Harvesting the Future: The Scope of AI in Smallholder Farming

Understanding AI

The key to realizing the potential of AI within agriculture is understanding what it is and how it works. AI is an umbrella term for machines that we code to solve problems and learn in a manner like humans. 

The AI available in the markets are considered as ‘models’; they are programs that have already been trained on data to accomplish specific tasks.

These models are built up of different components and concepts from within the field of Artificial Intelligence. When thinking about AI-related innovations in agriculture, it’s important to know what capabilities our models could have. AI models can use one or multiple of these concepts based on what the goal of the product is.  

• Machine Learning (ML) focuses on allowing machines to learn from pre-existing data and making decisions/predictions based on that data. Generative AI (GenAI) is a type of ML that was developed more recently and is extended to generate new content based on the data that it was trained on. Deep Learning (DL) is another type of ML, that specializes in recognizing complex patterns from large datasets, just like the human brain. It is also built similarly, using deep neural networks; all this together makes models using this system excellent in image and speech recognition. 
• Natural Language Processing (NLP) is particularly concerned with the interactions between computers and humans. This system allows machines to understand, interpret, and generate human language. A very prominent use of it is in chatbots such as ChatGPT.
• Computer Vision enables technology to understand and interpret visual cues from images and videos. This includes identifying and classifying objects and people, understanding image context, and tracking movements. 
• Robotics is the motorization of AI. Based on real-world feedback, robots can change their behavior when interacting with the world. They are not just programmed to complete specific physical tasks, but also learn from each interaction. 
  
• Expert Systems are another form of enforcing and specifying within which parameters an AI model should function. Instead of being trained on datasets like ML, the model is trained on different facts, rules, and knowledge provided by an expert in the AI model’s field.

These techniques come together to create models that can offer a range of innovative solutions to address the challenges faced by farmers. AI’s transformative power in agriculture comes from its ability to collate vast amounts of data, analyze it, derive meaningful insights, and share this with its user in an easy-to-understand and actionable manner.

Common Smallholder Issues

Farmers are expected to be experts in various fields such as crop protection & growth, animal care, climate resistance, agri-legislation, and the economics of agriculture. The odds are stacked against them in a swiftly changing world. All these areas require one key thing: information.

Smallholder farmers face challenges relating to access to information and resources. They may not understand how to find and use farming metrics to their advantage due to the digital divide. Oftentimes, farmers have limited network coverage and digital literacy. This hinders their ability to access information, make informed decisions and optimize their farming techniques. 

They typically rely on neighboring farmers, villagers, and government extension officers for information. For smallholder farmers in rural locations, extension officers are a crucial player within the ecosystem and key source of agricultural information.

Often constrained by meagre resources and access to information, farmers stand to benefit significantly from the potential of AI to revolutionize farming techniques and enhance productivity.

From Intuition to Informed Decisions

Farming traditions have been passed from generation to generation or influenced by the outcome of someone else’s farm. Traditionally, farming has relied on experience and word of mouth, with some farming products being recommended this way seeing an approximate 45% market participation increase.

It is estimated that by 2050, an average farm will be producing approximately 4.1 million data points daily. The emergence of AI is ushering in a new era of agriculture. These advanced tools process vast amounts of data that can be strategically applied to farming, allowing for data-driven improvements. Smallholder farms can also strive for this, provided they have appropriate solutions and adequate support. 

Smallholder farmers could have access to guidance on critical aspects of farming like crop rotation, optimal fertilizer usage, irrigation schedules, pest control strategies, and even harvesting. A deep learning model like Recurrent Neural Networks can be used to predict how various factors such as temperature, precipitation, and sunlight affect crop growth. This predictive capability enables farmers to make informed decisions about planting times and cultivation practices. Recommendation of appropriate measurements of fertilizer, water, and pesticides ensures limited resource wastage and lower emissions.

For example, a web-based Rice Crop Manager that was tested saw an increase in yields, reduced cost of 50-100 USD per season per hectare, and reduced environmental impact (through reduction in leaching and groundwater pollution) leading to improved sustainability. 

Agriculture robots could also aid in providing direct help to farms through Computer Vision and sensors. They can analyze real-time soil data and the visual state of plants. Drones are also being used for monitoring crops, irrigation, weed/pest identification, and wildlife. These robots collect data, whilst also reducing labor hours and costs. This also combats against the falling supply of farmers due to less and less of the youth wanting to become farmers.

Though tools for data analytics always existed, it has always been more accessible to larger farming operations rather than smallholders. Through GenAI models, those with less digital access and literacy can unlock similar functions. It can identify the most suitable weather conditions for different types of crops or analyze soil data to predict the nutrient needs of crops and identify soil deficiencies. 

