A Multi Modal Geologist Copilot GeoCopilot: Generative AI with Reality Augmented Generation for Automated and Explained Lithology Interpretation While Drilling

In well drilling operations, the rapid interpretation of geological data is crucial for optimizing drilling processes, ensuring safety, and understanding the characteristics of geological formations and reservoir fluids (Blue et al., 2019). Traditionally, these analyses depend on cuttings description, a manual and non-deterministic procedure carried out by teams of geologists in the field, combined with the analysis of drilling parameters and logging-while-drilling (LWD) data when available. However, characterizing cuttings samples to describe well lithology is both time-consuming and prone to human bias at various stages, from sample preparation to the actual description. Using it poses a challenge both to the traditional method used while drilling, as well as to incorporating this kind of information into any automated or semi-automated workflow that uses Artificial Intelligence techniques. Recent advancements in Machine Learning (ML) and Artificial Intelligence (AI) have shown promise in enhancing data reliability and real-time lithology prediction. The early explorations by Rogers et al. (1992), Benaouda et al. (1999), and Wang and Zhang (2008) laid the groundwork, utilizing well-log data to develop predictive models. As the field advanced, more refined ML models for lithofacies and permeability prediction emerged, employing techniques like artificial neural networks (ANN) and support vector machines (SVM). Researchers such as Mohamed et al. (2019) and Nanjo and Tanaka (2019, 2020) applied ML models and image analysis methods to address real-time lithology prediction during drilling operations. Recently, Khalifa et al. (2023) achieved a remarkable accuracy of 95% in identifying some lithologies with an ML-base approach, demonstrating significant advancements in real-time ML workflows for lithology prediction. However, the new advances of AI, more specifically in the field of Generative AI (GenAI) and Large Language Models (LLMs) have not yet been explored in such applications. And although GenAI faces its own set of challenges such as data scarcity, interpretability issues, scalability, and trustworthiness, it might offer a new frontier for further enhancing lithology prediction and assist in optimizing drilling operations. Therefore, the purpose of this paper is to advance the field by validating a methodology that integrates GenAI, LLMs, with geological data for assisting in the description of cuttings samples and interpreting lithology while drilling.