Artificial Intelligence in Archaeological Geophysics: Testing ChatGPT-4o on Magnetometer Data From Sapallitepa (Uzbekistan)

Abstract

Archaeogeophysical prospection methods have become essential for the preparation and implementation of archaeological projects. This study evaluates the potential of ChatGPT-4o, a large language model (LLM), to process grid-based magnetometer data collected near the Late Bronze Age site of Sapallitepa in Uzbekistan (ca. 1600–1200 bce). The aim was to assess whether natural language interaction could serve as a viable alternative to traditional software pipelines such as Geoplot 4.0. Both approaches—Geoplot and ChatGPT—were used to process the same field dataset, resulting in detailed magnetic anomaly maps that revealed subsurface archaeological features. While Geoplot required expert operation and domain-specific configuration, ChatGPT processed the data based solely on textual prompts. In this study, ChatGPT-4o is treated as a natural language interface for applying standard geophysical processing steps. Our findings indicate that ChatGPT, when guided by well-formulated instructions, can deliver results comparable to those of specialized geophysical software. While domain expertise remains essential for meaningful interpretation, such tools could lower the barrier to entry for archaeological data processing. These results suggest a promising future for integrating AI-driven methodologies into archaeogeophysical workflows, improving both accessibility and automation in data analysis.