By leveraging these insights, farmers can optimize resource allocation, improve crop yields, and reduce the risks associated with crop management. Integrating AI into agricultural practices not only enhances efficiency but also promotes sustainable farming methods essential for long-term agricultural resilience.

LLMs and Their Role as Extension Officers

An agriculture extension officer is a professional who helps farmers and rural communities to access information and resources to improve their agriculture practices. They act as middlemen between important information and farmers often providing capacity building (training) services in the agri-ecosystem. Typically, they are employed by government agencies or work with non-profit organizations, research agencies and private companies to develop and deliver agricultural information, educational programs and services. They can also connect farmers to specialists and subject experts and advise on regulatory requirements and business support opportunities including alerting farmers to the latest technology and data tools.  

Despite their critical role, many extension systems globally face challenges such as limited staff capacity and inconsistent outreach. For example, in Zambia, there is roughly 1 extension officer for every 1,000 farmers while in Kenya, it's 1:1,800— well below the recommended ratio of 1:400.

To address these gaps, Large Language Models (LLMs) can be used. LLMs trained via Expert Systems can emerge as virtual extension officers, utilizing their NLP capabilities to provide smallholder farmers with immediate access to essential agricultural information. Farmers can gain insights into the latest agricultural practices and sustainable methods tailored to their specific conditions.

Challenges of Using AI for Agriculture

While AI has the potential to revolutionize the way farmers manage their crops and increase crop yields, there are also certain limitations and challenges that must be addressed.

The digital divide is a primary concern. Smallholder farmers often lack reliable internet access, which limits their ability to utilize AI effectively. Additionally, digital literacy among these farmers may be low. Training and support are hence crucial to ensure that farmers can effectively interact with these models and understand the insights they provide.

Data availability and cost are also significant hurdles. For models using ML and GenAI, generating accurate predictions necessitates extensive data on crop performance, weather patterns, and soil conditions. However, implementing the data-collecting infrastructure may require additional investments in robotics such as sensors and drones for data collection. This financial burden is considerable for smallholder farmers operating on tight budgets, requiring affordable solutions or adequate financing to encourage widespread adoption. It is also crucial to consider finding the right business model for these new AI-based services, especially since monetizing from smallholder farmers can be challenging.

LLMs also have their own limitations that stem from the model still being in development. Firstly, they are predominantly trained on datasets in widely spoken languages, potentially excluding local dialects and languages spoken in rural areas. This restriction limits access to vital agricultural information for rural farmers. The model also still lacks understanding of subtle causal relationships. For these models to be successful, they should not be dependent on the ability of farmers interpreting the insights provided, and instead can give clear reasoning and solutions. Lastly, LLMs require extremely large data sets and might be more beneficial at solving a more varied set of agricultural tasks that require nuance. This means they are both energy and money-consuming.

What opportunities exist?

There is a growing need for smaller, low-latency LLMs that can be accessed on phones or even installed on edge devices. These models should be specifically trained for agricultural-related knowledge. This approach could significantly enhance accessibility and usability for smallholder farmers in remote areas. There should be language-specific adaptations to maximize the model's utility. The AgTech industry needs to continue investing in AI research and development to resolve these issues.

For being an information-conveyer like extension officers, Small Language Models (SLMs) might be more useful for smallholder farmers compared to LLMs. SLMs require fewer large data sets and use less resources to function, meaning that they are more sustainable and are more specialized to specific tasks. Microsoft’s Phi-3 is a great example of this; it is affordable and can diagnose crops while providing guidance.

Lastly, building acceptance of AI models is crucial for their utilization among smallholder farmers. Oftentimes, farmers are hesitant to change their methods drastically for good reason: any possibility of unreliable crop growth affects their livelihoods greatly. Addressing skepticism through transparency about model operations (regarding privacy, surveillance, data ownership) and ensuring reliability in recommendations are pivotal steps toward fostering trust and maximizing the benefits of LLMs in agriculture. 

AI stands at the forefront of integrating technology into agriculture, offering farmers powerful tools for sustainable and efficient farming practices. By empowering farmers with accessible information and enabling precision agriculture, AI can revolutionize smallholder farming. As this technology evolves, it promises to provide farmers with the knowledge and resources needed to thrive in a rapidly changing agricultural landscape.

The adoption of new agricultural technologies presents both opportunities and challenges for smallholder farmers. Overcoming barriers such as financial constraints, resistance to change, and limited agricultural consideration requires a comprehensive approach. These efforts will guarantee that all farmers, regardless of scale, can leverage AI and other AgriTech innovations effectively.