Automating expert decision-making is not the same as supporting it. A case on AI-driven climate control in Dutch greenhouses shows that when systems have to work alongside years of grower expertise success depends on model accuracy as much as whether growers retain control, understand how decisions are made and can trust what the system tells them.
Saheli De, an expert from the field, shared her experience implementing CropController, an AI system for autonomous climate and irrigation control in Dutch greenhouses. She discussed the case during a recent KINTalk, a series of events organized by the KIN Center for Digital Innovation to connect practitioners and academics.
The case highlights an increasingly important challenge. As experienced growers approach retirement, organizations face the risk of losing valuable tacit knowledge about how crops respond to changing weather, humidity, temperature and other conditions. CropController seeks to complement this expertise by combining localized weather forecasts and historical greenhouse data with strategies defined by the grower. The system then anticipates future conditions and calculates how the greenhouse climate can be adjusted accordingly.
Yet the talk showed that technical performance alone is not sufficient for adoption. Growers need to understand and trust the system, particularly in a context where poor decisions can directly affect crop quality, disease pressure and yield.
Several design principles proved important: maintaining grower agency, using explainable models, providing transparency through dashboards, tailoring the system to each greenhouse, and developing a shared language between data scientists and growers.
The case illustrates a broader point about AI implementation: successful adoption is not simply a matter of automating expert decision-making. It requires designing systems that can interact meaningfully with existing expertise, practices and professional judgment.