KANSAS CITY, MISSOURI, US — Formulating and manufacturing animal feed requires vast amounts of data, the sort of task artificial intelligence (AI) was created to feast upon and digest into usable information that could address universally shared challenges within the feed industry, such as quality and efficiency.

AI supporters herald a new era of data application that draws on exponential computational speeds linked with humanlike thought processes to produce original outcomes. The feed industry operates in a space of tight profit margins challenged by a desire for improved quality, energy efficiency, sustainability and workforce readiness. 

Industry professionals who spoke to World Grain see AI as an exciting, albeit early step, toward even greater innovation for the roughly $600 billion global feed market.

“The feed industry is a pretty traditional industry, with a long history, often not as quick in the adoption of fancy new solutions like AI,” said Philipp Hug, global head of animal nutrition at Uzwil, Switzerland-based Bühler Group. “But given the size of the market, still quite some early adopters are open enough to improve the feed industry (through AI use).”

The ultimate proof of the value of AI or any new technology is found in a company’s profit and loss statement, said Ben Allen, chief executive officer of feed supply chain technology company BinSentry, based in Kitchener, Ontario, Canada.

“Technologies that improve efficiencies and cut costs are the most likely avenues for adoption,” Allen said. “Teams that use a lot of spreadsheets and manual processes for scheduling, procurement, ordering, etc., will find quick returns by offloading tasks to AI.”

Dirk Maier, professor of agricultural and biosystems engineering at Iowa State University in Ames, Iowa, US, said expectations for any new technology should be managed.

“What is realistic? What should we be counting on? Where do we wait for something to come our way versus needing to be proactive as an industry?” he said. “I think those are all things that we are going to continue to work through over the coming years.”

It’s those very questions optimists might anticipate asking AI itself in the very near future.

Defining AI

The concept of AI goes back several decades to the earliest computers, when programmers pondered the very idea that a machine could replicate human thought patterns and intuitively learn new skills or generate unique ideas. As the processing power of computers grew, so did expectations for AI technology.

The 2022 emergence of OpenAI’s ChatGPT, an AI system that uses a large-language model to engage users in humanlike conversation and churn out topical written documents, seemed to jump-start global excitement and questions about its potential.

However, ChatGPT and other language models are but one subset for AI.

According to international technology firm IBM: “AI in its simplest form is a field that combines computer science and robust datasets to enable problem-solving. It also encompasses subfields such as machine learning and deep learning. Generative AI refers to deep-learning models that can take raw data from multi-data sets and ‘learn’ to generate statistically probable outputs when prompted. At a high level, generative models encode a simplified representation of their training data and draw from it to create a new design or approach that is similar, but not identical, to the original data.”

At the International Production and Processing Expo (IPPE) in January, G. Leo Schilling, senior scientific services manager for Eurofins, a global leader in scientific testing services based in Luxembourg, presented “Artificial Intelligence — The Next Wave of Innovation for Animal Feed and Livestock Production.”

Schilling listed four categories of AI:

  • Reactive machines: AI systems that simulate real time and are designed to respond to inputs that can be put into action.
  • Limited memory machines: AI monitors objects and situations over time.
  • Theory of mind: Third-level AI that allows machines to understand and respond to mental states like humans.
  • Self-aware machines: AI systems that understand their existence.

“Through analysis of complex datasets, AI can optimize production that leads to improved efficiency,” Schilling said. “AI automates repetitive tasks, reducing human error and ensuring consistent quality. AI analytics enable proactive problem-solving in production by predictive analysis and allow real-time monitoring and control systems that further enhance accuracy.”

AI in play 

Feed manufacturing is a competitive industry with tight profit margins. Raw materials prices, including grains, extraction meals and oil cakes, fluctuate. Utilizing locally sourced products and byproducts and an emphasis on sustainability create greater efficiency variables for pelleting. A well-educated, experienced workforce has become harder to sustain, particularly in rural areas. 

All of these issues offer AI an opportunity to advance what current robust automation and digital technologies currently address. In his IPPE presentation, Schilling said there are many animal feed and livestock production players in the market developing sensors and monitoring systems and pushing cost reduction and innovation.

“Automating animal feed production processes involves using AI systems for precise ingredient mixing and formulation,” he said. “These systems optimize feed composition based on animal requirements. AI-enabled sensors and cameras monitor animal well-being, including health and welfare.

“Efficient inventory management is also vital for smooth supply chains. One example of this is that AI forecasting accurately predicts customer demand.”

Allen said AI soon will be applied to ingredient inventory tracking and procurement challenges that directly impact formulations and profit margins. He noted that most ingredient bins today are still checked manually by humans, which provides poor datasets at irregular times. AI will give feed manufacturers the ability to monitor ingredient levels 24/7 and coordinate ingredient purchasing and feed demand, allowing for formulations that are more consistent from a cost perspective.

BinSentry recently received a new patent for a method to use AI to accurately determine the volume of animal feed in a bin using a time-of-flight sensor or other similar hardware.

“In the future, this ingredient data will be tied to, and driven by, finished goods data from every feed bin that a mill serves,” Allen said. “This coordinated data stream will reduce mill changeovers and emergency delivery orders, both of which drive operating costs up.”

Gero Zimmermann is head of technology animal nutrition at Büher AG, which provides complete solutions and services for sustainable animal feed production. He sees AI tools that can boost the performance of the feed mill manufacturing execution systems (MES) by their unique capability to deal with nonlinear and multivariate data systems.

“AI can clearly help to continuously optimize the pelleting process,” Zimmermann said. “Furthermore, if AI is applied to all core processes — including mixing, dosing and grinding — it opens up enormous potential in terms of production stability and consistent product quality.”

