Artificial Intelligence

AI in the food industry – what you should know about the technology and its application possibilities

Interview with Professor Stephan Schneider and Anna Dhungel

Artificial Intelligence (AI) – hardly any other topic attracts so much attention, is being discussed as much and yet is so misunderstood as this field of computer science. Time to shed some light on the subject. We asked two experts what AI is, what it actually can do and where its limits are. The interview focuses on the application possibilities of AI in the food industry.

Stephan Schneider is professor of business administration and business informatics at the university of applied sciences in Kiel. His research area covers data science, AI as well as business analytics and intelligence. In her master’s thesis, Anna Dhungel investigated how AI can be used to predict the crop yield of bell peppers in greenhouses. The research project was a cooperation between the university, agiles and the producer association Reichenau Gemüse.

There’s not the one definition of AI 

The topic AI is on everyone’s lips. However, it’s often not quite clear what the popular buzzword actually means. Could you define the term for us?

Anna: Let me back up a little. There’s not the one definition of AI. We already lack a uniform understanding when it comes to human intelligence. That’s why we can’t define machines behaving like humans as intelligent. The approach to consider machines as intelligent when they do something only humans could do, also is controversial. In my opinion, the ability to learn is crucial. Processing data and drawing conclusions about future behavior applies to humans and AI.

Furthermore, a distinction is made between narrow and general AI. Narrow AI refers to machines used in specific areas, digital language assistants, for example. General AI is what we know from science fiction movies, i.e. robots capable of everything. It’s discussed if and when we will reach that point. Currently, we only use narrow AI, i.e. machine learning models and algorithms that are trained for special purposes.

What do you think, Stephan?

Stephan: Anna explained that very well. Of course, it’s unsatisfying from a scientific point of view if you can’t define something exactly. It helps to consider the two terms artificial and intelligence separately. Artificial is defined as a man-made artefact, different from natural systems. A common definition of intelligence was given by the US-American psychologist David Wechsler. He defined intelligence as the “aggregate or global capacity of the individual to act purposefully, to think rationally and to deal effectively with his environment”.

Among other things, intelligence means information processing through perception and emotions, executing and accessing certain cognitive processes and thinking in an abstract, associative, analytic, creative, analogous and adapting way. It’s crucial that the retrieved information is useful. It is if you act goal-oriented. With regard to machines, the question is which type of information processing and actions they must be able to perform to be considered intelligent.

Stephan Schneider is professor of business administration and business informatics at the university of applied sciences in Kiel

Stephan Schneider is professor of business administration and business informatics at the university of applied sciences in Kiel.

The spearhead of AI are artificial neural networks

Many people associate AI with science fiction, but it actually already is part of our everyday lives. Where is AI currently used? Could you give a few examples?

Anna: I can think of digital language assistants, chatbots or bots in general, autonomous driving, image recognition, face recognition or fault detection.

Stephan: That’s again related to how we define AI. Of course, you tend to associate science fiction with it, something with a wow effect, according to the motto “it’s magic”. However, from a scientific and strictly definitional point of view, even the simplest IT systems performing context analysis are AI. For instance, AI is used in the section “Customers who bought this item also bought” on the Amazon website. The application of these association rules is intelligent behavior. Most of us don’t associate that with AI and find it hard to consider it as AI.

AI is a very broad field. What exactly is your research focus?

Stephan: My research focuses on the so-called spearhead of AI, i.e. artificial neural networks. Machine learning plays a crucial role here. Although the terms are fuzzy and I’m reluctant to use them for academic reasons, a division into three mayor fields is common in practice. First, regression, second, classification and third, clustering.

Regression deals with metric dependent variables, for example forecasting the sales of a product. In classification, certain objects are assigned to already existing groups. In clustering, there are no groups yet, but they will be revealed. Target group analysis is an example. Artificial neural networks can learn how to place certain features in order to create homogenous groups. Our research covers all these areas.

An artificial neural network predicting the crop yield of bell peppers

In one of your current research projects, you investigate how to predict crop yield using artificial neural networks. Could you tell us more about it?

Anna: In my master’s thesis, I trained an artificial neural network to predict the crop yield of bell peppers produced in greenhouses. I experimented with different architectures and with the data to be considered. The results of our final model were good. The question with these problems always is which data to use. Nowadays, a lot of data is recorded, for example by modern climate control computers. However, it’s not necessarily the best approach to take this bunch of data and to feed it into the network. You have to find out which data influence the results and which data the network needs to learn.

What are the basic requirements for producers to be able to use neural networks for predicting crop yield? 

Anna: In any case, you need to know the harvest quantities of the last years for the relevant area. After all, that’s what we want to predict. Which other data you consider has to be discussed. Of course, climate data such as temperature and humidity can be helpful, maybe also the pH-value and the hours of sunlight. However, often the horticulturists know best which factors are decisive.

Certain management data can also be helpful, for example if the number of flowers or fruits is counted. That’s something the horticulturists always monitor anyway. Another requirement that should not be underestimated is the openness of the staff on site to the technology. Germany is known to be rather skeptical about technology, especially when it comes to AI. You can build the greatest models, but they also need to be taken into account in praxis. In our case, however, people are very interested.

