The Exchanges

Q why it's actually going to be tough, and in some ways why that first wave, as you said, haven't, hasn't Played out as expected. Like, why is the electricity sector a hard one? Why is it not the first sector? Like, there are lots of sectors wherein AI is completely transformative already today. I think of, like, the legal profession, for example. Why is electricity probably not one of those?

A Right. Yeah. I, I think it, oh, the, the, the primary reason why, you know, we're not on a path Towards, you know, full automation, and, and there's, you know, been a spade of use cases launched is just modeling grid physics accurately is incredibly difficult, right, for an algorithm. So, you know, solutions on power flow obviously need to take into account, you know, Ohm's law and Kirchhoff's law and, um, accurately, um, um, predict or, or make real-time decisions. So there's a limit to where AIs can be deployed currently, um, for, for real reasons. I, I think, um, Just the complexity of the system. Like, if you look at a Sankey of any utility, right, and you've got the data sources, and you've got the data management platforms, and the visualization platforms, they are some of the most complex diagrams that you could witness. So it's not an easy operating environment for, um, for solution providers, or, or for solutions to be embedded. And I, look, I think there's the typical industry issues around just rate basing new technologies, right? Um, um, but more broadly, just for some of these use cases, the data's not available, Shale. Um, it's not frequent enough, right? So we don't have the millisecond, um, at, um, granularity that's needed for some of these use cases. So that makes it all a little bit difficult. I think it's also, you know, there's a torturous sales cycles for …

AI assessment note: “the primary reason why... is just modeling grid physics accurately is incredibly difficult”

Q a lot, and it means different things to different people. You know, there's the LLM, ChatGPT version of it, and all this generative AI is a new phenomenon, but the, but AI sort of interspersed with machine learning and other tools, it's all over the place. As you think about it, specifically in the power sector, like, what are the subcategories of AI that we should be Thinking most about?

A Yeah, that's a good question. And I'll, you know, and that's the, you know, the very start of a conversation with a utility or with any stakeholder in this industry. Defining AI is, you know, it's, it's an interesting, um, um, question in terms of how you frame it. I think one way that we've been looking at it is defining it through a set of capabilities. So you can understand it from a business perspective, right? So looking at machine learning and predictive analytics, right? So these are, Algorithms that detect patterns or can make decisions. That's one set of, of solutions. Computer visions. So these may be cameras that interpret or understand the visual world, right? Infrared cameras and so on. Another set. Natural language processing. Interpreting text, voice, human language. Robotics. Anything that's doing a task in a robotic nature that a human would do, right? Perimeter security by, by a robot would be classified as AI. I think, you know, areas like digital twins. So these are, you know, virtual replicas that move beyond things like anomaly detection into scenarios and decision making. You mentioned LLMs. We're seeing some application of LLMs across the industry, but usually at the enterprise level around regulatory documents around, you know, generating text or, or guidance. Less on the, on the actual power sector side of the operation itself itself. Um, I think we lo…

AI assessment note: “I think one way that we've been looking at it is defining it through a set of capabilities.”

Q Right. Okay, well, Powerflow, you mentioned we'll get to one of those. For number three, can we talk about something that is in the sort of core operations world of, of Powerflows?

A Yeah, we, so yeah, there's, there's two here, right? I think in core operations, the, the, the first one I mentioned, because this is highly important for utilities, right? This, you know, substation, asset management, it doesn't light everyone's world on fire, but there's huge dollars associated with this, right? So, um, you know, you've got vibration sensors, partial discharge sensors, gas sensors, temperature sensors, you've got inspection drones and, and robots, and they're looking for, you know, wear and turn equipment. They're looking at SF, Six gases. Um, they're looking at, at insulation breakdowns and so on. And what's interesting is, again, the algorithms have got so much better in the past five years. So these computer vision platforms, right, they're analyzing video feeds from fixed cameras, and they're able to detect corrosion, right, on these assets. Um, the machine learning platforms, which can be built on, on, on some of these open source tools. Um, and then made proprietary are able to, um, um, look at some predictive maintenance strategies, build it into existing digital twin software where you're moving beyond anomaly detection into, um, into digital replicas and scenarios and so on. And I think that's, that's, um, a huge value use case for, um, for utilities, right? Like it can't be understated, right? You can reduce downtime Of some of these critical compon…

AI assessment note: “I think in core operations, the, the, the first one I mentioned”

Q and GE and Siemens and ABB and the, the typical large utility vendors who ended up rolling out most of the new technology. There were exceptions, but, um, but that, that's, you know, tends to be how it goes in this sector. As you see this new wave of innovation coming, who are the suppliers Delivering the new solutions, and is it any different from how it has been historically?

A That's a great question. I, I think it is a little different, Shale. Um, I, I mentioned that, you know, the, the cost to develop these solutions has kind of collapsed a little bit with these open source libraries and frameworks and packages. So, and, and we were purposeful in this work to look beyond the conglomerates and the grid joints to, to really, you know, um, get a temperature check on the market. And what we found was a, a, a series of startups here looking at discrete problems, right? So, um, um, and across all of the, the areas where, where we mentioned it's the most pressing, even across the use cases, we mentioned there's a series of startups. So there's, there's a lot of activity. We looked at, I think, 350 to slow deployments since 2001. Um, and we had a, a, a focus on startups. Um, and I think what we, what we found was That startups, um, received 1.5 billion, right, from 80 rounds of funding in, uh, related to those deployments since twenty-twenty-one. Um, and that, so that, that's, that's a pretty, um, sizable capital inflow.

AI assessment note: “I think it is a little different... what we found was a series of startups”

Q it faster. Um, the third is actually, yes, I think, I think something will fundamentally change and the opportunity will be there to like, I don't want to say rip and replace, but, you know, execute a fundamental transformation of how electricity is delivered, um, in a relatively short period of time. Like of those three options, Or a fourth one that I haven't thought of. Where do you sit?

A Right. Yeah. Good question. I think there's, you know, there's possible and preferred and potential futures across all three, right? And they're all beholden to different characters. I certainly think that it's potentially possible. We're a few years away from being a few years away from that though. So I think with the timeframe is, is really important. Um, I think we could look, if there was the right investment Um, and, and the right technologies, and we can talk about some of the challenges to get there. Um, that, you know, your third option is possible. I, I, is, is, um, is, it could be a reality. I think what, I think number two, I think the incremental change over time, um, will lead us to a system that, um, eventually becomes very automated with the support of regulatory infrastructure, um, and the, and the support of proven technologies through rate cases and so on. I think we'll get there. Um, but, but, Shail, it's really important to underline in this that, you know, the, the data right now is, is not available, right, for, let's say, real-time decision-making across the grid, right? Like, what we have is, um, AMI, right, which is minutes to hours. We, we have SCADA, which is, you know, seconds to minutes, and then we have PMUs, which are kind of millisecond, um, um, frequency and granularity. I mean, for that future, you're talking about what we'd need is like a A h…

AI assessment note: “I think number two, I think the incremental change over time, will lead us”

Q a topic that everybody's talking about in every sector. So let's start with, like, why this one? And maybe talk a little bit also about why maybe not this one, but starting with why this one, like, why is the electricity sector, or electricity and gas, I suppose, the world of utilities, like, why is it an especially interesting one with regard to the possibility of new applications using AI?

A It's a good question, and it, it really is, um, kind of the, the, the biggest question that, um, you know, folks ask is, how, how does AI apply to the electricity sector? It's, Maybe not something you equate immediately, right? Um, but I, I think the thing to say, Shaila, is that there's been a ton of investment, right, in, in IT infrastructure and OT infrastructure, and, you know, all of these smart meters were at over 70% smart meters in the US. There's all this, you know, digital infrastructure, and this is from years of, of, of vendors and rate cases and so on. So there's, there's an appetite right now to, To kind of, you know, prove out some of those promises from over the years, um, and, you know, utilize all of this data and whatnot. So I think, I think that's interesting. I think what's more interesting is the actual problems that AI is trying to solve in the sector. Um, and, you know, if, if you, if you kind of stand back and you look at, you know, what the power sector is trying to do and take a leading role in decarbonization, and you think, right, well, if AI Um, is, is supposedly is a set of computational processes that can perform a task at human intelligence. I mean, the scope is actually enormous in that regard. So, um, AI has been put to use on, on various parts of the grid across the whole value chain. There's a whole set of use cases, but primarily it's been …

AI assessment note: “there's been a ton of investment, right, in, in IT infrastructure”