AI can help in farm HR
Can AI help your farm’s workforce development? Wolfgang Heuweiser, DVM, Ph.D., professor and director of Quality Milk Production Services in the Department of Population Medicine & Diagnostic Sciences in Cornell’s College of Veterinary Medicine, presented on the topic as part of Cornell’s “Boots in the Barn” webinar series.
“The point of this is we should not be only filling positions on a dairy farm but to grow and train the talent behind those positions,” Heuweiser said.
He thinks that practical AI can help extracted takeaways, summarize text and generate ideas.
Heuweiser referenced Hermann Ebbinghaus’s “forgetting curve,” which asserts that memory declines as the number of days increases since a person has heard something new.
“People have the tendency to forget knowledge within the first two days by the rate of 70% to 80%,” he said. “Only if you repeat it within the next few days will they remember.”
The concept of spaced repetition relates to the adage “repetition is the mother of all learning.”
The age of listeners also matters. “Adults learn differently than young students,” Heuweiser said. “They want to play an expert role and want to bring experience to the table. It has to resonate with a scenario.”
Evaluating the effectiveness of learning begins with students’ reaction, which progresses to learning, behavior and then results.
“We’re attempting to change behavior in the milking parlor,” Heuweiser offered as an example. Dairy farms want a lower somatic cell count. To achieve this or any goal, “we should make the outcomes measurable and the outcome should be ideally to change the behavior.”
He mentioned that learning consultant Mike Taylor suggests that the SURE framework (Source, Understand, Research, Evaluate) offers a way to make learning content easier.
“Whenever you create something, it should be simple, useful, resonant and easy to skim,” Heuweiser said. “I think it’s a pretty convincing concept.”
Something that is simple can be said in few words. Useful information is relevant to the learner’s job. To be resonant, the information should connect to a real goal or pain point. Information that is easy to skim can be understood within five seconds.
Farms that have standard operating procedures (SOPs) can better communicate with employees, create improved consistency, boost quality assurance and ease onboarding of new employees.
“Only very few dairy farms do have SOPs,” he said.
However, a survey by JDS Communication a few years ago indicated that many farmers are interested in SOPs but lack the time and know-how to write them.
SOPs should exist for many areas of farmwork. In a survey, Heuweiser asked farmers if they had SOPs for cleaning, disinfecting, handling, primary care, feeding colostrum, emergency care, drench, colostrum quality, pasteurization, disbudding, sick calves, automatic feeder and weighing. Most had no SOPs for these vital areas of animal care, yet most farmers wanted SOPs for standardizing these tasks.
A checklist-based approach to measure milker behavior before and after training resulted in behavior change in the milking parlor.
“When I look at SOPs or protocols, I always think, ‘How can we do a better job than written job descriptions?’” Heuweiser said.
He had the opportunity to participate in on-job training sessions, which he called a learning experience of thinking of ways to train materials.
“We wanted to incentivize our participants to sign up and provide some very limited demographic information,” he said.
Participants could choose different media and topics and use English or Spanish versions. All the modules were two minutes long, as he has discovered that long training sessions tires participants. Smartphones were the primary access point for users.
The training videos included a scenario, a relatable character and a “what’s in it for me” factor. For example, if performing a task with SOPs saves the farm money, it’s not as motivating for entry-level employees as realizing that SOPs makes their job easier. Training modules also include spaced repetition to reinforce the information.
When Heuweiser began using AI to make slides and other work, he realized he “had a hard time learning to write good prompts that increase the probability of getting good outcomes,” he said. “The most important part is to learn prompting is how a kid learns something.”
He encouraged attendees to “make mistakes and learn from them. Ask the large language model for feedback. You can ask a large language model to refine it.”
Heuweiser always uses Grammarly to check spelling first and then engages with the large language model AI system. He tends to use AI for prompting, summaries, study limitations, persona and stories.
For effective prompting, Heuweiser starts with experimenting, exploring, creating and allowing himself to make a few mistakes. Next comes analysis and refining the prompt. That kind of feedback helps him prepare for fine-tuning the structure of the work, such as its role, task, context, examples, output and constraints. He encourages using only verifiable sources.
When using AI, it’s important to be specific to what one wants, such as SOPs, research abstracts, training scripts or emails. “Actional key takeaways” may become summaries.
Heuweiser offered a prompt engineering framework:
1. Define the role, the expert persona, audience and communication style to calibrate exactly who is responding to whom.
2. State the task, which specifies the goal using an action verb and sub-objectives to tell the model exactly what it needs to achieve.
3. Provide context as the single source of truth by containing the necessary documents, data and background details.
4. Show examples, such as what “good” looks like through few-shotting to improve consistency and tone.
5. Define output. Surgically define the response’s formatting, length and structure.
6. Apply constraints, which establishes guardrails, style rules and boundaries to explicitly define what the model must avoid.
“I’m excited about the capabilities about AI uses in our field but I’m also frightened about what’s going on,” Heuweiser said. He has concerns about AI diminishing young people’s ability to develop critical thinking skills.
