With approximately 2 billion people affected by food insecurity currently and estimates only worsening as climate change intensifies, artificial intelligence and machine learning in agriculture promise significant benefits. Already, AI and ML have been used for the in situ assessment of soil composition, disease diagnosis, prediction of yield, post-harvest handling and traceability and more.
Despite this, in a new perspective published in Nature Machine Intelligence, a team of international researchers is urging caution and consideration of all potential risks before new technologies move forward.
“The idea of intelligent machines running farms is not science fiction. Large companies are already pioneering the next generation of autonomous ag-bots and decision support systems that will replace humans in the field,” said first author Asaf Tzachor, leader of the Global Food Security project at the Centre for the Study of Existential Risk (University of Cambridge). “But so far no one seems to have asked the question ‘are there any risks associated with a rapid deployment of agricultural AI?’”
In their paper, the global team from the UK, Colombia and Nigeria raise the alarm regarding data acquisition, access, quality and trust. While national and international agricultural research institutions collect abundant data that can in principle support ML models, the researchers say these data are too often not discoverable, interpretable or reusable.
As with any network, the digitization of agriculture opens the door to cyberattacks, including ransomware and denial-of-service attacks, as well as interference with AI-driven machinery, such as self-driving tractors and combine harvesters, robot swarms for crop inspection, and autonomous sprayers. Just last year, for example, a huge cyberattack was launched on JBS, the world’s largest meat processor, as well as NEW Cooperative, which provides feed grains for 11 million farm animals in the United States.
The scientists envision the adoption of advanced technologies worsening the socioeconomic inequities that already permeate global agriculture, including gender, class and ethnic discriminations and child labor. Small-scall farmers, specifically, will be at a heightened disadvantage, the researchers say.
“Small-scale farmers who cultivate 475 of approximately 570 million farms worldwide and feed large swaths of the so-called Global South are particularly likely to be excluded from AI-related benefits,” the researchers write in their paper. “Marginalization, poor Internet penetration rates and the digital divide might prevent smallholders from leveraging such advanced technologies, widening the gaps between commercial farmers and subsistence farmers.”
Unintended consequences also need to be taken into account. For example, an AI system programmed to deliver the best crop yield in the short term might ignore the environmental consequences of its endpoint, leading to overuse of fertilizers and soil erosion in the long term. Overapplication of pesticides in pursuit of high yields could poison ecosystems, while overapplication of nitrogen fertilizer would pollute the soil and surrounding waterways.
To navigate these risks and more, the researchers suggest initial deployment of agricultural AI in low-risk hybrid cyber-physical spaces, which they refer to as “digital sandboxes.” These sandboxes would allow multiple stakeholders to supervise prototyping and piloting of novel techniques, ensuring they are safe and well-secured before moving forward. The space can also be a safe playground for technology suggestions—both those that work and those that do not.
“Anonymizing data relating to failed deployment attempts and sharing it with agricultural AI communities will allow lessons to be learned and accelerate safe and secure innovations,” the research team concludes. “[Digital sandboxes] can also help inform rules and regulations for rolling applications out responsibly. Government agencies could give special, interim exemption to…digital sandboxes before developing targeted, customized regulatory frameworks.”