Focus on Food Safety: Using AI in food safety management

Artificial intelligence (AI) is radically shifting food safety and quality management in food manufacturing. As companies come to understand and apply AI within their workplaces, there’s an inevitable curiosity about the technology’s possibilities. SAAS companies will launch new food safety applications within the near future, but for now, most users are using standard generative AI platforms for single queries.

A big part of successful results in generative AI is the design of prompts and questions. For example, asking AI, ‘What are the risks found in hot dogs?’ will result in a generic response, as opposed to a more specific query: ‘What microbial risks may be present in the industrial manufacturing of hot dogs prepared with ground chicken, water, spices, sodium phosphate and sodium nitrite?’ More details lead to better quality responses.

Along with my students, I tested if generative AI platforms, such as ChatGPT, Microsoft Copilot and Google Gemini, can be used for routine food safety tasks. This was a fun activity, and worth trying out with your own work teams. For reference, we imagined making pasteurized, shelf-stable apple juice in glass bottles.

Hazard identification

Query: What are the potential food safety risks related to the manufacturing of apple juice packaged in glass bottles?

The quality of responses varied widely across platforms. Using AI in this case helped with brainstorming and thinking about potential hazards. AI could be used to start a table top risk assessment process. It is still essential for personnel performing risk assessment to review incoming ingredients and perform site-specific risk assessment related to the processing of the product.

Writing SOPs

Query: Prepare a standard operating procedure for measuring the Brix value in pasteurized apple juice using a prism refractometer.

Some of the Gen-AI tools did a remarkable job on this task. When preparing SOPs, writers block is a challenge. AI can speed up the initial writing process, and then allow manufacturing technologists or quality technicians to do equipment or site-specific editing to ensure compliance.

Drafting HACCP plans

Query: Prepare a HACCP plan for pasteurized shelf-stable apple juice packaged in a glass bottle. The apple juice will be pressed from fresh apples. We are writing using the SQF code as our HACCP guidance document.

One large language model followed the core principles of HACCP in its structure including the development of a theoretical process flow, and use of Codex decision-making processes for identification of CCPs or process controls. However, within the risk assessment and decision-making tree process, hazards were not allocated to elements within the code manual; responses were generic. Another Gen AI tool did not use the classic HACCP template, but the identification of operating prerequisite programs and process controls was detailed.

For simple products AI can speed up the process and help create generic models. None of the tools had the ability to map out a HACCP code manual for operational prerequisite programs within the hazard analysis decision-making process. The primary utility here is to kickstart HACCP program development. Human intervention is still needed for site-specific description of processes, and evaluation of feasible and appropriate process controls.

Identifying microbial criteria

Query: What microbial criteria should we test for in pasteurized apple juice packaged in a glass bottle?

We were disappointed by all the responses. While it’s not exactly a microbial criterion, only one tool identified measuring patulin as part of the protocol. A range of pathogens that are not exactly relevant to pasteurized apple juice, such as Listeria monocytogenes, were suggested. It would be far more useful to work with a process authority and food testing laboratory possessing microbiology expertise than using AI.

While AI has a lot of advantages for speeding brainstorming and drafting basic documents, there is still a need for human expertise and intervention. While the technology matures, human insight is key for successful food safety management.

Dr. Amy Proulx is professor and academic program co-ordinator for the Culinary Innovation and Food Technology programs at Niagara College, Ont. She can be reached via email at aproulx@niagaracollege.ca.

This column was originally published in the Aug./Sept. 2024 issue of Food in Canada.