Beyond the Buzz of AI for Energy Assets—What’s Available Now, The Future of GenAI

Introduction: Beyond the buzz of AI for energy assets, what's available now, and the future of Gen AI. AI is not just a hot topic or a concept. Understand that current and emerging use cases for AI in energy and other asset-intensive industries is critical to future success. With the amount of data produced by your assets, processes, people, and other systems across your enterprise, your organization needs tools that can help you draw insights faster than you ever have before. Today, we're seeing our customers and their peers not just look into predictive diagnostics, but also recommendations, generative AI, agentic AI, and prescriptive recommendations to help meet the challenge of evolving data needs. At the same time, in these industries, there's a need to keep people on the ground involved, what we refer to as human-in-the-loop, to meet efficiency, value, output, and safety goals with confidence within those models. In this audio blog, you'll hear my colleague, Mazen Younes, and I discuss that evolution in AI, and how our platform supports AI integration in the real world.

Hello, everybody, my name is Ryan Finger. I am the Director of Product Marketing here at GE Vernova's Electrification Software business, primarily working on our Essentials product, which is inclusive of our platform, bring your own analytics, advanced visualisation, AI, and our microservices. And I'm here with Mazen Younes.

Mazen Younes: Yeah, thank you, Ryan. Glad to be here with you. So, Mazen Younes here. I lead product management for our APM Essentials and platform products, as well as our AI strategy for the Power and Energy Resources software and the Electrification software at GE Vernova.

Ryan Finger: We're just going to get right into it and really start off with how the energy space is using AI and machine learning today. And I always like to remind people, machine learning is a subset of AI, it's one big environment, right? So, when you hear us use AI or machine learning or gen AI, just remember, it's all just getting the most out of your data at the end of the day. And the big area I wanted to start off here is just a very simple view on how people are using AI and machine learning today. And for both the plant level, as well as the enterprise and energy, what we really find is the space, when we talk to customers and partners and then internally, the space is still at a place where they're working to grasp just how to use this technology the best way possible. Things like similarity-based modelling are very prevalent. You have your historian systems that you're able to do some simple analytics on with AI and machine learning. But customers and organisations are still really dealing with how to build the right infrastructure to get more out of it.

So, today, energy companies are doing a great job using machine learning, building trends, doing analytics, descriptive and diagnostics. But what we're seeing is there's a pretty heavy shift on the industry moving more towards predictive, prognostic, and then now obviously generative and agentic and prescriptive recommendations. So, ultimately, the flow that we're seeing in the space is very heavy machine learning with the trend kicking over to AI. And what we're really looking at is, as enterprises look at AI, it's now at the point where, how do we use AI in the front office and back office? And what I mean by that is how do we use AI and machine learning on our assets to get more insights? And how do we use AI and machine learning centrally in the organisation to get more from our data? So, we're seeing a shift from a very predictive way of working on the asset side into, how do we grapple with this emerging technology, which Mazen is going to talk a little bit more about.

Mazen Younes: Yeah, Ryan, absolutely. And from my perspective, there are various uses of AI/ML solutions within the energy industry, mostly centred around driving uptime of critical equipment up, enabling efficiency, all the way to reducing costs, promoting our safety, from an uptime perspective, from an equipment failure perspective, and for personnel, and then from supporting sustainability initiatives nowadays. So, you touched upon a few things, right, predictive maintenance, for example, the use of machine learning to analyse sensor data, predict equipment failure based on specific failure modes is one of the key aspects used today as a solution. It reduces downtime, optimises maintenance costs, and avoids any safety risks. And even within GE Vernova, we have examples of that with our product called SmartSignal.

With carbon management systems as well that you're seeing out there, AI is used for data accuracy. When you're pulling in Scopes 1, 2, 3 type data, today we're using linear regression models, ensuring the emissions data collected is accurate, that there's no issues with the data itself. Let's say if a sensor is down, you can basically predict what the data would be based on other sensor tags, based on correlation, regression models that can be used to come up with the correct value and make sure that there's no data anomaly. So, I think that's a nice part of where it's being used within the carbon management system. And besides doing strategy and planning, we're basically using optimisation model techniques that provide recommendations based on constraints and assets available, for where customers want to invest more into decarbonisation.

You also have optimisation systems that are out there, that are in use today. Think of closed-loop type systems that are using machine learning today at the controls level of turbines, where they can auto-tune these turbines dynamically based on certain parameters, whether adjusting flame temperature or fuel splits, based on defined constraints and environmental conditions. So, in the carbon management space, we have CERius as a product with GE Vernona that we use. And then, for optimisation, we have Autonomous Tuning that is used for aeroderivative turbines. And then, you put into this mix robotics. You have robotic inspection that's nowadays also being adopted in the industry. You have these autonomous robots that are equipped with all these sensors and the computer vision, that are being deployed in industries such as oil and gas and power and utilities, that are conducting these frequent inspections of these assets.

So, essentially, these robots are promoting worker safety, they can access hazardous areas or hard-to-reach areas, and they can increase the capacity basically of where people may not always be available basically to conduct these checks. So, computer vision models are becoming more of a norm and a trend in the industry. This is being used already today to assess the health of equipment, it can alert based on triggered anomalies detected. And within GE Vernova, we have autonomous inspection that's a product that we're using that can integrate into autonomous robots. Think of ANYbotics, the animal robot, for example, that we have a partnership with.

So, yeah, Ryan, I think in terms of where we stand today, there's definitely good traction and usage of AI/ML within the industry. But as you noted, there's agentic AI, there are new trends that are coming out there that are definitely going to push the industry into further innovation and further acceleration of productivity within the space.

Ryan Finger: I think on the interesting point, one of the things of note that we're seeing from the GE Vernova side, and this is across all of our business units that are working in software is, there's still the need for the human-in-the-loop, right? There's a lot of interesting trends. I actually saw a report, everyone's sharing it across LinkedIn, and everyone knows who's listening, I'm a LinkedIn fiend, a recent MIT study that came out that said about 95% of these new AI projects, people are having a really hard time quantifying the return. So, there's this interesting top-down pressure in energy, where folks are coming into the energy space from outside energy, maybe they have an IT background, maybe they're coming from consulting, and they're trying to drive a really large transformation effort. Then you have the folks that are keeping the plants up and running, onboarding new assets, producing the energy, and they have a completely different use case, and they want to go bottoms-up. So, what we're seeing now is this convergence of what's possible and what's reasonable.

So, when we look at things like autonomous robotics, the way we look at it on our end is, how do we keep the human and the expert in the loop? Because right now, everybody coming out from, whether it's undergrad or advanced degrees or just who are inquisitive, we're starting to see the emergence of obviously, I'll call it citizen development or citizen coding, right? The term out there now is 'vibe coding'. You've got people like me who are typing away trying to pull something together. So, there's an interesting blend. And the way we look at it from a Vernova perspective is we want to give the flexibility for these AI/ML use cases, but there's also a really important need to have parameters set, because a lot of these initiatives, with the emergence of gen AI or agentic, it can ramp up costs and the return might not be that easy to determine.

So, we're really focused on keeping the AI use cases tight to the assets, tight to the people, and allowing those subject-matter experts to get more out of our system. So, it's a really interesting dynamic. And coming from the banking side, front office, back office, automation, full workflows being automated, we're seeing there's a hesitancy in energy. So, it's really about finding that right blend of emerging technology and really considering a tight use case, especially in these critical areas like oil and gas facilities, power generation facilities, that are producing our energy.

Obviously with that, there comes challenges. I think this is one of my favourite topics is the challenges with this technology. There's a shift that's happening now in energy that I saw that was occurring in banking probably four or five years ago. There's cloud technology, there's SaaS, there's these new services, there's AI, there's agentic, there's these big promises. And one of the challenges with really truly adopting AI machine learning, again to tie it back to how do you realise value from this technology, a lot of asset-intensive organisations have a lot of back-end work to do before really being able to get the most out of it. You have legacy systems, you have control systems, you have hardware, you have people, you have different sites, you're making acquisitions, you're bringing on new asset types, and all of that comes with a hardware and software component. And in that hardware and software component, how do you get that data accessible to be able to use this technology? So, we're seeing the emergence of data lakes, we're seeing the emergence of machine learning operations, AI operations, data fabrics, all of them come with a really, really big promise.

But a lot of what we see on the asset-intensive side is it does take feet-on-the-ground work on infrastructure and how you're set up as an organisation, to truly get that data where you need it. So, there's challenges that are overcoming them of just having the right systems, because you're thinking about someone operating a wind turbine 1000 miles away from your headquarters. How do you get that data? How do you manage that data? So, overall, really what we're seeing and what I always refer to as shadow IT, and everyone says, "Oh, that sounds mythical and scary", and as a software person, sometimes it is. Because when AI is coming in, and people can now go to ChatGPT and say, "Hey, write me a Python code", that doesn't mean that code is going to be accurate. People want to build their own applications. If you don't have the right infrastructure, it's going to lead to a lot of broken processes and workflows, potentially data issues.

So, even though AI and machine learning has a lot of promise, my biggest takeaway when I talk about this topic is, you really need to think bottoms-up on how you're actually architected in your systems before unleashing, I guess, this next wave of technology. So, it's all really exciting. But on the IT and OT side, it's really about, how do you get your systems in the best shape to use that data? And if you miss steps or take shortcuts, you might end up with not being able to give an ROI or potentially having some cybersecurity concerns, or your costs might go to the roof in certain things. So, that's what I'm seeing, Mazen. I know you probably have some deeper insights there as well.

Mazen Younes: Yeah, absolutely, Ryan, and you're spot on. So, I mean, it all starts with the challenge of eliminating silos across the disparate systems, right? So, historians, DCS, SCADA, EAM systems, CMS systems, how you bridge and merge the OT and IT world, converging that together and presenting it as a unified data model to ensure you have consistent tagging for quality and usability of that data, so that then, whenever you're bringing in these new AI use cases that you're able to develop, you mentioned that data fabric layer, right? So, just being able to tap into that and tap into this wealth of data and figure out how to build things, that, to me, is like step number one in the foundation of anything AI-related, is the data itself and how you're presenting it. I think solving this will go a long way in terms of setting that foundation.

Now, I think there's also the regulatory and ethical standpoint that are complexities that companies should overcome to ensure that AI decisions, for example, are transparent. You're following certain standards and embedding domain expertise to mitigate poor outputs or poor results or quality issues coming from these AI models. That's also a very important challenge and consideration that companies should keep in mind as they develop these within the energy industry. And like, think of things like the EU AI Act that's coming out, right? Companies must adhere to that, be on the lookout for any new regulatory systems that are designed to guide these AI innovations and encourage the responsible design and adoption of these technologies. It's very important to keep that in mind. And you touched upon cyber and shadow IT. I think from the cyber standpoint, the AI systems naturally will bring in an increased attack surface within the enterprise, especially if these systems are exposing or are exposed themselves to these critical infrastructures. So, I think there's definitely a need from a security measure standpoint to have strong controls in place and strong governance that's in place.

I think in the industry though, like in the energy industry, it tends to be a slower-paced industry in terms of adopting new technologies. You also have probably the aspect of skills shortages, of people who can intersect maybe between having domain knowledge within the energy industry and having knowledge in AI technologies. I think that's also one of the challenges and considerations that companies must keep in mind as they push towards advancements within the space. But I think the slow-paced adoption may be a good thing, I would say, in this energy space, because sometimes moving slower and adopting these technologies the right way is probably the right path forward, especially in this fast-paced environment that we're in today.

