The HDG Survey Group expects artificial intelligence (AI) to play a considerable role in quality control. Many users, though, are not ready to take this step. AI's quality is also not yet sufficient. "We've noticed that available applications can't record quality adequately and correctly. Also, digitization isn't as advanced everywhere in the world as in Europe. To create awareness in other countries, you sometimes need an intermediate step in their growers' digitization process," begins HDG Director of Innovations & Analytics Bas Lok.
That is why this European-based company focuses on properly designing the first step: data collection. Here, the quality of the data is vital. "Data collection is crucial. If it is done incorrectly, the data becomes less valuable, lowering the value of a prediction made by an AI application. So you must ensure people in the field know how to use the software properly."
HDG Survey Group, thus, initially focuses on basic analyses as a stepping stone to more advanced solutions. "Dashboarding is becoming increasingly important for growers to gain insight into how a product develops over time. However, not all growers have the money to buy the most expensive analysis software," says Bas.
"By offering basic analyses – which you can break down by things like quality score, variety, grower, and origin – we expect growers will gradually move into deeper technological development. That's the first step towards much broader analyses and a better foundation for improving food quality and safety."
Collecting photos
Bas explains that the company also focuses on preparing to introduce AI technology. "Take, for example, assessing whether a banana is green or yellow. It takes many photos to teach the AI system how to determine the exact color based on a photo. We're currently cataloging all these photos per observation. Once the right technology is available, you can upload thousands of photos directly into the system and it can be up and running quickly. Our inspectors have 40+ years of experience, and we, therefore, have the right experts in-house to teach AI with no errors," he says.
Error avoidance is critical when teaching AI because the margin of error directly relates to trust in AI, Bas notes. "AI is still in its infancy, so it's hard to trust. Currently, there are hardly any field tools geared towards variable conditions and always predicting correctly with 80 to 90% certainty. That's why things are done in such a way that corrections can be made so the report is still correct. Ideally, though, you want reports that are always 100% correct."
Intervening with the correct know-how
Tweaking an AI system is risky, Lok points out. "You need people with the right know-how to intervene in the system. If you make the wrong corrections, and that happens too often, it leads to the system making incorrect assumptions. That will only make the AI system worse," he explains. You, therefore, need expertise to teach or correct an AI system. "If you don't use the right expertise, there's a risk that the reports the system generates will become increasingly unreliable. And AI reports no longer being trusted because they're incorrect too often, poses a risk to the entire fruit and vegetable supply chain."
"That's what AI still lacks: insufficient expertise built into functionalities that people can directly apply. If a tool produces a negative report, people could say, 'I don't trust AI technology.' That's the most important thing right now: trusting this type of technology and the various software parties that (will) offer it," Bas reckons.
Coordination in the internal chain
That is why he especially sees companies that control large parts of the chain as potentially using AI for quality control, at least in the short term. "A company that, say, cultivates and imports, can apply this technology much more quickly. Using AI for quality control internally builds mutual trust. I believe using it in this way will take off in the next two years, but it will probably take much longer for the rest of the market. I think it will be more like five years or even longer before individual growers, importers, and retailers will truly widely apply AI technology."
Bas thinks that calls for more and better communication throughout the chain. He sees this necessity as being separate from AI, for instance, to monitor the consequences of the ever-decreasing crop protection product package. And to guarantee long-term food quality and safety. Financial margins should be part of that discussion.
"Small margins may not have an immediate short-term impact, but it does mean growers won't be able to invest in what's needed in, for example, five years. That could jeopardize food quality and safety. Ultimately, much more communication is needed chain-wide to ensure the quality and safety of food can be maintained at the level we all want in the long run," Bas concludes.
For more information:
HDG Survey Group
Tel.: +31 10 2441414
lok@hdg-surveygroup.com
www.hdg-surveygroup.com