Using artificial intelligence (AI) as an assistant to answer questions has the potential to increase your impact, but to get there, you must separate potential from unrealistic promises.
At this point, large language models (LLMs) and other machine learning are simply prediction machines based on training neural network models (think hand-held calculator). The prominent LLMs, such as OpenAI, Anthropic, Google and Meta, have used public information to build continual feedback loops and predict the next word in a sentence, the next action based on a question or how to reorganize information. They’re highly dependent on public information and detailed instructions. Like a calculator, you won’t get the right answer if you don’t put in exactly the right query.
On-Farm Application
Similar neural networks are being used for agricultural data to suggest everything from market predictions to planting dates. The farm-specific data you use isn’t available online today, so it’s impossible to build that data into an LLM. These models make assumptions of how the agriculture system works within fixed variables, such as the price of corn yield minus input cost. LLMs don’t factor in the ecosystem, market potential, nutrient index, etc., which limits their value. Most importantly, they are not able to think like a farmer.
Remember the two biggest assets you have in the game of agriculture are your brain and your land. A university partner might be able to build a model that helps you predict some components with more accuracy, but it’s difficult because so much of the data is common sense and localized.
It is too early to understand how synthetic data and self-training will help the models learn and react to real-world scenarios, so for now they should be viewed as a trusted assistant to your brain.
Here are a couple ways to experiment:
- Download ChatGPT and do a few searches. I did this by asking, “When should I plant corn in West Tennessee on my no-till field with 4% slope and 60°F soil temp in mid-April after a ½" rain?” I continued to narrow the parameters by telling the model there were 25' trees on the north and east sides of the field, a stream along the south side and that next week there is up to 1" of rain expected, followed by a two-week hot, dry spell. I got a lot of good suggestions, but never the recommendation that I should consider planting soybeans.
- If you’re seeking AI-enabled technologies, find ones that build on a good understanding of how your farm works at a granular level and can answer specific questions. As always, partner with other innovative farmers to share and evaluate new technologies and with startup founders to use AI technologies to their full capability.
Your Next Read: Harvesting Insights: How AI Crop Scouting Is Driving Decisions
