Generative artificial intelligence (AI) is bringing massive potential into the oil and gas industry, but it’s not without its risks, experts said at the IMAGE conference in Houston on Aug. 29.

The technology has quickly garnered tons of attention because of the impact it has on task performance, the speed of its development and its ability to democratize other forms of AI, said those on the Generative AI and the Future of Energy panel.

At the same time, there are concerns around security of the data used in generative AI systems, as well as tendency for it to deliver inaccurate information. Training is needed—from the top of organizations all the way down to the end users.

Jose Borges, principal director of energy at Accenture Strategy, said a recent study by the firm found that about 40% of energy tasks will be affected by generative AI.

“That does not mean the tasks will be eliminated, but the tasks can be augmented by gen [generative] AI,” he said. “Some of the reporting side of things, some of the chat bots, these things, we expect that to be completely automated with gen AI using natural language.”

On the other hand, generative AI could summarize image processing logs for use in well design. Pulling that data using natural language and summarizing it should improve the quality of the next well designed in that workflow, he said.

“It's important for people to understand that gen AI is not a technology gadget,” Borges said. “Gen AI should not be seen as one app you install in your phone, and that's a fundamental change [for] the way you perform a workflow. So it's very important for people to be aware of what gen AI is, getting educated about what gen AI is and most importantly trying to think how that fits into the overall workflow, how that can transform the overall workflow.”

Jimmy Fortuna, chief product officer at Enverus, said the company’s software engineers noticed a dramatic increase in productivity by using generative AI plugins to their software development tools.

“They're sitting there writing code, and before they can finish a block, the software development tool has accurately predicted what the block needs to be. If you're a software engineer, this is actually surprising and delightful,” he said.

Fortuna touched on the idea that generative AI may ultimately enable “basic science discoveries in the geosciences as a result of processing this data” against many different variables that previously had not been considered.

“I'm going to give it a non-zero probability that might actually happen,” he said.

Training needed

Xiaojun Huang, chief of modeling, optimization and data science at Exxon Mobil, urged all geoscientists to learn how to use AI as a tool. Exxon Mobil has curated online courses that its employees can use to learn more about AI.

“Our data scientists actually are acting as teaching assistants” to help the broader organization learn about using AI, she said.

“Physics is always going to be very valuable,” she said, suggesting experienced geoscientists to change how they think about AI and the open source software engineering space—as well as the problems they are trying to solve so they can break down the workflow and see how to best leverage AI to make a difference.

Understanding different AI platforms are necessary for different parts of the organization, Huang said.

The leadership of an organization needs to focus on how to get a return on their investment for using AI technology, she said, while the rest of the organization needs to know how to use and work with AI effectively.

“What I think is very important is, you actually need to have the bottom-up as well as the top-down design. Without the bottom-up, you can't get the organization excited about AI. But without the top-down design, we have a lot of micro chaos,” she said.

While there are benefits of generative AI, dangers exist, particularly around data security and privacy.

“There's some strong cautionary tales out there of very eager, understandably eager employees that have taken company secrets and making (them) someone else's property as a result of casual use,” Fortuna said.

At the same time, generative AI technology is developing quickly whereas regulations will take longer to develop.

“Whether or not the regulatory framework catches up to” the potential of generative AI technology is another question, Fortuna said.

Borges said one of the big questions is around who owns the intellectual property (IP) when generated content is involved.

“If some knowledge or some workflow is derived from a large language model that generated that and that becomes IP, who owns the IP? The company using the LLM (large language model) or the company who developed the LLM. So the regulatory bodies are scratching their heads a little bit from the IP side,” he said.

Disruption potential

“I got to see the before, during and after of (the dot com) era as the internet kind of washed over everything, and it was really hard to keep your bearing during that time because, for those of you that were there for it, you see these companies that nobody's ever heard of that are all of a sudden worth a zillion dollars and then they're gone just as fast,” Fortuna said. “It was very disorienting.”

And the world is very different now due to the internet, he said.

“Living through that era and comparing it to the present day, I've never seen anything move as fast as this stuff does. Nothing even close,” he said. “We were having discussions internally where what was impossible two weeks ago is now being done literally in the span of less than a month.”

And during brainstorming sessions of the past, usually half to two-thirds of ideas are shot down by engineers who explain “that it won't work or won't work well or nobody can afford it,” he said. Now, “every single time we ask, ‘What if you could?’ the answer was it's either already done or it looks like it's about to happen. Never had that experience. And by the way, we're really early in this … Check (back) in two or three years.”