What does it take to teach AI to understand the complexity of the subsurface?
TGS is exploring that question by developing a multimodal Subsurface Foundation Model on AWS, bringing together artificial intelligence, geoscience expertise and TGS’s extensive subsurface data.
In a conversation with AWS host Ali Khoja, TGS Data Scientist Meher Gajula explains how the model is being developed to learn complex spatial relationships across 3D subsurface data and apply that knowledge to a range of interpretation tasks.
Building a more complete view of the subsurface
A key part of the approach is connecting different types of geoscience information, including seismic and well data, to work toward a shared subsurface “world model.”
The model can support interpretation tasks such as fault inference, salt segmentation and channel identification across seismic volumes.
Scaling subsurface AI
Developing models of this scale requires the infrastructure to efficiently access and process large volumes of data. Meher and Ali discuss how technologies including Amazon S3, MDIO and advanced GPUs support model training and inference.
The conversation also looks at the potential role of AI agents in research, including helping researchers assemble datasets, run experiments and explore new model architectures.
Creating new opportunities from data
For clients, the opportunity extends beyond the model itself. Combining TGS data and models with operators’ own information and expertise could provide new ways to investigate uncertainty and identify potential prospects.
The work highlights a broader opportunity for AI in geoscience: using technology to accelerate the analysis of increasingly complex datasets while keeping geoscientists at the center of interpretation.
Watch the full episode to hear how TGS and AWS are working to advance AI for the subsurface.