In late 2022 several organizations began to release a type of artificial intelligence publicly called “generative artificial intelligence.” The artificial intelligence that’s received much attention is called “ChatGPT.” ChatGPT – the name of a product by OpenAI – is a type of generative artificial intelligence known as a large-language model or LLM. Large-language models can understand human language, and can use our spoken or written words and ideas to analyze situations as well as give output and feedback that’s also in the form of human language.

Those artificial-intelligence models are trained using information from the Internet. An large-language model like ChatGPT is doing its work based on having previously read and processed billions of documents, websites, videos, news reports, technical documents, educational textbooks, university bulletins and other sources of knowledge. When the user asks a question or “prompts” a large-language model, the question is processed and the vast data warehouse is used to structure and formulate an answer that sounds humanlike or natural in tone. That seems impressive to most people who use a model like ChatGPT or others. But the model only uses super-high-speed computer processors to create an answer based on statistics and what the training data would most likely predict as an answer. The large-language model uses statistical probabilities to structure and craft answers word by word.

Beyond text that might appear on a computer or phone screen, generative artificial intelligence can also generate other types of information like pictures, video, music, voice or other forms of communication. Depending on the artificial-intelligence model, they might also be able to detect objects in a photo or video, describe them and present information about the object. Examples include the task of describing the maintenance requirements for a specific machine, or care recommendations for an animal, or resources that specify control options for a crop pest. The best way to learn more is to tinker around and learn what’s possible with those models.

Companies and public organizations are just now beginning to embed the technology into new products or services. For example the AI-Farms project at the University of Illinois has trained a large-language model to answer agricultural crop questions. It’s trained on several thousand pieces of Extension-service information developed by University of Illinois researchers and educators. Private companies are also diving deeply into artificial-intelligence-based products to support the ag industry. At the University of Wisconsin researchers are beginning to create specialized, trained models to answer questions and serve as a technical support system for agricultural-safety issues, farm-level regulatory compliance and management best practices for woodland landowners.

Here are seven farm- and ranch-related use cases that we’ll see in the coming months or years. Some of these are happening now.

Recognition of specific pests like weeds, diseases, and insects and recommendations for appropriate control – it’s increasingly easier to train a computer using tools such as machine learning to recognize an object’s unique features. The software can examine shapes, sizes, colors and other distinctive features of a particular object – like the spot pattern on a bug or the number of leaves and leaf-length-to-width ratio on a plant. If the software can identify a particular pest or any object with a good degree of certainty, the task is straightforward of pulling the model’s past “training data” that might include university fact sheets, product labels or other information and presenting it to the user. It’s doubtful that those tools will ever replace the need for a skilled and experienced professional. But they can act as a well-trained assistant and save time and resources – especially in common, time-consuming and repetitive tasks.

Summarizing farm and ranch news and information communicated through print, podcasts, YouTube videos or other formats – whether a big-farm or -ranch operator or a smaller market grower, agriculture is becoming increasingly specialized. Information overload is a real issue, often adding to producer stress and mental load. Many of the newer large-language-model tools can read text from a report or newsletter that the user inputs, and it will provide short, concise and accurate summaries. It’s also possible for these tools to summarize YouTube videos, podcasts and other familiar news sources for the industry. Imagine having a series of 30 or 40 half-hour webinars regarding a complex topic like robotic applications in dairy farming or vegetable production. Artificial intelligence allows a person to spend an hour showing the artificial intelligence tool the links and sources of information. Then the software can prepare bullet-point summaries of crucial information or even-more-sophisticated tasks like outlining common threads or differing viewpoints among videos, presenters, pros and cons, and other considerations.

Machinery and equipment setup, maintenance, troubleshooting and safety – in theory it’s now possible for a large-language model to be given access to all the information in a particular operator’s manual for a tractor, harvester or other machine. This will likely be something that manufacturers will offer in the future. If the operator’s information is read into and processed by artificial intelligence, rather than reading the manual and searching for the correct information, the user can “ask questions” directly to an artificial-intelligence tool and be given answers that are rooted in the manual. As is the case with all artificial-intelligence training information, it’s essential to understand that many sources of information and data are proprietary and protected by copyright law. But this application will be commonplace in the coming five years or sooner.

Interpreting complicated regulations, standards, and rules, and providing farm-specific action plans for compliance – whether they are state, federal or local codes, rules, regulations or standards, the documents that dictate best practices and regulatory requirements are densely written and sometimes challenging to understand. A large-language model can digest old or new regulatory documents and provide summaries, including easy-to-understand checklists or outlines that make compliance more straightforward.

Helping with communication such as complicated family conversations, drafting emails or letters, or training programs for hired employees – as is the case with all these use cases, generative artificial intelligence can act as a powerful consulting “assistant” with many unique or routine communication tasks. It should never replace the thoughtful consideration of a real person, but it can help ensure you communicate with clarity and completeness. For example say you know your family is due for a multi-generational conversation about estate planning. If a large-language model like ChatGPT 4.0 is given a detailed story, context, history and considerations, it can be used to provide ideas on how to structure conversations, anticipate emotions and responses, and suggest ways for everyone to participate. Similarly it can be used as an idea starter or fact-checker for letters, emails and reports. As is the case with any artificial intelligence, it’s important to be able to explain “what you want” from the software but also be careful not to disclose specific private data or information that might be sensitive or protected.

Language and technical-information translation – increasingly there is interest in using artificial intelligence to rapidly translate educational information, communicate workplace policies or just help with routine person-to-person communication. The new generative artificial-intelligence tools are being tested and shown to be just as, if not more accurate, than paid apps or online “translation” tools. Look for more developments in this area of use because a large-language model can be used to translate text or spoken word. But it’s possible to “speak” the translated content in a human-sounding voice using generative artificial intelligence. That will open amazing possibilities in the workplace and be an essential tool for travel.

Brainstorming such as assisting with strategic planning, evaluating “what if” scenarios, or considering all appropriate points in looking at the pros and cons of a specific decision or alternatives – when we know we must make a difficult decision or take an essential step in business, often the most challenging part is starting. Overcoming inertia is one step. But there’s often also a lingering sense that you won’t have all the information you need or that you might overlook an important consideration. Again, artificial intelligence can assist you in starting. It will not do the work for you or “make” a final decision. But if you carefully describe your situation in detail it might suggest ideas, starting points or pathways you’d not thought about previously. It might also suggest additional resources, sources of expertise or viewpoints to consider.

As has been mentioned several times in this article, generative artificial intelligence in the form of large-language models like ChatGPT and others must never replace human wisdom and decision-making. But it’s a tool that can save us considerable time and lead to more-well-rounded, holistic and complete “thinking” about important issues. It can also digest vast amounts of complex information and then produce output that’s translated into more-easily-acted-upon steps or practices, or literally translated into another language that’s more-easily communicated to others.

This summer I will be preparing a step-by-step tutorial that any person interested in agricultural applications of artificial intelligence can try using “free tools” that are available on the web through various open-source platforms. The tutorial will include “walkthroughs” that farmers and ranchers can use – based on the seven use cases described in this article – that you can easily tailor to meet your own needs and situation. I will also include best practices for “prompts” or carefully formatted and described “questions” that will encourage the large-language model to answer in a helpful and useable way. Email me at shutske@wisc.edu to be put on the notification list and to receive a link when the tutorial becomes available this summer.