4 Ways Generative AI can be used in Agriculture.
Large Language Models(LLMs) and Foundation Models, are trained on hundreds of billion and trillions of tokens, and can understand questions posed in natural language. They have the capability to personalize responses to questions, create images, videos, and music from text-based inputs, and summarize documents or meetings/voice calls. Generative AI, popularized by ChatGPT, has wowed users with its ability to provide personalized and high-quality responses to a broad set of questions. Similarly, image and video-generating tools like Midjourney have shown phenomenal benefits in terms of time and cost to generate unique, high-quality graphics. These models are disrupting many workflows in every industry and providing better solutions in a cost and time-effective manner.
Although chatbots have existed for a while, the LLM’s ability to provide accurate and personalized answers that build on feedback has made them attractive. Several use cases are emerging for how Generative AI can be used in agriculture. These use cases are particularly important given the work the Ministry of Agriculture in India is doing on AgriStack, which is creating a set of agriculture data available as a digital public good. LLM models can be trained on such data sets and provide several new public services to farmers.
LLMs increase the importance of DPGs, and the faster we create these, the faster we will see innovation in this sector. Here are some examples:
1. Personalized Advisory to Farmer
One of the pain points for a farmer is a lack of advisory in all stages of crop management.
Over time the role of a government agriculture officer, who used to provide consultative services in the past has become more focused on subsidy disbursal alone. We also saw that the farming practices within a village for the same crop vary across a farmer, due to this gap. The prescribed package of practice is often not optimally followed. This has left a gap in the market around farmers’ awareness of newer techniques and farming in alignment with changes in nature — water, soil, and market requirements of production.
In a voice or text input a farmer can provide the details about the question they have, and then provide additional detail like an image of a plant, soil, etc. if needed, and the system, through consultation can provide contextual advisory.
Further, the generative AI is an interactive interface, so it can build and refine based on a series of questions/answers and essentially a discussion with a farmer.
Interestingly, four pilots are already underway in this space, as some companies with existing infrastructure or those who built it quickly have launched it quickly.
1. Samagra launched a pilot Ama Krushi in Odisha, just a month back.
2. DigitalGreen with Gooey.ai, launched Farmer.chat
3. Apurva.ai has launched a pilot for National Digital Extension and signed an MOU with Ministry of Agriculture.
4. Kissan GPT by Titodi
2. Eligibility for Schemes
Scheme eligibility information on government sites is available in all different formats. The generative AI LLM models are a great fit for training an LLM model with Govt scheme data.
Then it can answer a farmer or any consumer, which government scheme am I eligible for. Or I am facing certain issues, is there a scheme to help me with this?
While there is a solution provided by Haqdarshak, this space can expand rapidly with several providers providing this information in an easy-to-use manner. The beneficiary can then take assistance from a local entrepreneur who may offer the service of helping someone enroll for the scheme.
3. Monitoring the problems faced by farmers in real time, and proactively managing those at state level
Most states have a phone number for farmers to call for help, based on the issues faced. By using voice-based ML, and doing a summary of a call, one can monitor and categorize the daily calls coming to the call centre automatically into the type of issues faced by farmers.
Such a dashboard that provides the top challenges faced by farmers in real-time can be a great nudge for better service to be provided by various agriculture departments. This could also be put in the open and private and NGO players could assist on a need basis.
An example of this is — a pest attack on a crop in each region. Lack of seeds, fertilizer in a region, etc.
4. Personalized training in agriculture — educational content in Agriculture
Current Generative AI provides a agent which acts as a personalized tutor to the specific question a farmer asks. Since the LLM model is trained on almost all the data available on the internet, and a custom LLM for agriculture will have data provided by various agriculture universities, and all the data sources in agriculture, this virtual agent can be built further upon as a personalized trainer for a farmer.
Examples –
1. a farmer who is doing paddy cultivation wants to move to grow cotton. It can ask the system to learn about the same.
2. If some farmer wants to follow the GAP (Good Agriculture Practices), for their products to become eligible for export, it can learn this from the system.