That consistency could be established across the entire value chain from field to end-use, creating an even tighter process that squeezes better outcomes for profit margins, said Bühler’s Hug.

“AI could, beyond automation, help to capture data in the whole value chain from raw material, weather conditions, rainfalls, etc. to final product, even predict how fast an end user is going to use their feed reserves,” he said. “With this data, one will be able to improve process parameters to almost optimal feed nutrition parameters while processing.”

It is that super-fast, humanlike “observation” quality of AI that intrigues Maier at Iowa State, which recently completed and began operating its Kent Feed Mill and Grain Science Complex. While Maier said the new facility contains state-of-the-art automation technology and processes, it would not be described as employing AI.

Where Maier sees generative AI’s potential is its ability to learn and create from knowledge already available while observing processes at a feed mill. It would apply acquired data and make predictions on how the process could be further optimized and improved beyond what people who are busy with day-to-day operations might be thinking. AI would free up personnel to make the important decisions and create new ideas.

“It would be like, AI is sitting there saying, ‘I have been monitoring your processes and have noticed a pattern in your feed rations, that if you adjust it in this manner that I am suggesting, you would get 10% more efficiency out of your process,’” Maier said. “That is the power of something like generative AI technology.”

Humans remain paramount in such a scenario, making those final decisions aided by new information generated through AI, Maier said. 

While finding skilled employees for the feed industry is challenging and automation has helped, “lights-out” fully automated facilities seem to be a bit farther down the road. Maier noted that even at today’s most advanced facilities, including the Kent Feed Mill, employing predictive machines still requires human programming, observation and action.

“There’s a human being, software engineer that’s programming these control systems and how the sensors are read and how the data is recorded and how a decision might be made based on a sensor reading,” he said. “I don’t think we’re going to see that much in terms of AI replacing that operations technology automation that we’re already capable of doing. The diagnostics of helping people with predictive analytics or maintenance is where AI could be helpful.”

Where Maier sees AI potentially useful in the most human sense is as an interactive repository of knowledge gained by longtime employees that could be accessed by newer workers with questions about how to approach a problem within the mill. Perhaps this new AI could even interpret all that gained knowledge and personal experience in new ways that become more useful.

“In the end, we’re still producing food, fuel and fiber that is going to come from a field that’s going to have to be aggregated someplace, that will have to be processed in whatever form it needs to be for utilization,” Maier said. “These are supply chain physical processes that we, hopefully, can do in the future smarter and more efficiently, but the physical processes still have to happen.”

Allen said that “lights-out” plants with AI that never sleeps, while aspirational, remains over the horizon as the industry takes initial steps toward incorporating AI within its processes today.

“I do believe that AI can help ease the labor challenges that exist within agriculture by taking over repetitive tasks and freeing up human time,” he said. “The hard problems will remain with humans, but coordinating schedules, for example, isn’t really that hard of a problem for AI. If the majority of scheduling can be completed by AI, then human labor can focus on negotiating with vendors and solving complex challenges.”

Zimmermann sees AI-powered and governed mills as an almost necessary development down the road.

“In view of the widening gap between the complexity of compound feed production, instability of supply chains and raw materials quality on the one hand, and lack of qualified personnel on the other, this development is inevitable,” he said.

Challenges to adoption 

The potential for AI within the feed industry to help improve outcomes seems evident to the feed industry. Adoption, however, faces hurdles that include data security, cultivating new supplier relationships and a traditionally conservative industry in which AI will need to relentlessly prove its efficacy before mill owners would decide it is worthy of their financial investment.

Allen, whose career includes expertise in hardware, software, agricultural tech companies and investor relations, said many ag companies run dated or custom software solutions. He said that adding AI would mean developing new relationships with vendors offering software integrations. These vendor decisions will matter as much as the technology decisions.

“Waiting for existing vendors to develop AI capabilities will take too long in most instances, forcing companies to expand their relationships in areas they may not be fully comfortable evaluating,” he said. “This makes project staging vital to controlling risks.”

Maier said another aspect is simply a tendency for any business to take a wait-and-see approach: What can AI really do for mill operations in terms of improving efficiency, sustainability and cost-effectiveness?

“The equipment and technology suppliers are trying to see how much they should be invested in AI-related things as it relates to equipment,” he said. “How much does a mill invest in doing this itself in-house versus waiting for others to develop this, and can I buy that from them? That’s the classic question with automation control and other technology.”

Zimmermann said the advantages of AI are generally well understood, but doubts about its robustness under real-life conditions remain, as do potential shortcomings with security, given the vast amounts of data needed.

Hug noted that top management, middle management and lower management within firms tend to disagree about the levels of usefulness of AI to operations.

“Top management is often pro-AI, just hesitant on security, and middle and lower management does not really believe in AI capabilities,” Hug said. 

Ensuring accurate and quality data collection for AI models is crucial for the feed industry, Schilling noted. Managing privacy and security of sensitive data in AI systems is critical to prevent misuse of that information.

“It requires robust data management systems for training (of AI) that is necessary for relevant and reliable data processing,” Schilling said.

Adaptability will be paramount in this new environment. Smart monitoring and diagnosis, real-time evaluation and optimization of the entire feed process from raw materials to final product is expected to be improved through integrated use of AI in current systems. As businesses become more comfortable, AI heralds the next generation in feed milling, many believe.

“Computer chips follow Moore’s Law, doubling their computing power every 24 months,” Allen said. “This exponential growth has been one of the fastest technology curves in existence — until AI. In some cases, AI capabilities may be doubling every six months, leading to a significantly steeper exponential curve. AI isn’t coming. It’s already here.”