How AI is solving real utility challenges [partner content]

Laurent Boinot, a power and utilities leader at Microsoft, remembers the moment he discovered the power of artificial intelligence. Years ago, as a student using a basic AI model to assess World Bank project risks, he was amazed to discover the technology outperformed human experts. "With a very basic AI tool, we were able to replicate that risk analysis and do a better job than the human graders did," says Boinot. Today, Boinot works with utilities to implement AI solutions — an industry traditionally resistant to new technologies. "Our customers want to be fast followers, and in order to be fast followers, you need to follow someone, so someone has to go first. That's definitely an issue in this sector in particular," explains Boinot. Despite this hesitation, Boinot sees AI transforming everything from regulatory processes to grid management. In this episode, produced in partnership with Microsoft, Laurent Boinot talks with Stephen Lacey about where AI can have the biggest impact on utilities. They discuss how AI is supporting field workers, enabling scientific breakthroughs in battery technology, and helping companies like Constellation provide "24/7 matching" of green electricity — moving from "net zero to actual zero emission." This is a partner episode, brought to you by Microsoft. Learn more about how Microsoft is working with utilities and other energy companies to deploy AI to accelerate the energy transition.

• 0:07Speaker 00

  Was there any particular moment when you woke up to the power of artificial intelligence?

• 0:13Speaker 01

  It's a gradual process. I remember reading a book about Alan Turing when I was a kid, and frankly, not understanding a lot of it, but getting some sort of vertigo from thinking that maybe one day, machines could be as sophisticated as our own brains.

• 0:27Speaker 02

  That's Laurent Bueno, a power and utilities leader at Microsoft, talking with Stephen Lacey. Laurent spends a lot of time thinking about how AI can help utilities run the grid cleaner and more reliably.

• 0:38Speaker 01

  It's a really healthy balance between understanding technology, understanding business problems, and bringing in a vision for where technology is going with AI. It's a fascinating role, an intersection of all this.

• 0:56Speaker 02

  Laurent studied philosophy. He was once a chartered accountant, but he's always had a fascination with artificial intelligence. And in 2010, he took a class at Stanford to learn the backend of AI systems.

• 1:07Speaker 01

  I was amazed by how powerful those logic system were. And I had done logic as a philosophy student. And it's incredible how computer science takes those basic math slash philosophy principles and makes them useful for humanity.

• 1:23Speaker 02

  Years later, as a Harvard student cross-registered at MIT, he used a simple AI model to assess environmental and social risk of projects funded by the World Bank. The results surprised him.

• 1:34Speaker 01

  And we looked at the grades that were given by human risk experts, I would say. And then we compared that to actual outcome of the projects. Again, with a very basic AI tool, we were able to replicate that risk analysis and do a better job than the human graders did.

• 1:55Speaker 02

  Laurent learned three lessons about how to make effective use of AI, lessons he now applies to the power sector.

• 2:01Speaker 01

  One is that AI doesn't have to be hugely complex to be very powerful. The second thing is that for any AI system to be powerful and have impact, even before you start thinking about the algorithm itself, it needs great data. And then the third thing as well is that there is, for any AI project to be useful and add value, it needs to be integrated into a workflow.

• 2:25Speaker 00

  You're working on a lot of specific use cases with AI adoption inside utilities, in partnership with utilities. How would you characterize the way utilities are treating these tools today?

• 2:36Speaker 01

  Utilities show some resistance. Often, one of the difficulties we have is that our customers want to be fast followers. And in order to be fast followers, you need to follow someone, so someone has to go first. That's definitely an issue in this sector in particular. But in terms of what can be achieved, it's going to be around integrating renewables. You're going to have examples around grid modernization. It's going to be around enhancing the customer experience as well, operational innovation, security and compliance. All the use cases are going to be around sustainability, affordability and reliability. And that covers everything.

• 3:22Speaker 02

  There are dozens of compelling applications for AI on the grid, but utilities are still in the early stages of adoption. In this episode, produced in partnership with Microsoft, Laurent Boinot talks with Stephen Lacey about where AI can have the biggest impact. They discuss utility adoption trends, barriers to widespread deployment, and why some of the simplest use cases can be the most powerful.

• 3:54Speaker 00

  You just listed off so many applications for AI. There are dozens and dozens of applications. Some of them are really cutting edge in terms of fault detection and distributed energy management. But some of the most powerful ones can be fairly basic or maybe not as exciting, one of which is like using AI in the regulatory process.

