24.01.2025

A recent study conducted by the Leibniz Centre for Agricultural Landscape Research (ZALF) demonstrates how modern artificial intelligence (AI) can drive digitalization and sustainability in agriculture. The study focused on making AI tools accessible for everyday agricultural tasks, such as counting coffee cherries from images, an important step for yield estimation. The tools could also help to monitor pests and diseases in crops and support the reduction of pesticide use. It is therefore important to assess how well they can fulfill these tasks. The results were published in the journal Smart Agricultural Technology.
The data used in the study was collected from approximately 1,000 smallholder farmers in Colombia and Peru, who took pictures of coffee cherries using their own cellphones. This citizen science approach uniquely integrates real-world data from the Global South, offering insights into how AI can address agricultural challenges in diverse environments.
Researchers from ZALF, in collaboration with Brandenburg University of Technology Cottbus-Senftenberg and international partners, examined three approaches: a general-purpose AI model (ChatGPT-4) that can analyze uploaded images with simple questions like "How many cherries are in this picture?", a foundation AI model (T-Rex) designed specifically for counting objects, and a conventional AI method (YOLOv8), which requires annotated datasets and programming but efficiently handles large volumes of data. These cutting-edge methods, particularly ChatGPT-4 and T-Rex, eliminate the need for programming skills or preparing large training datasets, making advanced AI tools more accessible to non-technical users.
Key Findings: New Opportunities for Democratizing AI
The study found that ChatGPT-4 showed potential but required further improvement, as its accuracy was moderate and improved only with user feedback. On the other hand, the foundation model T-Rex delivered highly accurate results, even outperforming a conventional AI method (YOLOv8), which required extensive training data and programming expertise.
This indicates that AI tools like T-Rex could transform agriculture by enabling farmers and agricultural workers to use advanced technologies without technical skills. By simply guiding the model using a mouse to indicate what objects to count, users can achieve highly accurate results with minimal effort.
Societal Relevance and Future Outlook
This research highlights the potential for AI to democratize access to advanced technology in agriculture. Although the study focused on counting coffee cherries, such tools could, in the future, help farmers with tasks like yield estimation, crop monitoring, and disease detection. With further development, this technology could potentially contribute to chemical usage planning and supporting biodiversity in agriculture by enabling enhanced disease detection. Advancements in multi-purpose AI or foundation models could, in the future, make these tools more accessible to farmers, supporting sustainability efforts such as yield prediction, targeted pest management, and reducing reliance on synthetic chemical pesticides. By promoting practices that minimize chemical impacts on ecosystems, these technologies could play a crucial role in preserving biodiversity and fostering environmentally friendly agricultural systems.
By integrating real-world data from the Global South, this study demonstrates the value of combining citizen science and AI to tackle agricultural challenges in diverse and resource-limited contexts. Eliminating barriers like programming and extensive data preparation could help farmers worldwide to benefit from AI-driven solutions, opening the way for more efficient, more inclusive agriculture.
About this Text:
This summary was created with the assistance of artificial intelligence based on the original text: Konlavach Mengsuwan, Juan C. Rivera-Palacio, Masahiro Ryo (2024): ChatGPT and general-purpose AI count fruits in pictures surprisingly well without programming or training. Published Open Access under the CC BY 4.0 license.
Partner institutions: