JKMRC Friday Seminar 2026: Learning Earth’s Hidden Structure with Subsurface World Models

JKMRC Friday Seminar 2026: Learning Earth’s Hidden Structure with Subsurface World Models

by Gary Oliver

Applied Sciences & Engineering Physics PHVApplied physicsPHVGGeophysics
🎙 Gary Oliver 👥 6K 📅 May 11, 2026 ⏱ 60 min 👁 126 📄 expert opinion 🧭 2026-08-16
Available in: English (current) Français

Keywords

world modelsmineral explorationsubsurfacegenerative AIgeophysics

Summary

Gary Oliver, Chief Scientific Officer at Fleet Space Technologies, presents a seminar on subsurface world models for mineral exploration. He begins by acknowledging the traditional owners and introducing Fleet’s work in seismic technology. He highlights the increasing demand for copper and critical minerals, driven by AI data centers, and notes the declining discovery rates and increasing depth of deposits. He critiques current AI applications in geology, citing examples of ‘slop’ and the limitations of LLMs in 3D spatial understanding. He discusses the challenge of sparse labeled data in mineral exploration, contrasting with self-driving and robotics industries that use synthetic data and world models. He introduces the concept of subsurface world models, which are generative AI systems trained on synthetic geological models and geophysical simulations to infer 3D subsurface structures. He reviews prior work, such as Noddy models and diffusion-based approaches, and outlines challenges including realism vs. scalability, heterogeneous constraints, and computational efficiency. He emphasizes the need for uncertainty quantification and suggests that the industry must build specialized models itself. The talk is conceptual, with few results, but provides a compelling vision for AI-assisted mineral exploration.

185 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable insights into the application of AI to mineral exploration, highlighting the limitations of current methods and proposing a novel framework. The argumentation is coherent, drawing parallels with robotics and self-driving industries. However, it lacks concrete results and empirical evidence, making it more of a conceptual proposal. The speaker’s industry experience lends credibility, but the absence of detailed case studies or quantitative analysis weakens the argument’s strength.

Scientific Rigor, Source Quality, Title Accuracy

The presentation demonstrates a good understanding of the field, but the scientific rigor is moderate. The speaker references a few works, such as Noddy models and a diffusion-based paper, but does not provide detailed citations or URLs. The title accurately reflects the content. The talk is based on expert opinion and industry experience rather than peer-reviewed research, which limits its scientific rigor. The adequacy between title and content is high.

155 words

Title / Content Match

The title accurately reflects the content, which introduces the concept of subsurface world models for mineral exploration.

Quality & Reliability

7/10

Presentation by an industry expert with relevant experience, but lacks detailed citations and peer-reviewed references. The talk is conceptual and forward-looking, with limited empirical results.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • Noddy models (Mark Jessell et al.) — Mentioned as prior work on synthetic geological models for machine learning training.
  • Diffusion-based geological modeling paper — Referenced as a recent example of using diffusion models to generate geological models from constraints.

Concurring Sources

  • Noddy models (Mark Jessell et al.) — Supports the idea of using synthetic geological models for training AI.
  • Diffusion-based geological modeling paper — Aligns with the proposed approach of using generative models for subsurface inference.

Contribution & Novelties

The talk introduces the concept of subsurface world models, a novel application of generative AI to mineral exploration. It proposes a framework that combines sparse hard constraints (drillholes) with dense indirect observations (geophysics) to infer 3D geological structures while quantifying uncertainty. This approach addresses the challenge of sparse labeled data by using synthetic models for training. The talk also highlights the need for the industry to develop specialized models rather than relying on general-purpose AI.

Pour aller plus loin :

108 words

Radar Profile

The radar profile shows moderate scores across all dimensions, with slightly higher scores in information quantity and quality, reflecting the conceptual depth but lack of empirical results. The low technical level score suggests the talk is accessible to a broad audience, while the moderate reliability score indicates a need for more rigorous sources.

Reliability 6/10

💬 No comments were provided for analysis.