• 4:14Speaker 01

  Thanks for flagging this. Because yes, there's a lot of non-exciting things that are really, really key. One of the biggest issues right now for the grid around the world is interconnection. Another big topic is how do you get fast approvals for new projects? And with that in mind, yes, we have a generative AI for permitting solution, which customers can deploy on their Azure subscription, which allows teams to work together on permitting. And permitting, if you take nuclear, for instance, let's take one example. The tool also works for renewables, but nuklears, the permitting can run into the tens of millions of dollars yearly.

• 4:56Speaker 01

  Permitting in nuclear is not only for building, for decommissioning, but it's also operationally for bringing in new fuel, for the transport of spent fuel and for other operational aspects of the plant. Really big teams producing documents every step of the way. With a GNI tool that has ingested all those documents and all the regulations, you can have not only very strong first drafts of the regulation, but you also have the ability to query existing documents to figure out very quickly if what you're looking for is in the document and in a collaborative way make sense of the massive documents

• 5:39Speaker 01

  that a power plant would send to get approval for anything it does.

• 5:43Speaker 00

  So beyond regulatory applications, where are some other promising use cases in core areas inside the utility? What's getting the most traction?

• 5:52Speaker 01

  So we see copilots, so M65 copilot, for instance, which is the GNI layer within the office suite. We see that being used across the board, both for office-type work, but also for field workers. The field workers have with them, for example, a phone or a tablet that will guide them to achieve what they need to achieve with a copilot layer. So you can interact with your phone or even with AI glasses and while you perform a task, not only make sure that you're performing the task in the way that it's anticipated, but also if anything doesn't go according to plan, you can ask more questions that can alert you to potential dangers. A lot of applications that are quite deep across the board.

• 6:37Speaker 00

  I want to move into scientific discovery now, which is really exciting. Microsoft recently demonstrated that AI could screen tens of millions of candidates to discover new battery materials. And I'd like your thoughts on how AI is being used for scientific discovery to accelerate clean energy solutions and then also to help with utility operations and procurement.

• 7:06Speaker 01

  Yeah, this is fascinating. So what you're referring to is a project where AI was leveraged by our research teams to screen over 32 million candidates for a new electrolyte for advanced batteries. Thanks to AI inference, you know, the R and D scientists were able to go from that number to then 500,000. And then with fast AI screening and also high performance compute simulation, go to 100, again, go through a different layer of screening, including high performance

• 7:45Speaker 01

  compute of molecular dynamics, so really simulating the actual molecular behavior with the latest few, 18, to the one with expert guided AI models and human informed validation. So the ability to discover new compounds is really powerful, not only because it allows for potential new material, better batteries, which would go a long way to make the grids more stable and integrate distributed energy resources, but also if you think about the big power plants, some of the output of a nuclear power plant can be leveraged and are, you know, the output is being used today for cancer treatment.

• 8:28Speaker 01

  Those isotopes are treated by medical companies and then used in radiation therapy. And you can imagine having high performance compute simulation as well there to potentially find new usage or find different isotopes. The material science aspect of this is really interesting.

• 8:48Speaker 00

  Absolutely, and you can see that there's will directly influence companies that are developing new technologies, but bringing it over to the utility, how can these high-performance simulations maybe help utilities evaluate new technologies? Are there ways that they can integrate some of this into their own thinking about the energy transition?

• 9:10Speaker 01

  I'll mention to Hydro Quebec, for instance, as a whole menagerie of robots for inspecting power lines, for inspecting, so the actual line, but also for inspecting vegetation around power lines, preventing fires. All that, again, you know, very boring, kind of maintenance-type conversation, but they have a really, really big impact when things go wrong. And so all those robots, you know, there are a lot of AI in the background to code those robots, to interact with those robots. Another customer where we have a tight partnership is Constellation, and with Constellation, we have a project called 24/7 Matching, which actually not a

• 9:50Speaker 01

  project, it's a product. And with that, Constellation is able to tag its green electricity hourly, which means that when you buy green electricity from Constellation, you're not buying net, you're actually buying actual green electron. So the difference is, the electron you're using is green, and so that's really powerful as well, to go from net zero to actual zero emission.

• 10:16Speaker 00

  So anyone who has worked with utilities or watched the utility industry really understands that it's a very conservative industry for obvious reasons, security reasons, they're operating critical infrastructure, which is often under threat, they're really focused on reliability. What would you say are the biggest technical or organizational hurdles that you're facing when exploring these tools in partnership with utilities?

• 10:42Speaker 01

  So some of the hurdles are regulation or the perceptions of regulation. So for instance, if you think about NUC SIP, which is the North American Electric Reliability Corporation's Critical Infrastructure Protection Standards.

