Converting agricultural products into food requires a mixture of precise engineering and exacting science. Along the way, food processing companies generate vast amounts of data. Food safety programs require data. The efficient function of processing equipment and logistical systems also require data. Statistical analysis is a common way to analyze the day-to-day data, the results of which can remain internal to the operation or be summarized for other purposes. Food safety-related data is used to monitor and document processes. What if there was a different way to look at the data, though? What if the totality of data could be examined quickly and efficiently?

The goal of this new approach would be to use data analytics to discern some pattern, trend, or association that remains hidden when using normal statistics, but to do so in real time or near-real time.

The descriptive term used for extremely large data sets is "big data," which can be computationally analyzed to create insight. The concept of big data has been around for decades. There are three types of big data. All are collected and analyzed in different manners:

1. Structured data (e.g., quantitative data, dates, times, serial numbers, etc.)
2. Unstructured data (e.g., machine data, text, email messages, survey responses, transcripts)
3. Semi-structured data (e.g., graphs, tables, PowerPoints, transcripts, video, audio, XML documents, etc.).

Collectively, all three types of data share five characteristics, including:

• Volume: There are vast amounts of data resident in the databases of every food processing plant and within every food-related corporation, which are often running multiple processing facilities and varied food products.
• Value: There are hidden gold nuggets scattered in the resident data, but these insights must be discovered, and then extracted and processed, so that they can serve operations and decision-makers.
• Variety: Many varieties of big data can be examined individually or as an aggregate of the whole. Food safety-related big data could, for example, be analyzed for its own value, or it could be combined with other types of data to determine if there are any larger patterns, trends, or associations. For instance, what is the return of investment on each implemented food safety-related decision? Structured, unstructured, and semi-structured data can also be combined in new ways and analyzed as an aggregate.
• Velocity: Big data can be analyzed historically, but with the right technology it can also be analyzed in real time, or near-real time, depending on the need. Food processing plants are increasingly filled with sensors that are gathering data at incredibly fast speeds. Traditional analytical techniques are often inadequate for discovering immediate answers.
• Veracity: Big data follows the old computer adage of "garbage in, garbage out." Data must be validated as correct and real (meaning unaltered, or reflecting "ground truth"). This feature is an increasingly important requirement because of the growing threats of cyber adversaries, who are targeting not just Information Technology (IT) and Operational Technology (OT) systems, but also sometimes maliciously altering the data collected.

Insider threats are an increasingly serious business concern that can deleteriously affect food product safety, physical security (i.e., food defense), and brand quality. Real time or near-real time data could, for example, be used to identify when an anomaly occurs in the system, perhaps providing evidence of a disgruntled employee seeking to adulterate an ongoing process step.

What is common across all of the examples given is that traditional methods for data analysis are inadequate to accommodate the speed and breadth of challenges found in pre- and post-harvest domains today. One solution for dealing with the data is to utilize artificial intelligence (AI) to assist those charged with analyzing the data.

Application of AI to Food Safety Data Analytics

AI refers to the simulation of human intelligence processes by computer systems. It encompasses a range of technologies and techniques that allow machines to perform tasks that typically require human intelligence, such as learning, reasoning, problem-solving, and understanding natural language. In the context of big data analytics for food safety, AI can be categorized into the following varieties:

• Machine learning (ML) algorithms enable systems to learn and improve from experience without being explicitly programmed. They can analyze large datasets and identify patterns, anomalies, and trends relevant to food safety. For example, ML can be used to predict potential foodborne illness outbreaks based on historical data and environmental factors.
• Natural language processing (NLP) focuses on enabling computers to understand, interpret, and generate human language. This is useful for analyzing text data from sources like social media, customer reviews, and regulatory documents to identify potential food safety issues and sentiments expressed by consumers.
• Computer vision technology allows machines to interpret and understand visual information, such as images and videos. In food safety, it can be used to inspect food products for contaminants, to detect spoilage, or to monitor food processing operations for quality control.
• Deep learning is a subset of ML that uses neural networks with many layers to perform complex tasks. Deep learning models can be used to improve image recognition in food quality inspections and enhance predictive models for food safety incidents.

Big data analytics, when combined with AI, enable the food industry to better monitor, detect, and prevent food safety issues. A few examples of how the different varieties of AI can be used in food safety big data analytics, using technologies available today, include:

• Predictive analytics: AI can analyze historical data on food safety incidents, such as outbreaks, recalls, and inspections, to predict potential future issues. Machine learning models can be trained to identify patterns and risk factors, helping authorities and food manufacturers take proactive measures to prevent problems.
• Quality control and inspection: Computer vision and deep learning can be used to automatically inspect food products for visual defects, contamination, and spoilage. Cameras and sensors in production lines can identify irregularities in real time, helping ensure product quality and safety.
• Supply chain transparency: AI can help track and monitor food products throughout the supply chain by analyzing data from various sources. This ensures traceability and enables quicker responses to any potential contamination incidents.
• Sentiment analysis and social media monitoring: Natural language processing can analyze social media and online content to identify emerging food safety concerns and consumer sentiment. This helps authorities and companies respond to public concerns and implement necessary measures promptly.
• Regulatory compliance: AI can assist in monitoring and ensuring compliance with food safety regulations. By analyzing data from inspections and audits, it can identify non-compliance issues and trigger corrective actions.
• Personalized food safety recommendations: AI can provide personalized food safety recommendations to consumers based on their dietary preferences, allergies, and health conditions, ensuring that they make safe food choices.

