This episode of the Pipeliners Podcast features a discussion with Christopher DeLeon and Rhett Dotson from D2 Integrity on the role of artificial intelligence in integrity management. The conversation explores whether AI is a useful tool or just a buzzword, emphasizing the importance of data quality, AI-assisted decision-making, and real-world applications in pipeline operations. The episode also touches on emerging AI-driven technologies, industry trends, and the challenges of integrating AI effectively.

Artificial Intelligence – Buzzword or Real Technology? Show Notes, Links, and Insider Terms

• Rhett Dotson is a Chief Engineer with D2 Integrity, LLC. Connect with Rhett on LinkedIn.
• Christopher De Leon is the Director and Principal Consultant with D2 Integrity, LLC. Connect with Christopher on LinkedIn.
  • D2 Integrity, LLC was founded in 2022 by Christopher De Leon and Rhett Dotson. The company has grown into a B2C and B2B ILI focused engineering, consulting, and advisory firm. They specialize in delivering expert ILI based engineering assessments, consulting pipeline operators and technology companies to achieve best in class ILI based pipeline integrity assessments, and support investors with ILI technology and market SME advisory services. Their principal objectives are to create value for our clients and provide an enjoyable professional working environment.
• DeepSeek is an AI research lab dedicated to advancing large language models (LLMs) and generative AI. It focuses on developing state-of-the-art natural language processing (NLP) technologies, improving AI reasoning capabilities, and optimizing model efficiency for various applications.
• Integrity Management is a comprehensive approach to ensuring the safe and reliable operation of pipelines, involving risk assessment, inspection, maintenance, and regulatory compliance.
• Upstream is the operation stage in the oil and gas industry that involves exploration and production. 
• Midstream is the processing, storing, transporting and marketing of oil, natural gas, and natural gas liquids.
• Downstream is the process involved in converting oil and gas into the finished product, including refining crude oil into gasoline, natural gas liquids, diesel, and a variety of other energy sources. The closer an oil and gas company is to the process of providing consumers with petroleum products, the further downstream the company is said to be.
• ILI (In-line Inspection) is a method to assess the integrity and condition of a pipeline by determining the existence of corrosion, cracks, deformations, or other structural issues that could cause a leak.
• AI is Artificial Intelligence, the simulation of human intelligence processes by machines, especially computer systems.
• ChatGPT is an AI chatbot that uses natural language processing to create humanlike conversational dialogue. The language model can respond to questions and compose various written content, including articles, social media posts, essays, code and emails.
• Generative AI is a type of artificial intelligence technology that can produce various types of content, including text, imagery, audio and synthetic data.
• Machine Learning (ML) – A subset of AI that uses algorithms to identify patterns in data and make predictions or decisions without explicit programming.
• Neural Network – A type of machine learning model designed to simulate the way the human brain processes information, often used in pattern recognition and predictive analytics.
• PHMSA (Pipeline and Hazardous Materials Safety Administration) is responsible for providing pipeline safety oversight through regulatory rulemaking, NTSB recommendations, and other important functions to protect people and the environment through the safe transportation of energy and other hazardous materials.
• NTSB (National Transportation Safety Board) is a U.S. government agency responsible for the safety of the nation’s major transportation systems: Aviation, Highway, Marine, Railroad, and Pipeline. The entity investigates incidents and accidents involving transportation and also makes recommendations for safety improvements.

Artificial Intelligence – Buzzword or Real Technology? Full Episode Transcript

Russel Treat:

Welcome to the “Pipeliners Podcast,” episode 375, sponsored by EnerSys Corporation, providers of POEMS, the Pipeline Operations Excellence Management System, Operations and Compliance software for the pipeline operator to address safety program management, control room management, and field operations. Find out more about POEMS at enersyscorp.com.

[background music]

Voiceover: The Pipeliners Podcast, where professionals, bubba geeks, and industry insiders share their knowledge and experience about technology, projects, and pipeline operations. Now your host, Russel Treat.

Russel:

Thanks for listening to the Pipeliners Podcast. I appreciate that you’re taking the time, and to show the appreciation, we give away a customized YETI tumbler to one listener every episode.

This week, our winner is Dave Mann with Phillips 66. Congratulations, Dave, your YETI’s on its way. To learn how you can win this signature prize, stick around till the end of the episode.

This week, we speak with Christopher DeLeon and Rhett Dotson from D2 Integrity about artificial intelligence. Is it a buzzword or is it actually useful technology?