Ryan Finger: And that's a good segue over to the cloud and SaaS side of things. And I think it ties naturally into what we just talked about, which is getting your infrastructure set up in a way that this data is useful, that is actionable, and you can access it. And when I think about the use of AI/ML, and when I was in a conversation with a few folks, with our customers and prospects, and they were battling through, "Hey, I want to do AI and machine learning, but I also am heavily on-prem, my data is on-prem, I don't want to leave on-prem, but I want to do AI and machine learning", the way of the world there is shifting. When you think about the speed, the access, your data usability, how fast you can pull that data in, how quickly you can model, we're shifting very heavily and we're seeing that in our customer base and across the globe, is this heavy shift into whether it's SaaS, which is GE providing software through a tenant for our customers to leverage, or even private cloud deployments of our software to help manage this data. And there's a big, big shift happening there.

What comes with cloud, there's a few things to consider. And what usually scares people is there's this high initial or perceived initial cost of going to the cloud or going to SaaS, right? You have to modernise your systems, get your data in order, figure out how you're going to migrate your data over, you might have some feature and functionality changes in your applications that you're going to have to retrain on. Plus, on top of that, you're looking at this initial bill to make this move. But when you're talking about AI and machine learning, my lens on that is, if you're trying to run AI models on-prem, all those things that Mazen just talked about with your systems and the data and security and just the access, you're going to struggle to show return. Because what we're seeing now is an ask and a requirement from a lot of customers is, how can your data work with the other data we're collecting today, whether that's in a data lake or data ops platform? We want to merge all that together and run models on it to compare things that maybe weren't compared in the past. So, to do that, the move to cloud and SaaS is highly important.

I know Mazen will get into it in terms of the technology that fits there, but when we made some investments probably three years ago, when we released APM V5, a huge focus was the monolith-to-microservice story, right? How do we break up some of these rigid cloud services and make them more flexible? And when you see things now like AWS Bedrock or Mistral or all these models that are popping up seamlessly overnight, you need to have the right services to use that technology. So, we made a pretty strategic investment a few years back to re-architect, move to a microservice infrastructure, and what that's really helping with is speed and performance of applications. And with that, as you start to add AI or machine learning or generative AI or agentic, you need that speed and scale to do it.

The good news is the initial cost, what we've seen as an industry, and when I say industry, I mean from a cloud perspective is, initially, years past, it was very hard to manage and control your cloud costs. You would buy in, you would say, "This is great". And then, next thing you know, you get your first monthly bill and someone was running analytics in an office that you didn't know about, and your consumption's through the roof. And it was an issue in the past. What we're seeing is a way better ability to control that now with the data you're using. So, if you want to use AI, it's a little less scary in terms of the observability and how you can control and manage that cost. So, ultimately, that shift to SaaS and the cloud computing, I live in Washington DC, I'm in the data centre hub of the world at this point, it's incredible what is being put in. And I know Mazen will hit on it in terms of the security and adaptability that you can get in cloud now, where we see a really interesting lens for energy customers to go and get more out of their data.

Mazen Younes: Absolutely, Ryan. I'd like to complement and add to what you're saying. I think the energy industry companies, as they look into the space of modernising and transforming their digital landscape, moving to the cloud, as you said, offers multiple benefits, whether it's from a security, adaptability, upgrades, agility standpoint, scalability. So, from the security aspect, the cloud providers already are offering strong security controls out of the box. They're offloading that need of having to manage data centre security, among other things. You don't have to figure out how to go and implement these strong security controls. For the most part, most of the services that are offered by the cloud vendors are NIST 800-53 compliant. You have FIPS 140-3 compliance, which is very important for the nuclear industry, as you see more and more nuclear companies that are more open, for example, to utilise the cloud. And that's mainly driven from the robust security framework that's in place with these cloud service providers. They're really enabling the highest standards and level of security that are generally required by the energy industry.

So, to me, security is mostly a solved issue. It comes to adaptability and agility. You're providing highly scalable computing resources, they can fluctuate based on demands, auto-scaling. You can basically, especially now that you're seeing more renewables and smart grid assets that are coming online, there's a need to make sure that you're able to keep up with the volumes of data and information that's getting pumped into the cloud. So, it also allows us for rapid deployment, testing of new technologies, especially in the AI space; whereas, as you said, the on-prem is really a limitation or poses more of a challenge to get set up and get started, without having to spend that upfront and for investment. It really helps that cloud computing or cloud vendors, SaaS-based products are really helping respond faster as well to changing regulatory requirements, also adapting to customer expectations. You may have requirements to be multi-region availability for your cloud computing resources. So, it really allows you to be more agile as you deploy there.

