How AI is evolving in oil and gas: insights from industry insiders

Production operations are inherently all about physics. The laws of nature govern the dynamics of the physical processes involved in a well, generating massive amounts of structured data. Machine learning is extremely helpful in managing that data to drive value — solving regression-type problems, classification-type problems, time-series forecasting and many other challenges that we face in operations.

Traditional machine learning can also supplement pure physics-based models where needed; for example, if the physical process is too complicated to model in a physics-based simulator due to complex geometry; uncertainties in the model inputs; or when the computation time would take too long and a real-time operational response is needed.

But GenAI can assist with unstructured data needs by automating natural language search and summary, for items such as daily well-work reports; analyzing images of equipment to assist with failure diagnosis; and document classification, such as invoice coding for field work.

So when you combine machine learning and GenAI, you ensure that physics-based constraints are included in your model, since you are using real data generated from field processes. But you are also benefiting from GenAI’s ability to ensure reliability and accuracy over that data.

Our company recently completed an Ops AI trial at a domestic business unit. The majority of use cases for GenAI were from production engineering — for example, surveillance and artificial lift. But we also used GenAI for job summaries, preparing physics-based simulation such as wellbore hydraulics models and helping classify invoices for operations. All of these GenAI tasks drove efficiencies due to reduced workforce time.

Using GenAI as automation for preparing traditional physics-based models or physics-informed machine learning models is one interesting area to hybridize the two capabilities by providing a prompt-driven experience rather than performing button-clicking in a modeling app or writing code to build a model. Using GenAI as automation, like during conversational search to interrogate results generated from traditional machine learning models is another area to hybridize the two capabilities.

Scalability will continue to be a priority for all of us, because while end users say they want tools they can use, acceptance still lags. So there is a challenge for us in building solutions that solve real problems but that are also easy to learn and implement. Organizations can help with acceptance by developing well-thought-out governance structures that support innovation and help end users trust the tools that have been created.

I think we’ll see more companies realize that data governance must precede AI governance. AI teams need clear guidance on which data they can use — and how they can use it — before beginning any project. And they must have confidence that the data they are incorporating is accurate and timely. Proper governance also ensures that we aren’t duplicating workflows across the organization.

As these issues sort themselves out, we will likely see traditional roles in oil and gas evolve as technology evolves. GenAI will automate routine tasks in engineering and operations, enabling employees to take on new, optimized opportunities. We’ll see employees focus their efforts on areas where AI can’t yet help — issues such as optimization, debottlenecking and more. Employees who have critical thinking skills and can understand how to modify or combine models into problem-solving workflows will be in high demand across the industry. But we need to do a better job of explaining to employees what this future looks like — we need to explain in nontechnical terms how they will work with AI and how AI will support them in these elevated roles.

Without effective communication of that vision, there will be more skepticism than acceptance among front-line workers.