AI as a Catalyst for Cross-Disciplinary Breakthroughs in Soil Carbon Sequestration Research

Abstract

AI models have been proposed for generating scientific hypotheses; thus, our aim was to test their ability to drive novel research in soil science. We used an AI multiagent platform to generate research ideas for (innovative) practices capable of increasing Mineral-Associated Organic Carbon (MAOC) in soils. We assigned the AI multiagent system Manus two tasks: a general research-generation task and a specific task that required cross-disciplinary approaches. For the general task, the AI proposed well-documented strategies such as no-till farming, crop diversification, integrated crop-livestock systems, and organic and other amendments. More notably, the cross-disciplinary task generated novel ideas from materials science, bioengineering, chemistry, medical science, physics, marine science, geology, and computer science. The AI system prioritized three research areas: (1) Engineered mineral surface modifications to optimize carbon binding, (2) Controlled-release carbon delivery systems, inspired by medical drug delivery technologies, and (3) Biomimetic mineral engineering mimicking high-carbon natural environments. We critically assessed the proposals and determined that while some are plausible and align with concepts in soil science, others offer the potential to open new research avenues through interdisciplinary collaboration. Our findings suggest that AI can generate “outside-the-box” hypotheses and help test new scientific ideas, demonstrating its potential to drive innovation in soil science. We suggest a workflow for using AI for hypothesis generation to ensure scientific rigour and epistemic responsibility.