• 10:58Speaker 01

  There is a perception that what NUCIP does is prevent you from doing anything in the cloud. It's not the case at all, and our customers get it. But initially, there is this perception. You can think of it as traveling to an airport. In the past, you used to jump on an airplane, and now you go through TSA. If the TSA goes from 30 minutes to three days, nobody travels anymore, right? You need to find the balance between security and unlocking all the opportunity allowing you to actually manage the grid properly. And actually, AI has been used by a customer like Uniper, which is one of the leading multinational energy company based in Germany. And they use a tool called Security Copilot. And what that specific copilot does, it goes from, you know, flashing red lights on your screen,

• 11:41Speaker 01

  where as a human, you have to understand what the red light means, to actually telling you in a natural way that actually, you know, those two things that are flashing are connected. It's two doors in the same corridor that are not closed properly, and therefore, there's a risk of intrusion. Security Copilot can also allow you to make sense of your IoT devices connected with Defender for IoT. So again, with the complexity of securing those really constantly targeted customers, like the electric providers, having the ability to automate some of the day-to-day grind of reports, of alerts, makes a really, really big difference and empowers the human to do a lot more.

• 12:25Speaker 00

  Where would you say we are on the adoption curve?

• 12:27Speaker 01

  So it's a tough question to answer for two reasons. One, because customers are different steps. And also because the, when you say adoption curve, you mean getting to somewhere, and that somewhere is constantly moving. But if I think about, you know, generative AI for office workers typically, there's a lot more that can be achieved, for sure. We see in some cases, very significant efficiency gain. I have a customer that did a report where they found that after just a few weeks of usage, each employee had a, or achieved a day more for activity every week. So that's, that's very significant. And I expect that to be eventually rolled out across the, across the board.

• 13:08Speaker 00

  I know you've been thinking a lot about the energy environmental impacts of AI as well. And there's this ongoing debate, not just in the energy sector, but now in the mainstream about AI's energy consumption versus its efficiency benefits. And I'm wondering, based on what you see in the field, like these actual deployments and ways that they can help utilities become more efficient, adopt more clean energy, help companies explore new types of materials, how do you evaluate that trade off? With the obvious increase in energy from data centers, but the societal benefits of AI applications?

• 13:46Speaker 01

  Yeah, you're right. Data centers use a lot of energy, but you'll have to use a lot of very dirty energy to compensate how much more efficient you can make the process of identifying sites for clean energy. And I'll take one example. We have a partnership with an NGO called Terapraxis. And with Terapraxis, we are identifying coal power sites for repowering to nuclear and SMR, so small modular reactors. And that is really powerful because the coal sites are connected to the grid. You know, otherwise they wouldn't be there. They're already classed as industrial, so there's

• 14:30Speaker 01

  less of a risk of, you know, not in my backyard type behavior. And once those sites are identified with a lot of AI algorithms, then you can leverage the permitting with GenAI as well to make sure you are as efficient as possible in building the plant and getting it offline with carbon free electrons. And the other part of that conversation that is really interesting is that if you think about cutting edge research in AI, there's almost two labs in the same building, if you think about it. One lab where the goal is to do the most impressive thing, you know, those are the cutting edge models, right? In the other lab, you take the cutting edge model

• 15:14Speaker 01

  from two weeks ago and you say, okay, now this is the capability, how cheap and how light can this be? So there's almost two curves. There's the cutting edge of AI and what we can unlock with increasingly sophisticated models. But on the other end, there is like how cheap, simple and light and eventually is sustainable can compute become.

• 15:37Speaker 00

  So you've got this unique window into this array of solutions. What do you think is going to surprise people about adoption or capabilities?

• 15:45Speaker 01

  One other thing I'm thinking about, for instance, that we haven't talked about is agentic AI, where instead of connecting two separate software solutions with APIs or with some sort of convoluted connector, you can have an AI system working across the board to achieve a task across systems. So that's for sure going to get really powerful and really important. But maybe the biggest surprise would be that even with self-driving cars, even with some of the crazy sci-fi scenario that now we are starting to envisage, maybe the most surprising thing is that there's nothing new under the sun. Humans are still humans.

• 16:30Speaker 01

  We are still good at what we are good at. And like all other previous massive changes that we've gone through throughout history, we'll find a way towards a new normal and a new obvious. We are very good at finding things normal when we grew up with them. So that's what I would expect.

• 16:48Speaker 00

  Laurent, thank you so much. I really enjoyed this.

• 16:51Speaker 01

  Thank you, Stephen. Thank you for having me. Please reach out on LinkedIn if anybody wants to have an AI, gen AI, powered G2 conversation. I'd love to hear about other use cases people are working on, any thoughts you want to share. I'd love that. So thank you.

• 17:09Speaker 02

  To learn more about how Microsoft is working with utilities and other energy companies to deploy AI to accelerate the energy transition, go to microsoft.com/industry/energy.