Brian Walzel, senior editor, Hart Energy: Hi, I am Brian Walzel, senior editor at Hart Energy, and I'm here at Hart Energy's DUG Gas Conference in Shreveport, Louisiana. And I'm joined today by Peter Harding, founder and CEO of Kelvin.ai. Peter took part in a great panel discussion around AI applications and natural gas development. Peter, thanks for joining us.

Peter Harding, founder and CEO, Kelvin.ai: Happy to be here.

BW: Kelvin.ai has integrated human expertise with AI to deliver autonomous well operations not in control. Can you talk a bit about those results of that application?

PH: Yeah, so what we find is first, the capturing of the information is a big challenge, so making sure that we understand how those best engineers are actually doing their job. We've created a lot of tools to simplify sort of the access piece as the first stage, and then there's an opportunity to take those applications and test them. And the first part we start with is, what's the baseline really? What are those assets actually doing today? And being in a position where we can understand from the baseline what the impact is of those changes over time. So in upstream production, we've seen in certain cases single digit uptick in terms of production. There are certain cases where it's higher than that, but we also recognize that you're dealing with a depleting resource, a well that's sort of declining in value. And so we're looking to find situations where we can sort of affect that curve and be able to shift it up.

That's on the production side. Whereas on the cost side, we're seeing real changes in the amount of people it takes to run these fields and situations where we can get a lot more leverage and be a force multiplier. So you're able to have folks managing 2X as many wells as maybe they were before with the same workforce. And so those are the types of things that have real profit impact for our operators.

BW: And taking a macro view, where are AI applications having the most impact in the oil and gas sector?

PH: I think right now we're seeing a lot of work that's happening on the subsurface where folks are using it to go back and review data that may be 30, 40 years old that they've done from a seismic perspective and then marrying that with other information. I think that there's kind of a synergy piece of this that is just starting to take hold, but folks have been working in earnest on it for the last five years or so. We see on the operation side a lot of really good sort of interfaces with allowing people to talk to their data. So we think about, there's sort of this internal analogy around the C-3PO you need as a protocol droid to be able to actually talk to the data. And OpenAI and some of these other LLMs [language learnings models] are really good in terms of being able to create that connection. And then you need R2-D2 to be able to actually talk to your machines. And in that context, that's what we do. We want to be that bridge between the human information and the actual machines themselves. And that's what’s starting to really take hold is the ability to kind of bring both of those together so that as an engineer I can get to an insight faster and then I can take action and measure that impact very easily. And that's what we're seeing right now, which has been really powerful.

BW: I love the analogies. We have this conversation just now on stage. It still seems to be early stages for AI adoption in the oil gas sector, but thinking five years from now, to what scale will AI be playing a role in the operational side of oil and gas production and what will be the key drivers to those changes?

PH: I think you're going to see AI across virtually every workflow as it stands. And what does that mean? Well, back to the [question], can I have a conversation with my data? I think at some level that's a large language model that allows me as an engineer or an operator to speak in my language and actually extract the information I want. That's very possible today, and I think those tools are even getting better.

I think the second piece is going to be expressing the action. You want to be able to say that in plain text and turn it into code because that's the next piece where, we've needed developers, we've needed product managers to create that interface. And we're not going to need that as much anymore. We're going to need more people just being able to express what they want and being able to turn that into code.So that'll be something that we're going to see a lot more of.

I also believe you're going to see just more and more autonomy in these systems. And just drawing a distinction between sort of automation, which is traditionally rules-based, to autonomy, where you have actually systems that are learning, that's also going to be happening. We're seeing it right now, but we believe certainly amongst our early adopter customers that they're getting huge value out of that. Because there’s something about being able to make a decision, be able to measure the impact and then learn from the results. That makes sense to every engineer that I talk to. And so now it's about how do you bring that into your workflow so it's easy to use and something that really drives a lot of value.

BW: Yeah. Okay. Great. Well that wraps up our time here with Peter Harding with Kelvin.ai, thank you all for watching.

Comments