Subsurface Intelligence

AI will not replace geoscientists. It will change how they think.

AI agents can now read well reports, check logs, screen acreage and draft evaluations in hours. Subsurface Intelligence asks the question that matters to every exploration team: where do these tools fail, and how would a geoscientist know?

This is not a book of opinions. Each claim is tested. Using real well data from Kansas and offshore Norway released under open licences, synthetic seismic with known answers, rock-physics and basin models, and Monte Carlo simulations with every assumption stated, the book follows agents across the full exploration workflow: data preparation, seismic interpretation, well evaluation, basin and play analysis, prospect risking, portfolio ranking and bid-round screening.

Inside you will find:
How a single misread label cut a model's accuracy from 97 to 86 percent, and why no warning appeared.
Why a fault detector that scores well on its training data traced channel margins as faults.
How a predicted shear log can reverse the AVO signature of a gas sand.
Why shared play risk raises the chance of a dry five-well programme from 17 to 42 percent.
How the winner's curse inflates winning bids even when every estimate is honest.
A practical protocol for challenging, calibrating and auditing an agent's work.
What data-ownership and data-protection rules mean for AI use in the Middle East and North Africa.

Every chapter ends with a methods statement so results can be reproduced, and a case study describes building an offline regional screening system, including what went wrong.

Written for exploration geoscientists, managers and investors, Subsurface Intelligence shows which work can safely move to agents, which must stay with people, and how to tell the difference.