[Paper Review] Large language models can help boost food production, but be mindful of their risks

Djavan De Clercq, Elias Nehring+2 more|arXiv (Cornell University)|Mar 20, 2024

FinTech, Crowdfunding, Digital Finance4 citations

TL;DR

This paper examines the dual potential of large language models (LLMs) to significantly enhance food production through improved agronomic decision-making, innovation acceleration, and policy support, while cautioning against risks such as misinformation, farmer data exploitation, job displacement, and over-reliance undermining critical thinking. It calls for proactive policy frameworks before LLM integration becomes irreversible.

ABSTRACT

Coverage of ChatGPT-style large language models (LLMs) in the media has focused on their eye-catching achievements, including solving advanced mathematical problems and reaching expert proficiency in medical examinations. But the gradual adoption of LLMs in agriculture, an industry which touches every human life, has received much less public scrutiny. In this short perspective, we examine risks and opportunities related to more widespread adoption of language models in food production systems. While LLMs can potentially enhance agricultural efficiency, drive innovation, and inform better policies, challenges like agricultural misinformation, collection of vast amounts of farmer data, and threats to agricultural jobs are important concerns. The rapid evolution of the LLM landscape underscores the need for agricultural policymakers to think carefully about frameworks and guidelines that ensure the responsible use of LLMs in food production before these technologies become so ingrained that policy intervention becomes challenging.

Motivation & Objective

• To assess the transformative potential of large language models (LLMs) in enhancing agricultural productivity, innovation, and policy-making.
• To identify and analyze direct and indirect risks associated with widespread LLM adoption in food systems.
• To highlight concerns such as agricultural misinformation, farmer data collection, job displacement, and over-reliance on AI for agronomic decisions.
• To urge agricultural policymakers to develop proactive regulatory frameworks before LLM integration becomes entrenched.

Proposed method

• Synthesizing insights from media reports, academic preprints, peer-reviewed literature, startup innovations, and lessons from LLM adoption in other sectors.
• Evaluating LLM performance on standardized agronomy exams, including the U.S. Certified Crop Advisor (CCA) certification.
• Analyzing real-world LLM applications such as KissanGPT, Norm, Bayer’s LLM-powered advisor, and FarmOn for farmer-facing agronomic support.
• Examining risks through analogies from other domains, including automation dependency in aviation and code churn in software development.
• Reviewing emerging legal and ethical challenges, including AI-generated scientific content and intellectual property issues.
• Drawing parallels to publication spam and AI-generated scientific figures to illustrate risks in agricultural research integrity.

Experimental results

Research questions

• RQ1 How can large language models improve agricultural productivity through on-demand expert advice and decision support?
• RQ2 What are the key risks associated with integrating LLMs into food production systems, particularly regarding misinformation and data exploitation?
• RQ3 To what extent does over-reliance on LLMs impair critical thinking and decision-making in farming and research?
• RQ4 How might LLMs impact the integrity of scientific research in agriculture, including publication quality and peer review?
• RQ5 What policy and regulatory frameworks are needed to ensure responsible and equitable LLM use in agriculture before widespread adoption?

Key findings

• LLMs have demonstrated expert-level performance on agronomy certification exams, correctly answering 93% of questions on the U.S. Certified Crop Advisor (CCA) exam.
• Applications like KissanGPT enable farmers in low-literacy settings to access tailored farming advice via voice-based LLM interactions in local languages.
• There is a growing risk of 'publication spam' in agricultural science, with AI-generated figures containing anatomically incorrect or nonsensical content, as seen in the retraction of a paper over AI-generated testicle images.
• Over-reliance on LLMs in coding environments leads to increased code churn and reduced code quality, a trend that could similarly affect agronomic decision-making if not monitored.
• Pilots’ automation dependency in aviation provides a cautionary analogy: over-reliance on LLMs may erode farmers’ situational awareness and hands-on decision-making skills.
• Legal challenges are emerging, including the U.S. Supreme Court ruling that LLMs cannot be listed as inventors on patents, raising questions about IP ownership in AI-augmented agricultural innovation.