AI, with its various forms available today, can play a crucial role in enhancing food safety through big data analytics. It enables more effective and efficient monitoring, detection, and prevention of food safety issues, ultimately reducing the risk of foodborne illnesses and improving the overall safety of the food supply chain.

A Future View of AI

The future of food safety will continue to be increasingly influenced by AI. As AI technology innovations increase in frequency, what is available today will likely be very different from what lies in the future.

The U.S. government is prioritizing biosurveillance, which is a concept that the food industry will likely be hearing repeated, given that the food supply is an important component in larger public health. The U.S. Department of Homeland Security describes biosurveillance as the focused development of effective surveillance, prevention, and operational capabilities for detecting and countering biological threats.

Biological threats can be naturally occurring or intentional. In this way, biosurveillance is important not only for the purposes of protecting public health (i.e., prevention of foodborne illness) but also public safety (prevention of bioterrorism, intentional food adulteration, etc.). The fortunate thing about AI-assisted biosurveillance is that current AI-assisted surveillance systems can be adapted, further scaled, and serve both purposes. In this way, a single biosurveillance system can guard against conventional food safety problems, as well as deliberate attempts to adulterate food products.

Sensor-Driven Data

In the future, comprehensive, AI-assisted biosurveillance systems could be included in every stage of the food chain, from pre-harvest to post-harvest, through the logistical chains (e.g., cold chain, produce, etc.), thereby providing true "farm-to-fork" surety. Sensors could also be "layered" at ground level and at altitude (carried by drones), and even located in space.

In March 2023, in the Federal Register, the U.S. Department of Agriculture proposed that official identification devices for cattle and bison must be both visually and electronically readable. Multiple Radio Frequency Identification (RFID) technologies are being utilized, as well as GPS-enabled devices that will meet the requirement. Premises Identification Numbers (PIN), which are assigned to a geolocation, are required to receive official identification devices. These devices allow the creation of data sets that are available for analysis that can help establish source and age, detect reproductive status and disease conditions, measure movement and feed intake, and generally create pattern-of-life data sets for cattle and bison. This data is valuable for trade, food safety, biosurveillance, and food security.

Each sensor at each point of intersection with an agricultural product that is destined to become food would generate its own set of data. This data could then be aggregated across the entirety of the commercial enterprise and analyzed using AI; tuned to search for specific patterns, correlations, and findings of known aberrations (e.g., a contaminated product); or directed to search for any new anomalies in processes or outputs that might not have been previously encountered.

The real-time AI findings could then be transmitted for review by system operators and subject matter experts (SMEs), who could then interpret and confirm their meaning or refocus the AI analytics. Not all correlations have meanings of importance; therefore, it is critical that humans not be removed from the analytical process. A human analyst—using reason, judgment, and experience—is needed to assess and determine the significance of any finding. Within the AI-empowered analytical system, structured data, unstructured data, and semi-structured data could also be brought together to assist the analysis.

An example of such a data fusion would be if the systems-related data indicate a similar anomaly occurring at multiple corporate locations, which is subsequently noted within the internal company messaging systems. Trigger warnings could then be transmitted rapidly across the corporation, notifying other system operators of the problem that might affect them at some point. Cyber systems already commonly use such alerting procedures. AI can help push alerts in real-time or near-real time.

Next-Generation Food Safety Programs

Food safety programs have the potential to be revolutionized using AI. In a hypothetical example, Pathogen X is a human food safety problem that occurs in animal food products but is also shed in animal feces, which could in turn contaminate produce. Large arrays of ground-level sensors could be placed throughout agricultural production systems, whether in confinement rearing (chickens or hogs), pastures or feedlots (beef and dairy), or fields and orchards for plants.

What would these sensors detect? Many types of sophisticated sensors are available for multiple kinds of applications. Hyperspectral imaging (optical signatures) is currently being developed within post-harvest food processing food quality and safety programs.¹ The sensors enable food products to be rapidly optically scanned as they pass through the various process stages, checking for both quality and the presence of foodborne pathogens or other contaminants.

Another type of sensor uses spectral analysis to analyze effluents or environmental contaminants. Tuned properly, these sensors could be used to identify Pathogen X before it entered into the human food chain, enabling remediation at the source. Imagine the utility and cost savings of having a system that would identify food safety or animal or plant health problems in production agriculture. Likewise, airborne and space-based sensors could be used to track the movement of pathogens through the environment.

