Generative AI Surpasses the Limits of Geological Language

Austrian philosopher Ludwig Wittgenstein famously stated that “the limits of language are the limits of my world.” As geoscientists, the limits of our understanding of the subsurface can be viewed as the limits of our understanding of surface geological processes. In other words, rocks are our language.

Our limits are becoming seemingly boundless with generative artificial intelligence and new ways to use machine learning and deep learning. The business of oil and gas, and all subsurface energy, has dramatically expanded in the last few years, thanks to generative AI and super-powered machine learning and deep learning made possible with the new computing capacity enabled by innovations in chips, storage and algorithms.

Many of the new opportunities were discussed in last month’s conference, “Generative AI, Machine Learning, and Analytics for Subsurface Energy,” held Dec. 10-11, in Houston. Despite being located in Houston, the use cases were global. Operators of all sizes, researchers, solution providers and service companies discussed the ways that generative AI and machine learning help them turn pain points into real solutions.

Please log in to read the full article

Austrian philosopher Ludwig Wittgenstein famously stated that “the limits of language are the limits of my world.” As geoscientists, the limits of our understanding of the subsurface can be viewed as the limits of our understanding of surface geological processes. In other words, rocks are our language.

Our limits are becoming seemingly boundless with generative artificial intelligence and new ways to use machine learning and deep learning. The business of oil and gas, and all subsurface energy, has dramatically expanded in the last few years, thanks to generative AI and super-powered machine learning and deep learning made possible with the new computing capacity enabled by innovations in chips, storage and algorithms.

Many of the new opportunities were discussed in last month’s conference, “Generative AI, Machine Learning, and Analytics for Subsurface Energy,” held Dec. 10-11, in Houston. Despite being located in Houston, the use cases were global. Operators of all sizes, researchers, solution providers and service companies discussed the ways that generative AI and machine learning help them turn pain points into real solutions.

Highlights from the Conference

Kim Padeletti of AWS and Rocky Roden of Geophysical Insights kicked off each day with historical perspectives of just how powerful the paradigm shift has been already, and how AI is already accelerating the process of exploring for and developing oil and gas reservoirs. AAPG President Deborah Sacrey, who has been finding oil and gas for years using machine learning, has embraced the new capabilities of the new models to find even more oil and also to provide more accurate valuations for reservoir engineers.

AI is also making it possible to find oil that is left behind in legacy fields, either behind pipe, or in offset locations by reprocessing vast amounts of historical production, petrophysics, well log and other data, according to both NuTech’s Galen Dillewyn and S&P Global’s Toby Burrough. Sometimes the new resource discoveries are possible because of data that has been previously “dark” – hidden away in offices, storage units and garages in the form of paper, mylar and microfiche reports, maps, well logs, mud logs, drillstem tests and other non-digital media. Sabata’s founder Bryan McDowell has tackled the repositories in order to scan them, even when hauling the heavy paper records translate to hours in a U-Haul, traversing Texas cattle country.

Legacy data is one thing, but what about real-time operations? Chevron’s Jenni LaRue and NthDS’s Mike Ramirez focused on how AI is essential in fast-paced, automated geosteering, and DWL’s Dave Tonner described how AI enables sample collection robots to operate efficiently, as well as how they also make the sample collection process safer. Finally, Fervo’s Sirish Dadi described how generative AI and machine learning applications convert streaming real-time data to monitoring and automation functions for utility-scale geothermal power generation.

Let’s think again about how the limits of our understanding of the subsurface can be viewed as the limits of our understanding of surface geological processes. Surface images from satellites and drones can give us time-sliced snapshots of sedimentological processes in near-shore and offshore depositional environments. Time-series surface images can also give us intriguing insights into the impacts of erosion, aeolian transport, active faults and even dissolution (in the case of sinkholes). At universities such as Texas Christian University, AI is now being used to merge this highly detailed surface geology information to develop more detailed subsurface reservoir and basin models. These new subsurface models, report TCU’s John Holbrook and others, make older models look like cartoons.

Generative AI makes the new information and models quickly usable across a wide array of needs. It is possible to merge the surface and subsurface geological information with digitized reports, maps and tables to generate simulations, answer questions and evaluate regions for their resource potential, which can include oil and gas, along with geothermal, critical minerals-rich reservoir fluids, geologic hydrogen, carbon storage, and more.

Hacking the Hallucinations

Can you trust the results of gen AI? The answer is “not without a subject matter expert.” This is where domain expertise is a must. One of the most important skills of the future will be the ability to ask the large language model very specific questions that can be used to obtain accurate answers and also to continually refine the model as well as to identify where data may be flawed or miscategorized. Learning how to ask questions of all kinds – even questions deliberately designed to provoke “hallucinations” – will be a part of the training that all geoscientists will take as they probe the foundation models and the data integrity checks.

ThinkOnward’s Nate Suurmeyer and Mik Isernia enjoy leading Hackathons that give geoscientists the chance to produce their own AI hallucinations because it’s in that process that one starts to understand just how the data interacts, the patterns that become preferential, and the thrill of victory when you have outwitted the app (and in doing so, learned how best to work with it).

When Wittgenstein published his “Tractatus Logico-Philosophicus” in 1922, he was interested in the intersection of language, perception and meaning-making processes. His work came to form the cornerstone of Postmodernism, an exploration in literature, film and art, of making order from chaos, then turning the corner and making a chaos of order, just to reconfigure it again. The Greek god Proteus became emblematic of the constant metamorphoses, the multiplicity of interpretive possibilities. Now, with generative AI, the limits of data and the imagination are the limits of the geoscientist’s world.

You may also be interested in ...