Well, this is Russel treat from the Pipeliners Podcast, and I want to thank Sarah and Chris and Rhett for allowing me to use their booth. This time, we’ve called a truce, there’s no booth stealing going on. We’re playing nice nowadays.

Anyways, look, I wanted to talk to you guys about artificial intelligence. I hear a lot of people talking about artificial intelligence. One of the things I want to know is, who if anybody’s doing it well in integrity management.

Rhett Dotson: It’s really funny that…did you bring this up because of the DeepSeek stuff that just happened like a day ago?

Russel: I have no idea what just happened.

Rhett:

Fun fact on the news.

[crosstalk]

Russel: You’re talking about the Chinese model. Yes.

Rhett:

…that came out in all the stocks, including…it didn’t even affect the natural gas stocks. They dropped 10 percent because they were expecting all the natural gas to feed the plants. Anyway, that’s not why.

We’ve had two different guests talk about AI on our podcast. We had Adrian Belanger from TDW, and we had Thomas Boyker from ROSEN. I do know that they’re using it out there. I am loathe to comment on who is doing it well or better.

What I am more than willing to comment on is, this is like a trigger word for me because I feel like these words are thrown around. It’s gotten to the point now where when I’m listening to a presentation and somebody says the word neural network, I just want to walk out because…

Russel: Most people don’t even know what that is.

Rhett:

Or they’re even using it correctly, or they’re like machine learning. I’m like, “You did Y equals MX plus B.” That’s linear algebra. What are we doing here? That’s why…

[crosstalk]

Rhett:

…to me. I think what I’m more concerned with, Russ, when we talk about this is, the models are only as good as the data you feed them.

I mentioned, we’re here at PPIM. Sherry Baucom did a fantastic presentation. Looking at the performance of ILI data over the last, I think it was 15 years on all the data…

[crosstalk]

Chris DeLeon: I think it was all like 2003 or so roughly.

Rhett: She very clearly showed amongst all the vendors, they’re not all created equal. Some of them had much better performance than others. When you talk about…

Russel: It’s interesting. Why is that, do you think?

Rhett: Oh, man. So Chris and I, having worked for a vendor, I think we understand it. She got to it a little bit, but she was even gray. Is the ILI tool is not just a tool. You say this all the time, it’s a system. It’s the run conditions, it’s the analyst. It’s the algorithm identifying it. It’s more.

Russel: It’s the instrumentation. It’s health, it’s implementation.

Rhett:

It’s not a ruler you walk up to.

[crosstalk]

Chris: We’ve talked about this before like fatigue, the human factor of it.

Rhett:

Exactly. My thought on that, Russell, is as anyone chooses to use it in the industry, and I have my own thoughts, but I want to turn it over to Chris on this too is that, the quality of the data becomes important.

If you don’t have it, it doesn’t matter if you have a million data points. If your first 500,000 data points were you learning, and you know they’re potentially fraught with errors, and you feed those into the model.

Russel: I’ve talked about this a lot in machine learning around real-time systems. It all starts with a quality data set. You’ve got to have well-organized, well-structured, accurate data.

Rhett: Accurate.

Russel: If you don’t have those things, garbage in, garbage out. I absolutely agree, and we’re learning about what that looks like, for sure.

Rhett: Give your opinions, Chris.

Chris: AI is fun. I would say in our space, we’re being bombarded by AI. It’s everywhere.

Rhett: It’s going to solve all our problems.

Russel: If you don’t have it on your marketing materials, you’re probably falling behind.

Chris: We know. Did you hear that Sarah?

Rhett: This is an hour episode. I just forgot. Sponsored by AI. Sorry, it’s coming for your job, Ms. Spenser.

Chris:

One of the things that I’d like to align it with potentially some of our listeners is, it’s almost similar to the roadmap that we see with ILI and the value prop and how that’s presented to us. There’s a lot of ILI vendors. There’s a lot of ILI tools. There’s a lot of collateral out there that you can grab and try to interpret.

In fact, it’s probably easy to call somebody and say at an ILI vendor and say, “Talk to me about this.” There’s a wealth of information, but even still, we could argue that the average user doesn’t fully understand what the ILI system is capable of doing and what it’s actually doing. Yet we make a lot of decisions based on it.

Russel:

No, this is actually interesting. I’ll tell you a little bit about what I’m up to in this domain. We were recently awarded a PHMSA R&D project to take a gaming platform and generative AI and combine them and build a training scenario manager.

A multiplayer simulator for team training and emergency response, so that others outside the industry, like first responders or multiple stakeholders, multiple roles in a pipeline could run a scenario.

