Building AI Literacy

Artificial intelligence (AI) is moving into food R&D, safety, manufacturing, supply chain, and marketing faster than many professionals have been trained to evaluate it. For Hanyu Chen, clinical assistant professor at Purdue University, that creates a practical workforce problem: A company can invest in capable software, but employees still need to know when to trust its output—and how to challenge it.

At the IFT FIRST Annual Event and Expo, Chen shared lessons from developing Purdue’s AI Applications in Food Science and Industry course, which was designed for food professionals rather than programmers. Her central argument was straightforward: AI training needs to build durable competencies instead of teaching employees how to use one rapidly changing platform.

This should be a practical training blueprint, and it should not focus on tools.

Hanyu Chen

Purdue University

1. Teach Fundamentals, Not Tools

AI products and versions change too quickly for workforce education to revolve around a particular platform, Chen said. Food professionals instead need enough understanding of how AI systems work to assess outputs, recognize limitations, and make informed decisions as new tools emerge.

That does not mean turning food scientists into computer scientists. Chen said the training focuses on basic AI concepts and critical evaluation rather than Python or other coding skills. The goal is to help professionals understand what a technology can and cannot do before they rely on it in their work.

2. Fit Training to the Job

AI literacy will not look the same across every food industry function. In R&D, a scientist might use AI to help plan formulation trials. A quality assurance or quality control professional might apply it to risk analysis, while a manufacturing employee could use it to support scale-up decisions and operating parameters.

Chen recommends defining competencies around actual workflows. A team that spends most of its time on formulation, for example, will learn more from formulation-specific use cases than from generic computer-science examples. Purdue’s course follows the food product life cycle, from ideation and safety evaluation through manufacturing, logistics, and consumer-facing applications.

3. Make Data Literacy Nonnegotiable

Knowing how to use an AI interface is only part of the job. Users also need to judge the information behind the output. Chen said food professionals should be able to ask whether they have enough data, whether the data are accurate and relevant, and whether AI-generated information is appropriate to use as evidence.

Those questions matter in food science, where decisions may affect formulation, processing, quality, or safety. Poor inputs can produce convincing answers that are not dependable.

4. Teach People to Question—and Own—the Answer

Critical evaluation also means resisting the temptation to treat an AI response as a finished product. During the session, Chen pointed to fabricated references and hallucinations as familiar examples of why users need to verify what a system produces.

Accountability is just as important. Chen asked who would be responsible if an AI-assisted formulation failed or an AI-supported risk assessment contributed to a bad decision. Her framework puts responsibility on the people who decide whether and how to use the output. Organizations, she said, should make expectations around ethics, risk, transparency, and acceptable AI use explicit from the start.

“AI literacy is not another software training. It is the ability to evaluate the evidence, recognize the limitations, and remain accountable for what happens next.”

5. Learn From Failures

Successful AI demonstrations can show what is possible, but Chen believes failed implementations may be just as instructive. Each Purdue course module includes 5–10 industry use cases from large and small companies, including examples in which AI projects did not work as intended.

“A lot of the failure cases ... it’s not because the software doesn’t work,” Chen said. “It’s because people don’t know what is the best way to implement that software.”

That makes implementation literacy part of the workforce challenge. Employees need to understand not only whether a model is technically capable, but whether it fits the problem, the data, and the workflow where it will be used.

The Purdue course had only limited learner feedback when Chen presented at IFT FIRST. Early participants, she said, valued being able to understand a software provider’s claims and judge whether a tool made sense for their role without becoming programmers.

AI literacy, in that sense, is not another software training. It is the ability to decide whether a tool belongs in the workflow, evaluate the evidence it produces, recognize its limitations, and remain accountable for what happens next.

Explore IFT FIRST Coverage

Discover the latest reporting, interviews, and expert insights.

Read Now