Fully Funded PhD: Automated, Explainable Geological Interpretation of 3D Digital Twins Under Conceptual Uncertainty

Can AI learn to reason like a human - and recognise when it isn’t sure? This PhD tackles that challenge in a high-impact setting: interpreting 3D digital models of rock formations built from drone and laser-scan data. These models underpin decisions impacting net-zero ambitions, yet interpretation remains slow and subjective. As demand for low-carbon resources grows, we need tools that make geological reasoning faster, more transparent, and more consistent. Current AI systems struggle because they assume a single correct answer, when geological evidence often supports multiple plausible interpretations; so, training AI to copy one expert’s labels fails outside narrow cases.

The project investigates how to build AI to handle uncertainty - representing alternative interpretations, explaining the evidence behind them and supporting experts in making better-informed decisions. You’ll work at the frontier of computer vision, uncertainty modelling and knowledge representation, developing methods that help understand what AI sees, how confident it is and why, while studying how experts trust, challenge and refine AI-assisted interpretations.

You’ll gain expertise in 3D deep-learning, semantic technologies, explainable and uncertainty-aware AI, and human-centred evaluation, and test your work in VRGeoscience’s commercial platform VRGS. You’ll join an inclusive doctoral community with excellent training, collaboration and professional‑development support.

Objectives

This project aims to make geological interpretation of 3D digital outcrop models faster, more transparent and more consistent by treating it as reasoning under uncertainty rather than matching a single “correct” answer. The objectives include: 

• Build a benchmark dataset capturing how multiple experts interpret the same data, including their confidence in each judgement and the reasoning behind it.
• Develop deep-learning and 3D computer-vision models that detect features while representing a distribution of plausible interpretations.
• Encode geological relationships in a knowledge graph that stores alternative readings of the same observations with evidence and provenance supporting each interpretation.
• Study human-AI collaboration and explainability through user studies.