When you're looking at, for example, upgrades of the infrastructure, whether you're introducing more modern compute, SaaS providers are having capabilities of utilising modern upgrade techniques through CICD approaches that are out there that would minimise downtime for applications. And you're seeing emerging technologies that are being facilitated by these cloud providers with the fast moving pace of large language model releases, agentic frameworks that are out there, and even quantum computing. So, they're making available more powerful infrastructure at the tip of our fingers that we can just provision in the cloud, without having to worry of the ins and outs of how to go and set that up in an on-prem setting.

I think from an operational standpoint, the cloud improves data management. I think we were talking about all the siloed data that's out there. It's more of an enabler as well to help us consolidate all these large volumes of data from these diverse sources for unified analysis. It's reducing the need for on-prem hardware, maintenance, IT staffing. From a disaster recovery standpoint, the geographic redundancy of cloud providers is a major aspect as well that we should consider, especially in these highly-regulated energy industries that are requiring critical software to be up at almost all times. And I think, Ryan, this is probably also a good segue for us, as companies think of adopting cloud, they can also think of more of the future within AI and energy, right, since these cloud vendors now are able to provide the newer or latest technology that's out there, that basically can drive a shift in terms of the future use of AI within the energy industry. So, I would definitely love to know more as well your thoughts there, in terms of how you're seeing that progressing and where you see us going as well within the energy industry.

Ryan Finger: The future of AI and energy, to me, it's interesting in every industry. But I think energy in particular, it's interesting, because people are using AI today and don't really know it, right? We know here internally, machine learning is a subset of AI, they're doing things, they're modelling their data, they're doing work today, that is in that space. So, it's not like everything's getting started from scratch and it's a big greenfield opportunity. There's a lot of structure in place in a lot of these companies to get more out of what's coming. And what I mean by that is when you think about across these siloed systems, I will say full stop energy, if you start seeing, "Oh, we're going to run an autonomous energy enterprise", I know if anybody's listening and you're a reliability engineer on the maintenance side or OT, the alerts go up. And the arrival of generative AI and agentic AI is going to help gather insights quicker for the people with the knowledge of the assets.

I don't see a move where AI is going to be shutting off your turbines for you. I don't think anybody wants that. I don't see an area where it's completely removing the experts from the field. And I see a lot of those promises and I see a lot of these high-level conversations of, "Let's move to an autonomous enterprise. Let's remove X, Y, and Z. Let's give people more time to work on more critical work". We're in energy, all this work is critical. We have a situation now where demand is spiking, people need energy. There's no, I think as it stands today, time to mess around in some of that autonomy. I think where it's going to have a lot of value is to help move data where you need it. So, if you're working site by site or you're in the enterprise, I think agentic AI is a really interesting use case, similar to what RPA has done in a bunch of industries, which is just getting the data you need in the right place faster than you were able to get it before. So, instead of something taking 10 minutes, maybe it takes 15 seconds to get that data. And in energy, the difference of nine minutes is huge. So, I think those are the really exciting areas.

The one thing that's going to happen though, is these digitally forward companies, and this happened in banking, it happens in software, it happens in IT, it happens in every industry, especially with this technology, when you start deploying and using agentic and generative AI, the talent pool to go and do that properly in-house is so tight. You see the OpenAI news of their Head of AI getting a $10 million signing bonus, right? When you think to build these things internally, and you want to use this tech, one of the things to keep an eye out for in the future is like, how do you build that skillset in-house? How do you upskill? How do you train? How do you effectively give your experts the ability to use this technology? Because in some scenarios and for some organisations, it might not be feasible.