AI can be used to predict the crop yield of bell peppers

AI can be used to predict the crop yield of bell peppers.

The application possibilities are gigantic

Besides predicting the crop yield, are there other applications of AI in the food industry that will play a role in the future? 

Stephan: Yes, the application possibilities are gigantic. We need to broaden our understanding of how to use AI. The prediction of crop yield is only one example. Similarly, you can forecast the sales volume of product ranges in food retail or wholesale. The product sales, including production and delivery, involve the end customer, i.e. the entire supply chain is covered. Of course, in the case of products such as bread or cheese, other factors have to be considered.

You might also want to go one step further and analyze how price changes of one product affect other products. Are there any cross price elasticity or cannibalization effects? For example, there is no point in lowering the price of apple juice if its sales increase, but people buy fewer other beverages. With this campaign, you would shoot yourself in the foot. Revealing all these effects is a gigantic playground for AI.

In what way can producers and the food industry in general benefit from AI?

Stephan: Producers have to deliver minimum quantities. Retail and wholesale have to ensure that the products requested are in stock. This represents gigantic possibilities for intelligent behavior since there is not only an economic, but also an ecological and sociomoral obligation. You want to produce as few rejects as possible. Surplus perishable food could be donated to people in need.

Anna: One of the horticulturists said that they have to buy packaging material without knowing how much they will actually need. The following season, the packaging looks different and the old plastic packaging is left over. That’s something you could optimize.

Stephan: In my experience, the protagonists realize that a multilateral network is opening up. You can no longer look at individual factors such as sales volume or pricing in isolation from other quantities because you have to take the whole supply chain into account, from producer to final customer.

Are there any other benefits at the social or environmental level?

Anna: Especially in the food sector, the governments of countries where the food supply of the population is not always guaranteed could eventually use AI to determine in time when there will be a shortage and react faster. They could thus prevent famines affecting parts of the population.

No AI without humans

Humans are the limiting factor, chance, and risk in AI

Humans are the limiting factor, chance, and risk in AI.

In your opinion, what can AI do and where are its limits?

Stephan: Of course, this depends again on the definition of AI. Personally, I believe that AI finds its natural limit in the human factor. This might not be true without restrictions. I go out on a limb and say: currently, a reasonable AI can’t exist without humans. Humans are a very important driver of AI and more than mere embellishment. I would perhaps even go so far as to say: no AI without humans.

Anna: I would even say that this is also the risk. After all, it’s us humans who decide which data will be used. These data have a concrete impact on how an AI decides and which conclusions it draws.

Humans as a risk. Could you explain that in more detail?

Anna: For example, American courts employ an algorithm that predicts whether a defendant will reoffend. It has already been shown that this probability is rated significantly higher for Afro-Americans. That’s because a society in which racism occurs structurally also generates data containing that bias. If we feed AI with it, it will draw the corresponding conclusions. The AI did not directly request the ethnicity of the defendants. There are many similar cases, for example facial recognition which doesn’t recognize differences between Asian people. Not to mention the topic of women and discrimination in the field of AI (you can read more about women and AI on Anna’s blog).

What do you think, Stephan?

Stephan: Of course, humans are a very limiting factor, not only as users but also as developers of AI. Many companies take credit for buzzwords like AI, but that doesn’t make them experts. Experience and know how play an important role. Anna already mentioned examples showing how algorithms work. There are also more harmless examples from the field of image recognition. For instance, people were surprised how incredibly well an AI recognized horses. Then they found out that it hadn’t learned the body shapes or facial structures, but the watermark of the producer, which was on all horse pictures. What I want to say is: experience and expertise play an important role to avoid these types of mistakes.

If we knew which role AI would play…

Are there any other difficulties we face when it comes to AI?

Anna: Something I observe is that the legislation is lagging behind with respect to AI. Of course, that’s in the nature of things because everything is developing very fast and new use cases are emerging all the time. The legislation can’t define and resolve everything in advance. However, this can also give cause for concern, like for example in the case of Cambridge Analytica. They used AI to analyze Facebook user profiles. The extent to which this is allowed is not clear. We do have rules for data privacy, but the issue is so complex that it’s hard for individuals to figure it out. That legislation can keep up is an issue that we need to keep in mind.

If we look 20 years in the future – what do you think: Which role will AI play in the food industry?

Anna: Phew. In the past, people gave terribly wrong answers to this kind of question. In the early nineties, I believe, someone said that the internet was only a trend and that it wouldn’t catch on. Difficult. I don’t know. Stephan, what do you think?

Stephan: This reminds me of a documentary about the beginnings in Silicon Valley, in which different protagonists had their say, including a friend of Steve Jobs. He said that Steve Jobs asked him whether he could lend him 1000 dollars. In return, he would sign over 10 percent of Apple to him. The friend thought about it for a long time and finally decided not to lend Steve Jobs the money. We know the rest of the story. If we knew which role AI would play…

Thank you, Anna and Stephan, for this interesting interview. We wish you a lot of success and fun with your future research projects. 

More on the use of AI in the food industry

Are you looking for a modern software solution tailored to the needs of food production and trade? We are happy to help.

Contact us!