Session industrie
Visual AI in agriculture: opportunities and challenges
Hassan-Roland Nasser — Agroscope
Visual data are becoming ubiquitous in agriculture, driven by the widespread availability of low-cost imaging systems, drones, and satellite observations. This rapid expansion of visual data creates unprecedented opportunities to automate animal and plant phenotyping, monitor production systems over time, and extract actionable insights at scales that were previously unattainable. In this talk, we present a series of applied research projects in which visual artificial intelligence has been successfully deployed to address concrete challenges in agricultural systems. These examples illustrate how computer vision and deep learning enable non-invasive monitoring, improved decision support, and increased efficiency across livestock and crop production. Beyond current successes, we also discuss the key barriers that still limit large-scale adoption, including data availability, model robustness in real-world environments, and integration into operational workflows. Finally, we outline how closer collaboration between research institutions, technology providers, and industry stakeholders (particularly SMEs) can help close these gaps and accelerate the practical deployment of visual AI in agriculture.
When AI Meets Reality: Successes, Setbacks and Surprises
Olga Popovych — ELCA Informatique SA
Across organizations, interest in AI is high. However, resources, data maturity and change management capacity are often limited. Many organizations launch pilots without clear problem definition, realistic expectations or alignment with operational processes, resulting in prototypes that look promising but never reach operational scale. This presentation is built on a conviction that sharing failures is just as important as sharing successes. Only by examining what did not work (missed assumptions, integration hurdles, user adoption issues) can organizations avoid repeating the same costly mistakes. For instance, we will share: - How we successfully managed unrealistic expectations for an NGO - How we were challenged by lunatic LLMs on an external facing AI assistant - The importance of data quality and security to avoid surprises