New Trends in Food Safety – Artificial Intelligence and Predictive Analytics

 Artificial Intelligence in Food Safety

Artificial intelligence and Predictive Analytics are immerging in the food industry in decision making as interrelated technologies that are starting to have an enormous impact on the supply chain and whose impact will be growing over time. These technologies are most heavily used in the food supply chain sectors that have much historical data already captured and processed, such as farming and its related agronomic activities. The use of Artificial Intelligence in food safety is beginning to allow for more complex data analysis to be performed, where resulting conclusions have shown a less problematic and reduced requirement for verification. The confidence in the results allows businesses to make bolder decisions and move on to more ambitious goals. On the other hand, as confidence increases, less reliance on the use of personnel for data analysis and perhaps reduction in the workforce. Hence, the verification and approval process has to be human in order to decide what food is to be eaten or not and to keep quality, food safety, and regulatory requirements complied, where QA departments may have more busy days ahead.  Hence, access to food safety insights enables businesses and professionals alike to identify, monitor, and prevent any increasing risks or incidents that need global attention across the complete supply chain.
 
Nevertheless, predictive analytics has shown a bright future ahead with a more accurate confidence range when incorporated with AI, allowing businesses to take a more transparent and focused decision to proceed or alter direction. Thus, incorporating AI and Big data with PA is much more readily assessed and deciphered into business decisions leading to more successful business decisions. The modern food safety landscape is rapidly diversifying to allow the companies with better AI and PA systems to take the lead, where an onboard computer is becoming much more important than the engine or driver for success.

The actual capabilities of AI and predictive analytics are yet to be revealed. Still, they have exciting potential, where massive amounts of data are collected publicly and proactive surrounding food production, manufacturing, imports, exports, and more. However, taking all this data and utilizing it in productive ways is a massive effort that AI has incredible capabilities, where AI and predictive analytics will allow the food industry to act proactively against potential threats. Modern food safety practices are based on preventative actions; however, food safety issues are often already affected within the supply chain before the threat is known. On the other hand, AI and predictive analytics can be used to determine the factors that indicate threats to the food supply and allow threats to be stopped before an incident occurs more accurately. Thus, allowing the food industry to act proactively against potential threats.
 
Another interesting area of AI development can be seen in food contamination, with the potential consequences of pathogenicity and/or spoilage, which is a critical control point in food processing plants, where throughput is high and varied.  Hence, many niches for microbial growth may exist, which can impact large numbers of consumers in the event of a contamination. The microbial species practically do not exist in isolation, where many species can be found together, forming a community that is known as a microbiome. Understanding the microbiome's composition is the key to understanding the risk from microbial contamination, including specific organisms and their numbers. Many bacteria or other microbes which are non-culturable, either because they are unknown or they are not recoverable using current growth media and conditions and so are intractable to conventional techniques, can be assessed through DNA sequence data, which can be obtained without the need for culturing the microbes beforehand. Thus, predictive analytics can be used to achieve this, which overcomes the culturing problem and provide a catalog of the bacteria present in the particular microbiome under investigation.
 

Empirical data has shown that the presence of particular bacteria in a microbiome can create conditions conducive for the growth of other organisms, where the makeup of the microbiome can be predictive of the appearance of organisms of concern such as pathogens or food spoilage microbes. As a matter of critical importance, many processes used in food processing are used to kill spoilage and pathogenic micro-organisms. Thus, the ability to model those processes, the process variations, and the variations in microbial death can be achieved more efficiently from artificial intelligence. Hence, the benefits of combining simulation and machine learning (ML) techniques in food manufacturing will open up new headaches for manufacturers in the near future.
 
