Comparing large language models and human expertise for fungal mycelium growth assessment in biogeotechnics

The increasing adoption of generative artificial intelligence (GenAI) through large language models (LLMs) in complex image interpretation tasks traditionally requiring specialized expertise necessitates evaluating their effectiveness in geotechnical applications, particularly in fungal growth assessment for soil improvement. This study compares the performance of LLMs and human experts in assessing fungal mycelium growth from microscopic images in sandy soil across different inoculation methods. A systematic comparison was conducted between three LLMs (GPT-4, Claude-3.5, and Llama-3) and five human experts in analyzing microscopic images of fungal growth patterns, evaluated by independent reviewers using standardized criteria. LLMs demonstrated superior performance in standardized observations (87.27%) and terminology usage (89.07%), but showed limitations in comparative analysis (58.93%), while maintaining higher overall consistency than human experts (79.69% vs. 56.87% average performance). Statistical analysis revealed notably higher inter-rater reliability for LLMs (intraclass correlation coefficient (ICC) = 0.640, and p = 0.025) compared to human experts (ICC = 0.392, and p = 0.038), with significant differences in terminology usage ( p = 0.002), observational skills ( p = 0.004), and accuracy ( p = 0.012). However, no significant differences existed in the comparative analysis and objectivity. These findings suggest that assessment of optimal fungal growth can be achieved through a hybrid approach combining LLMs' strengths in standardized analysis with human expertise in complex interpretative tasks. This is particularly valuable for biogeotechnical engineers and researchers. Future research should focus on developing real-time monitoring systems integrated with LLM analysis and comprehensive validation testing across diverse environmental conditions to establish robust automated assessment protocols in biogeotechnical applications.