One of the things that we’re finding is, it’s all about, one, what data do you get to train the model? Two, how do you massage that data to train the model? Because the model is only going to respond in the way it’s trained to respond from the information it has to respond from. There’s a lot of stuff in the market that’s easily available, incidents, NTSB reports, things like that.

There’s a lot of stuff that’s not easily available, that’s probably way more valuable, like near misses that aren’t generally shared. Then how do you get that? Here’s my premise. The promise of AI is immense, but our ability to utilize it effectively is very much in its infancy. It’s all about what we build as interfaces to the models that provide something meaningful for the user.

Rhett: Now I have questions. This gaming platform, will it be like a Steam Deck? Because I have a Steam Deck. Will I be able to download it and play it like a first-person shooter on my gaming system?

Chris: On the plane? Disconnect it from the Internet. [laughs]

Russel: I think it’ll be a web delivery. I don’t know that it’ll be a console delivery.

Rhett: Do I get to walk around?

Russel: Not in the first generation. It’s going to be more of a narrative simulation.

Rhett: More like Oregon Trail.

Russel: More like Oregon Trail.

Rhett:

Everyone died of dysentery.

[laughter]

Russel:

Actually, it’s a lot like Oregon Trail. Exactly.

[laughter]

Rhett:

You failed to cross the river.

[laughter]

Russel: Oh, my God. That’s awesome. Here’s the interesting thing about Oregon Trail, though. Oregon Trail was one of the first strategy games that people really loved because it was simple enough to understand.

Chris: Hunt, cross the river, survive.

Russel: It’s very interesting because it’s not about the elegance of the UI that makes it meaningful, it’s about the quality of the story.

Chris: The experience.

Russel:

Then the UI can always be improved and made more engaging. Anyway, so I think what I’m interested in, or what I’d like to get you guys’ perspective on, from the standpoint of using AI in integrity management, one of the places that it offers a lot of promises, these tools are generating loads more data.

The work to get to the things I actually need to put a human brain looking at, can be radically reduced by an appropriate application of AI. I mean to tee that up as a question.

Rhett: Nope. I know what you want.

Russel: What’s your take on that?

Rhett:

I very much resonate with what you’re saying. When we talk about AI as replacement, I get really worked up. Again, it’s like second trigger, because I think we’re nowhere near AI replacing.

When we talk about AI assisted, then we just…That was what Thomas Boyker talked about and I think a lot what Adrian talked about as well, is not replacing an ILI analyst, but assisting the ILI analyst.

It’s like, I’ve got a truck now that’s got some of the self-driving features, but it doesn’t completely drive. In certain conditions, it will take control and change lanes. It’s more like a driver-assisted feature that actually greatly improves my driving experience, but there’s no way on earth I’m turning the car over and letting it take me home. It’s not there, and we wouldn’t do that.

Chris: A bunch of topics come to mind, so I’ll just shoot some of them out. One of them is, the purpose has to be clearly defined. The person has to know exactly what I’m using it.

Russel: The guy that understands the outcome I’m looking for.

Chris: I’ll give you an example. We’re looking into using Microsoft Copilot to enable us. It’s like, yeah, but it plugs into everything, so where do I start? That’s the first part is…

Russel: I see, and I wouldn’t use Copilot because it’s Internet-facing, and I would go with a private model that I could train.

Chris: Sure, but the point is…

Russel: I get it,

Chris: …again, it’s this whole idea of we’re bombarded by it. It’s everywhere. Another example of it it’s…

Russel: Copilot’s nothing but the paperclip guy repackaged.

Rhett: Oh my gosh. Clippy. [laughs]

Russel: I’m serious.

Rhett: Chris has lost his strain of thought. You need me to take over so you can recover.

Chris:

No, I didn’t. The other one is, I know what problem I’m trying to solve and I don’t have the data, but I trust someone did. How do you deal with that? I bought a solution from somebody that they gained access from the PHMSA database and they did great stuff on it.

Almost if you think about one of our podcasts that we did with Keith Lavis, where they went to the PHMSA database, extracted a bunch of data, did a bunch of neural networking and all this great stuff where they integrated AI and AR is separate, but it’s a trusted data set and I can trust this now.

When I tell it what I want, it does a lot of fun stuff and it spits out an answer. Can I run with it? It feels a little bit like, wait, maybe I don’t want to do that. Go ahead. What are you doing?

Rhett:

I got another application. Machine learning, AI, buzzwords together. We have a real-world example we’re dealing with right now, the subject of the paper I presented on, where in theory, we have these millions of data points which were fed to a machine learning algorithm to produce lower bound estimates of dent fatigue life, but the curves don’t make sense.