Artificial intelligence builds upon food risk assessments as the structure for evaluating data to make predictions, but it doesn't take the place of food risk assessments and don't answer prevention questions. As such, big data analytics and machine learning are very effective at frequently identifying that something has happened but are less effective at explaining why. That's where a risk assessment needs the human approach to enrich data insights and create tailor-made intelligence to achieve greater visibility across the global food supply chain. Hence, AI exposes questions that need to be answered more than providing answers. Further,  AI is used as a tool for enhancing the safety of foods in the supply chain, which will use risk assessments and data collected to predict outcomes and expose questions leading to task and system adjustments that result in safer foods, lower defects, earlier identification of problems, etc. all that results in safer food. Nevertheless, simulation is usually an efficient technique to make difficult problems more tractable that can contribute to elaborating new algorithms, supporting decision-makers, decreasing the risk in investments, and running the systems exposed to changes and disturbances more efficiently.
 

The presence of AI in the food supply chain is in an early stage right now, where the scope of impacts will grow as adoption grows. However, the use of AI on the inspection end of the supply chain shows tremendous promise for removing the compromised product from supply chains at ports of entry in a targeted way that allows identification of compromised products even in the midst of limited inspection capacity, where the role of AI in identifying compromised product early in production will grow as data collection models improve and gain acceptance/implementation. On the other hand, it appears that AI is most effective in the context of specific systems rather than generalizable across systems. Therefore, the structure of the system (i.e., risk assessments, operational plans, equipment specifications, etc.) will provide the foundation for AI within a workflow. The data sets needed will depend on the workflow in question, such as manufacturing line, processing facility, re-packing operation, etc., where the identification of key data points and time intervals within a workflow will drive the data collection needed to build robust AI capabilities.
 

Predictive analytics will be a new path for the access to quality, processed, applicable datasets used in building AI models, where AI, machine learning, and deep learning all require a large amount of information used for training, testing, and deployment to produce a positive result. Access to the appropriate data will be the prime accelerator or impediment to AI's widespread adoption in any sector. Further, Data and AI are two sides of the same coin, where availability of structured data will be a decisive factor in the adoption of AI, and the adoption of AI will drive the need for more data. Hence, adoption of AI and predictive analytics will be easier with companies who are able to successfully collect the sector-specific private datasets, use them in conjunction with public models, and commercialize these tools to solve sector-specific challenges. If more data is available for a given problem, AI methods can better perform, and the accelerant or impediment to widespread adoption will be how quickly these datasets can be established and become commonplace. Thus, the application of better Data sharing mechanisms and building methods for various stakeholders to feel safe sharing their food data will be accelerators to the development and subsequent adoption of AI and predictive analytics for food risk prevention.
 
For the development of a comprehensive risk assessment strategy, methodologies should be devised and tested for sensitivity, specificity, and accuracy, which can be explained in an example as a homology detection can be performed for inferring (toxic) properties from well-characterized biopolymers since homologous proteins can share similar structural architecture and functions. The obtained data can also be used in conjunction with already available and well-established methods in protein science, where additional strategies can be derived using the gathered data that is based on the use of multiple alignments, specific domains, and/or molecular signatures as well as hidden Markov models (hMMs)-oriented approaches. Artificial intelligence approaches could be used to assess whether a protein belongs to a family that is known to contain toxic proteins in order to try to increase the accuracy of predictive risk assessment for proteins.
 

AI and PA can create significant impacts on the accuracy, speed, and consistency of decision making globally, irrespective of the individual knowledge and experience, by allowing data-driven and sound decisions along the supply chain such as procurement, including predictions on available quantity and price of agricultural raw materials due to climate change or unexpected weather disruptions, full traceability to the farm, automated link of process equipment settings with product quality characteristics, optimizing preventive maintenance, consumer behavior, and preferences, etc. Difficulties in modeling production processes are manifold as a great number of different machining operations, multidimensional, non-linear, stochastic nature of machining, partially understood relations between parameters, lack of reliable data, etc. Some examples for applying simulation and Artificial Intelligence techniques for different fields of manufacturing can be given as modeling, simulation, and optimization of production processes and process chains, design, control, and reconfiguration of flexible manufacturing systems (FMSs), design and control of holonic manufacturing systems (HMSs).


References:
https://www.bigdatagrapes.eu/sites/default/files/Agroknow-BigDataGrapes-Discussion_Paper.pdf