Enhancing trust in agri-food AI: insights from the human-centred explainable AI masterclass ○ S+T+ARTS Hungry EcoCities

As part of the Hungry EcoCities consortium, a masterclass titled “Human-centred explainable artificial intelligence” was delivered last summer to support the project’s aim of creating a more healthy, sustainable, and ethical agri-food system. The session, led by Robin De Croon from the Augment research group at the Department of Computer Science at KU Leuven, addressed the critical challenge of making complex algorithmic decisions transparent and actionable for diverse stakeholders in the food sector.

Moving beyond the “black box”

The masterclass established that while technical accuracy is essential, it is insufficient for the successful adoption of AI in real-world scenarios. The session distinguished between a “narrow” definition of explainable AI (XAI), which focuses on making models understandable to developers, and a “broad” human-centred definition. This broader approach encompasses the entire user experience, ensuring that the output of an AI system provides the necessary context and reasoning for non-technical users, such as farmers, nutritionists, or consumers, to make informed decisions.

The importance of trust and actionability

A key focus of the presentation was the relationship between explainability and trust. For domain experts in the agri-food sector, the ability to “explain and be convinced” by an output is a prerequisite for adoption. If users cannot verify why a system recommends a specific intervention, whether it be a crop management strategy or a dietary suggestion, they are unlikely to trust the tool.

To address this, the masterclass outlined several design strategies to bridge the gap between algorithmic complexity and user needs:

● User-centred design: moving away from the assumption that “I am the user” and instead relying on observation of actual workflows and habits.

● Actionable explanations: ensuring that AI outputs do not just describe a prediction but empower the stakeholder to take specific actions.

● Interactive modalities: utilising visualisations and conversational interfaces to allow users to explore data and ask “why” questions, rather than presenting static, opaque results.

Relevance to Hungry EcoCities

These principles were relevant to the Hungry EcoCities “Paths to Progress” experiments. As the project integrates digital technologies within the agri-food sector, ensuring compliance with frameworks like the EU AI Act and overcoming user resistance is vital. The masterclass underscored that by adopting a human-centred approach, technical innovations can be transformed into reliable support systems that respect the professional judgement of experts and the values of consumers.

The slides from the session, detailing these frameworks and providing case studies on visualising health and energy data, are available below.

---