They do not follow any fundamental underlying logic. I think as we use this, we have to make sure…

Russel:

That’s either a fault in the model, or a fault in the algorithm, or a fault in the training. When I say model, I mean data structured into the model. I get it, man. Here’s…

[crosstalk]

Chris: I didn’t get to finish my point. No worries.

Russel: Told you it was a trigger.

Chris:

No worries. [laughs] The other one was, and then we also hear it highlight of suppliers who are using this in different ways. They’re saying, “You know what we do? We can now apply these algorithms and you give it a little bit of input and it spits out an answer.”

I think you said this earlier. In all of those scenarios where it’s Copilot, whether it’s I needed a tool that I’m trusting to trust a data source, or it’s I’m doing some kind of supervised machine learning. At the end of the day, you said this earlier. Are you changing the experience of the user to where we can have a better outcome? Is it reducing fatigue?

Russel: Is it allowing me to apply my analytical brain?

Chris: To solve the right problems and use energy the right way.

Russel: Is it supporting my intuition?

Chris: Yeah, because here’s what we did. We got to see some of these ILI algorithms being generated, and it was, you’d have hundreds of thousands of features where traditionally in ILI analysis, you think, I’m going to start at the launcher valve, and I’m going to go through every joint, click. Who wants to do that for 30 years?

Russel: Not me. [laughs]

Chris:

You bring in AI. You train it, assume that works well. What you really were able to do, is we were able to reduce the time, but make the experience a little bit more enjoyable and actually almost a little bit like a game.

You could present things in clouds, and you could sample the cloud, and then it produces results and it classifies things. You move on to the next thing. It becomes more of like a interactive problem-solving activity, where you’re still achieving the same goal, but the user’s experience was elevated and they were able to do something a little bit more fulfilling.

Russel:

In the control room world, this is all human factors. Where we were prior to the control re-management rule versus where we are now implementing API 1165, SCADA displays for pipelines, what we’re finding is, you really need to focus on what decision is the pipeline operator trying to make?

How do you support them to make that decision as accurately and as quickly as possible. It’s exactly the same thing, and that’s all about, how do I present and how do I have the human interact with the intelligence? AI is a buzzword.

Rhett: Machine learning.

Russel: It is a buzzword that’s being used to cover a whole bunch of things. Basically, AI is, “Hey, we have these new ways to deploy algorithms.”

Chris: Again, I want to reiterate this, because I feel like at least in our domain right now, I feel like this is where it’s most applicable. A lot of us are trying to solve technical problems. Maybe I can make this parallel properly. I’m in control room and I’m monitoring.

Rhett:

You’re getting tons of alarms.

[crosstalk]

Chris: Am focused couple of times and I have to have scheduled breaks so I can re-energize so I don’t get mental and physical fatigue. Or can we enable AI that allows me to still accomplish tasks, and it’s supplementing the energy and focus I needed so that, like you said earlier, the right things happen.

Russel: I get rid of the repetitive cognitive load, and I allow that cognitive load to be applied to the analytical task.

Chris: One of the things we think about is it’s all resources are finite. In this case, guess what’s finite? Cognitive load, energy focus. You’re allowing me to better distribute my energy and focus.

Russel: What distinguishes a really good integrity engineer from a competent integrity engineer. Why is really good different from competent? My answer is that…

Chris: The way I rendered that was performance versus competence.

Russel: What I’m driving at is intuition. I thought that’s where you might be going. We as human beings, we naturally recognize patterns, and as we begin to learn something, we see the patterns inside of it. Whatever that is.

Chris: Which can be good or bad. We can grow biases or we become efficient for performance.

Russel:

That’s what we’re good at. If the AI can present patterns to me that I can use, then it all becomes about, how do I visualize this interaction? The other thing I would say is, we got to quit thinking of AI as a be all end all, this is the tool.

It’s really like, I have a tool chest that I’m going to take this AI and I’m going to…Good analogy would be, I have a new set of power tools that are all battery operated, and I have one battery that powers all the tools.

Every tool has a different function and works a different way and looks a different way. The AI is more like the battery, but we don’t really have a lot of tools yet. Would you all agree with that?

Chris: I would, and it makes me think about this.

Rhett: I can’t speak anymore.

Chris:

We need to be careful, though, right? Because here’s what happens is it’s too often. I like using the ILI business as a parallel. Operators got really comfortable with what the vendors were doing, and they just started under-managing or neglecting certain things. If we’re not careful with AI, we will just start depending on auto-generated fill in the blank.

