Deep learning needs human and ecological touch to improve agriculture
Deep learning needs to go beyond automatic data analysis to better unlock potential gains for sustainable agriculture, according to a group of experts.
In a review for Journal of Sustainable Agriculture and Environment, researchers led by Masahiro Ryo at Leibniz Centre for Agricultural Landscape Research argued that bringing ecological knowledge and human experience into deep learning approaches can improve their usefulness, in the face of complex challenges.
In particular, the biological dynamics of soil are “critically understudied”, while despite high awareness of the possible impact of agricultural practices on neighbouring ecosystems, the authors found no work using deep learning to study such impacts.
Human knowledge can build trust
Expert knowledge on ecology and agronomy is necessary to improve the accuracy of models, they noted, while there is also more scope to incorporate farmers’ decision-making processes.
“How to address non‐digital information, especially expert‐based knowledge and opinions on ecology and biodiversity, in deep learning is key,” the researchers wrote.
They call for these technical and social issues to be addressed in order for deep learning to increase its potential to foster sustainability in agriculture.
“Many deep learning methods were just technically validated in the scientific domain, and we guess few farmers use them for daily use because they do not see the benefit or know how to make use of it,” they continued.
Greater trust in artificial intelligence needs to be fostered more broadly, they suggest, with better education and user-friendly apps.
Agricultural AI not ready for consumer involvement?
“I agree with the general idea that we should include users of technology and subject experts early and often into our research. This is often really hard to do,” said Ed Harris, senior lecturer in statistics at Harper Adams University.
However, he pointed to digital marketing as an area where there has been considerable human input in work by major companies such as Google and Microsoft, driving AI advancement, including that which might be used for agriculture applications.
He also suggested a need for sharper distinction between applied AI research, usually done by engineers for scientists or ag-tech industry, and real-world AI, potentially done after scientists are finished and which should definitely involve technology consumers. At the current stage of development, such involvement would not necessarily be expected, he said.
“Most of this technology has not yet matured to a state to enable feedback from real, expert humans like smart, tech-savvy farmers,” he added. “It is a great idea, but we are not quite there yet.”
Read the full review in Journal of Sustainable Agriculture and Environment.
This article first appeared on Farming Future Food