One example might be to use these sensors to search for pathogens that are shed in poultry houses. Rather than looking at the chickens on an individual production-house basis, airborne sensors, combined with space-based sensors, could enable monitoring locally or, more largely, on a regional basis. The amount of data collected in such an array would be staggering. No human could make sense of the volume or speed, making AI essential. Every data point would be precisely geospatially linked by AI so that data users would know the exact location of each potential problem, not only speeding up its identification, but also perhaps offering options for remediation, using case history.

Such systems sound like they would be confined to government use, but given the increasing availability of commercial technology, AI-assisted, comprehensive sensor systems could be owned and managed by individual companies, or alternatively by a commodity sector, housed perhaps in an Information Sharing and Analysis Center (ISAC).

Whether on an individual-business or sector-wide basis, data streams would be proprietary information just like any other business-related information. If collected by the sector, data could be segregated by company or potentially by a larger consortium of companies. From a non-legal expertise perspective, this data would not appear to violate antitrust laws since it would be collected for the public good, furthering public health and safety. Like other food safety-related data, it could also be used to document due diligence and good practice.^a

Given that food and agriculture are deemed a critical infrastructure, the data generated by such comprehensive sensor systems could be protected, if shared with the non-regulatory portion of the government (e.g., the Department of Homeland Security's Cybersecurity and Infrastructure Security Agency, or CISA). This protection would be effected using the category of information encompassed within the Protected Critical Infrastructure Information Program.²

AI is Only a Tool and Not Infallible

AI does not enable omniscience. It is rapidly advancing, although the technology is like all computer systems—only as good as the inputs and the algorithms used to drive the analysis. Any new food safety-related application using AI needs to roll out in stages. AI systems need to be "trained," meaning humans look at the output and then confirm or correct the system based on the results. Initially, all AI data conclusions would need be examined from the perspective of skepticism and further validated through conventional means, such as statistical analysis. Bias and error can also be unintentionally or intentionally inserted into the AI analytical structure. Given the relative newness of AI in post-harvest systems, one must guard against error. AI is not a panacea, regardless of the claims of its proponents. It cannot replace human decision-making. An excellent review of AI's use in food safety is available in Kristen Altenburger and Daniel Ho's article in Food Safety Magazine.³

In summary, it is accurate to say that AI-assisted analytics is a rapidly evolving science that, if nurtured with care, can be used to better protect public health and lessen the probability of foodborne illness, while also protecting the bottom line and brand quality. Care must be taken in adopting this technology, but once in place and properly tuned, AI can assist users in learning more about their own systems and products, providing the opportunity to identify not just hidden problems but also opportunities for discovering new efficiencies. If you are not already using AI, you will be soon.

References

1. Photonics Spectra. "The Food Industry’s Appetite for Hyperspectral Imaging Grows." June 2021. https://www.photonics.com/Articles/The_Food_Industrys_Appetite_for_Hyperspectral/a66946.
2. U.S. Department of Defense, Cybersecurity and Infrastructure Security Agency. "Protected Critical Infrastructure Information (PCII) Program." https://www.cisa.gov/resources-tools/programs/protected-critical-infrastructure-information-pcii-program.
3. Altenburger, Kristen M. and Daniel E. Ho. "Artificial Intelligence and Food Safety: Hype vs. Reality." December 16, 2019. Food Safety Magazine. https://www.food-safety.com/articles/6416-artificial-intelligence-and-food-safety-hype-vs-reality.

Note

• a. The authors are not lawyers or regulatory experts; therefore, this assessment should not be construed as definitive, but instead as opinion. Readers are urged to speak with corporate counsel for definitive determination of legal status of information related to the collection and analysis of said data.

Robert A. Norton, Ph.D., is a Professor and Coordinator of National Security and Defense Projects in the Office of the Senior Vice President of Research and Economic Development at Auburn University. He specializes in national security matters and open-source intelligence, and coordinates research efforts related to food, agriculture, and veterinary defense.  

Marcus H. Sachs, P.E., is the Deputy Director for Research at Auburn University's McCrary Institute for Cyber and Critical Infrastructure Security. He has deep experience in establishing and operating sharing and analysis centers including the Defense Department's Joint Task Force for Computer Network Defense, the SANS Institute's Internet Storm Center, the Communications ISAC, and the Electricity ISAC. 

Cristopher A. Young, D.V.M., M.P.H., Diplomate A.C.V.P.M., is a Professor of Practice at Auburn University's College of Veterinary Medicine and an Adjunct Professor at the College of Veterinary Medicine at the University of Georgia's Department of Pathology. He received his D.V.M. from Auburn University's College of Veterinary Medicine in 1994. He completed his M.P.H. at Western Kentucky University in 2005 and is a Diplomate of the American College of Veterinary Preventive Medicine.