So, I see it as a really exciting thing when the use case is very well defined. And to your point, Mazen, about the infrastructure and data silos, that needs to be handled first. And you need to be really, really confident that your data is in the right place and clean and ready to be used, because that's where we see, like, these MIT reports of 95% of gen AI and agentic AI programs aren't showing ROI. So, that has to be done well. But I think the future is exciting. But for our space, I don't see it getting to a point where, like I said, automated turbine shutdowns are removing people from the process. People are going to need to be more involved as these models evolve, because they're going to be dealing with more data. That's kind of where I see things going.

Mazen Younes: Yeah, absolutely, Ryan. And I think gen or agentic AI is gaining traction, as you said. But I think really, it all goes back to driving productivity as an assistant to the human operators, and you mentioned human loop several times already. I think that's key here, because it can help with actionable recommendations, insights. And as you think of the future, AI agents can help analysing large data sets, suggest load shifts, for example, based on grid dynamics, based on fluctuating energy prices. It can act as an optimiser for a human to take action. Now, yeah, in certain cases, maybe the agent can be autonomous, can maybe execute a load shift. But as you said, it's nowhere close to automating the whole end-to-end workflows that are out there.

So, I think essentially, with the help of agents or these technologies that are out there, you even have resource management, demand forecasting, you can ease integration into renewable sources, and how you shift the load from those to the more traditional power sources. But yeah, I think beyond just gen and agentic AI, which is interesting and a disruptor, we also touched upon robots and robotics. I think we're going to see more and more, I guess, autonomous robots that are out there, utilising them further into automating inspections, for maintenance and complex operations as well, right? Think of vegetation management as well for the grid. These robots are also becoming assistants to human operators. So, yeah, I would say it's definitely an exciting and interesting phase of, I guess, technological development. And I think there's a lot to gain within the energy industry, especially in terms of shifting, I guess, the focus of the operators into more productive work.

Ryan Finger: Yeah, and we'll wrap up with a few thoughts. We covered a lot. Hopefully, for those of you listening and following along, it was useful. So, just to summarise my thoughts here, and if I were to give advice or recommendations to asset-intensive organisations looking into this space, I love using non-energy examples. There is a lot to be learned from other industries. Energy is not the first to take the step into this space. There's a lot on consumer goods, banking, manufacturing, for an example, which obviously we cover that space as well. But when I think about the shifts that need to happen, there's a lot of learning that can be done. So, yes, energy is unique, AI and machine learning is being used today. But when you start thinking about how do we justify the investment in this technology, and then how do we prove the return on this investment, there's a lot out there in your AI and machine learning.

So, we're seeing return today. We're seeing our customers with catches on their turbines and avoiding shutdown; we're seeing, to your point, Mazen, autonomous inspection; we're analysing data in two hours compared to two weeks, with computer vision. So, we're seeing the value today and giving folks the insights they want with this technology. But as this next step and people start to add on this, "Hey, we're going to come in and transform the way you work. We're going to fix everything for you and we're going to do it with gen AI and agentic AI", go look at other industries, go learn from other industries. When you look at this technology, to Mazen's point, it's a disruptor. It has a lot of value when used on a very defined use case, where you're able to point back to, "Hey, we increased production. Hey, we minimized mistakes".

So, when you can start pointing back to not only the IT value, which is, "We're not over consuming in the cloud and we're not using too much data and we're not having security issues", when you can narrow this down to, "My people are working faster, they're working smarter, they feel more empowered", and then ultimately, that hard O&M cost with this is really where this needs to start. So, it's exciting. But take a look at some of the use cases out there. It's going to be very hard and continually hard to point back to a return, which is why those core fundamentals that Mazen went through at the start, which is your machine learning, your pattern recognition, we're doing a lot of this today. So, start with the fundamentals, get those right, make sure your systems are set up in a way that actually use and leverage the data you require, and then start thinking about some of this new fun stuff.

Mazen Younes: Yeah, absolutely. Thank you, Ryan. And the only thing I would add is data is the digital gold nowadays. So, with digital transformation and all these AI solutions that are out there, this technology is basically the enabler that's going to help us tap into this digital gold and a lot of value that's going to come out of that for the energy industry. And it's exciting times ahead of us.