world with a history over the past few decades for free, or paid data from services such as Bloomberg. Consequently the sheer number of educational, for profit, non-profit studies using AI in financial data is mind blowing. Such a centralized and high resolution data system does not exist in the Ag-sector. FAO estimated there are 570 million farms in around the world. How a data scientist could leverage that wealth of data for better yield? One way around this problem without hurting privacy laws is to use satellite imagery data. Historical satellite data with resolution as good as 30 cm is available to purchase, or free for lower resolutions for almost all around the globe. This means computer vision specialists and data scientist can extract useful features (such as NDVI and NDWI) out of historical satellite data to train their models for the predicting yield, disease, or stresses in farms.

If this approach is already getting used by wealthy traders and investors used by wealthy traders and investors to pick their next investment, why not farmers get benefited from it?

IV. Cost of Change Toward AI in Agriculture, Cost of Mistakes

The high cost of change is particularly true in the case of farming industry. This cost structure is very different from some other sectors that adopted artificial intelligence with open arms. If Uber runs an email campaign based on what their data science team suggested, and after a few days turns out that it does not work the way it was estimated, they can stop/modify it immediately. The risk factor is on a different scale in a farming system; following a wrong recommendation could mean losing crop yield for a farming season. One approach around this issue, which already is practiced in farming, is to adopt changes on smaller scales even if the farmer farms on large scale. The small scale means a small portion of a farm, or small changes toward the targeted change over multiple farming seasons while reflecting on feedbacks. Another approach, which inherently exists in data science and AI solutions, is probabilistic approaches. Predictions in AI are rarely deterministic, although they might be expressed a deterministic way. Accuracy of a prediction is always a probability, and that is the reason statistical confidence interval is used next to a prediction. An experienced data scientist rarely issues recommendations based on a model unless he or she reports the confidence in that prediction. This approach is widely used to evaluate and marginalize the risk associated with changes prescribed by AI solutions.

Bottom Line

The gap between the current state of agricultural practices and the potentials for adopting AI is shrinking very fast. There are some bottlenecks for the transition, nonetheless; a combination of global demand, availability/affordability of infrastructures, and national/international initiatives drives the fundamental changes that already started happening.

It is inevitable: the future of agriculture belongs to EARLY ADOPTERS of AI in agriculture.

Special thanks to Aron Cory and Ali Safilian for revising this post.

AI for Your Own Business

Book an appointment with us to learn what specific opportunities AI will provide to your agribusiness.

Book an Appoointemnet

Table of Contents