Now we’re not even competent. We think we’re doing a good job, but we’re actually being ignorant if we’re not careful about it.

Russel: The risk is we could get lazy. We could over-rely on the tools.

Rhett: It’s so rare that I don’t get a chance to speak. It’s so difficult, Sarah.

Chris: You’re in an uncomfortable space, aren’t you?

Rhett: I am. Like three extroverts, all of a sudden. [laughs]

Russel: You’re making funky signs to people walking by the booth. We’re allowing you to get away with that.

Rhett: If you’re watching the video.

Russel:

I guess that’s the rule. [laughs] Don’t let Rhett be quiet. As long as he’s talking you won’t…

[crosstalk]

Rhett: One of the things that I was thinking about as we were talking about AI is, it’s less like Jarvis and Iron Man, who’s like a one-to-one complement, and more like an assistant who thinks about things that you might not be thinking of. When you talked about integrity management, it’s easy for me to talk about the context of a control room.

Russel: Let’s both talk about what we don’t know, and it’ll make it really interesting. I’ll talk about integrity management, you guys talk about the control room.

Rhett:

In the control room, I’m going to use the incident that was related to the release in Kalamazoo. There was conflicting information coming in.

If you had an AI that could sift through the various alarms and maybe offer assistance to the decision maker of, “Hey, in this situation, under these consequences, there’s a 90 percent likelihood that this is associated with a release rather than that.”

I wonder if that helps the decision maker more objectively rely on his experience to make a correct decision, or when you have other alarms that are going off.

It can take the aggregate information from the system as a whole, rather than being able to absorb one pressure reading or one valve reading and looking at the system as a whole, here’s what the AI thinks might be going on, which gives another data point.

Russel:

I want to expand on that a little bit because it’s an interesting notion. The challenge with leak alarming is that you get false alarms, it’s just the nature of it. You typically would get something like 10 to 1 false alarms to…Real leak alarms are exceedingly rare.

The process is, I get a leak alarm that’s telling me there’s a hydraulic condition that could indicate a leak, and I have to go through a rubric to rule out all the things it could be like slack line or a startup, or a shutdown, or a valve movement, but it’s not leak.

The AI could be used to say, “Look, we looked at all these things. There’s not any of those.” It can rule things out. I think it’s an equivalent thing again, I ask the question. ILIs I’ve got these thousands upon thousands of features. Here’s the 9,000 you don’t need to look at and the 100 you need to pay attention to.

Rhett: It does that on the aggregate of not just the ILI data, but maybe your DCVG readings, maybe similar pipelines, it begins to expand and offer assistance.

Chris: Those are the different domains that we’re talking about.

Rhett: One minute.

Chris: It’s OK. I think one of them is, it’s facilitating a task, but I want to touch another one. In Kalamazoo, we had a conflict of authority there. We had information that was presented that was being subject to someone’s experience.

Russel: There’s a whole thing there.

Chris: What I’m trying to frame it in, though, is AI could also begin to have a place if we were thinking about controller management as a case and also ILIs, it’s not just the facilitation of task, but it gives grounds to people to really challenge that authority, or enable the authority to be able to check themselves from their biases.

Rhett: I don’t think you’re going to be able to stop them.

Russel: I would argue that.

Rhett: When do we pick this up?

Russel: We’re getting a signal that we are having to complete this conversation. Listen, we’ll just cut it there because we could run down this rabbit hole for a long time. It’s really interesting. This conversation is very important for our industry to understand. 10 years from now, 20 years from now, all of us are going to be using AI all the time.

Chris: All in AI, all in.

Russel: What we’re exploring right now is, what might that look like. I would offer, we don’t have a clue. It’s like when the car first came out, nobody knew we were going to have highways.

Rhett: We don’t have a clue. Thank you, Russell.

Chris: Also great time.

Russel: Gentlemen, thank you for letting me use your booth here. Thanks for having a conversation. It’s great to see you. I’m going to go walk the floor.

Rhett: See you. Take care.

Russel: That sounds like a country song, right?

Rhett: [laughs]

Russel:

I hope you enjoyed this week’s episode of The Pipeliners Podcast and our conversation with Chris and Rhett. Just a reminder, before you go, you should register to win our customized Pipeliners Podcast YETI tumbler. Simply visit pipelinepodcastnetwork.com/win and enter yourself in the drawing.

[background music]

Russel:

If you have ideas, questions, or topics you’d be interested in, please let me know either on the contact us page at pipelinepodcastnetwork.com or reach out to me on LinkedIn. Thanks for listening. I’ll talk to you next week.

[music]