AI’s Energy Paradox: More Compute Could Cut Industrial Emissions

What if AI’s biggest climate impact isn’t chatbots, but cutting real energy waste in buildings, grids, and factories?

In this episode of Climate Confident, I’m joined by Philippe Rambach, Chief AI Officer at Schneider Electric, to unpack one of the sharpest tensions in climate tech today: AI is increasing electricity demand, but used well, it may also be one of the tools we need for decarbonisation, emissions reduction, and a faster energy transition.

You’ll hear why Philippe argues that the real opportunity is not in chasing every shiny new model, but in applying AI to physical systems: reducing peak demand, optimising building energy use, supporting grid operators, and helping companies move from pilots to production. We dig into Schneider Electric’s work on using AI to cut energy waste, including the striking claim that in some energy-saving applications, the carbon emitted to run the model can be dwarfed by the energy saved.

We also get into the hard bits people love to ignore because apparently spreadsheets and wishful thinking are still considered strategy in some quarters. Why do so many AI pilots fail to scale? Why does domain knowledge matter as much as technical skill? How should businesses think about responsible AI, privacy, policy, net zero, and the operational realities of electrification?

This is a practical conversation about AI for energy, not AI theatre.

🎙️ Listen now to hear how Philippe Rambach and Schneider Electric are applying AI to real-world decarbonisation and climate action.

Sign up to Climate Confident+ for deep dive analysis of the major climate and energy stories of the day.

Transcript

Copy

Philippe Rambach: 0:00

So in short, using AI to reduce energy costs is working well. We have made computation that probably there is a ratio between one to 500 between the carbon emitted to run the model versus energy saved.

Tom Raftery-1: 0:23

That's the tension at the heart of this conversation, AI is increasing electricity demand, but it may also be one of the tools we need to reduce it. Good morning, good afternoon, or good evening, wherever you are in the world. This is Climate Confident Stories and Strategies that Cut Emissions episode 282. And I'm your host, Tom Raftery. My guest today is Philippe Rach, Chief AI Officer at Schneider Electric. Philippe makes the case for a more practical kind of AI, not just chatbots and giant models, but systems that cut energy waste in buildings, shift demand away from peak hours, support grid operators, and help companies move from endless pilots to production. Let's get into it.

Tom Raftery: 1:14

Philippe, welcome to the podcast. Would you like to introduce yourself?

Philippe Rambach: 1:19

Sure Tom, thank you. Super happy to be there with you today. So Philippe Rambach I'm the Chief AI Officer of Schneider Electric. I'm based in Paris, as you can hear from my beautiful French English.

Tom Raftery: 1:32

Great. And Philippe, you've run multi-billion Euro industrial businesses, and then you shifted into becoming Schneider Electric's first Chief AI Officer. What pulled you into AI?

Philippe Rambach: 1:47

Actually, I'm not sure that I pulled myself into AI. I received one morning, a very strange call from one of our execs saying to me, Philippe, you know what you have done a good job so far, but we would like to give you something totally different. Okay, what? We'd like you to run AI. And that was five years ago. That was 18 months before ChatGPT was even released. So that was a time when AI was fashionable, but a bit less. So I kind of asked our CEO of at that time, Jean Pascal, why did you exactly pick me? I don't know software, I don't know AI. I don't even know IT. I like my multi-billion current job. I'm not so sure I want to take a zero million job. So why me? And then that was interesting though. The answer was, Philippe I've been thinking who I need for that job. And I need somebody who first understand our business very, very well to do an AI that impacts. To not fall into let's do a shiny object, a funny demonstrator, but let's impact our employees. Let's impact our customers. I know somebody also will understand our company. AI will be certainly a technical game, but it'll be also a transformational game. To transform a company, you need to understand the company. And then the last point was, ideally I would like somebody who also know AI, but I don't think I will find somebody who knows the business, knows the company, and knows AI. You are French. I'm calling you end of July. You probably planned a four weeks long holidays. You've done a bit of mathematics when you are much younger, a lot of time to learn. So that's how I fell into AI. And I can tell you zero regret.

Tom Raftery: 3:25

Okay, good, good, good. And listen, I should have asked you earlier, but for people who are listening who might not know Schneider, can you give us just a quick 101 on Schneider Electric?

Philippe Rambach: 3:36

Schneider Electric is a multinational company, quite large, more than 40 billion revenue with two main businesses. We have one business of energy management and one business of industrial automation. But the two businesses are focused on the same thing. We want to make our customers more sustainable, which is a very broad ambition. We do that by helping them be more efficient, and we do that by helping them being more efficient into domain, how they use energy. We help them use less energy cheaper energy greener energy, and we help them in their industrial processes to manufacture things, to build things with less efforts, less investments, less energy. So energy management, industrial automation, focusing on helping our customers be more efficient.

Tom Raftery: 4:24

Okay, fantastic. And when you started the role in 2021, what did Schneider already see coming that most companies had missed?

Philippe Rambach: 4:33

I think that what we were seeing was the data journey had already started since a long time, and if I look for example at what we do for our customers, we have been trying to help them be more efficient since maybe, I don't know, 15, 20 years. We started first by providing them products to electrify, to automatize, to make things in a more efficient way with less energy losses, et cetera. And that's good, and that has to be continued. Then the next step was connecting those products to help our customers understand better their processes. Where do they lose energy? Where do they have inefficiencies? So we use these products to collect process data, and the process can be industrial process, but it can also be heating and cooling a building, powering a data centre. Then after that, we moved into visualisation. So we developed our softwares. We made some large acquisitions like Wonderware, where we gave our customer the ability to visualise and act on this data. Then we moved into storage of data with company like OSIP and our own development. You see me coming the next natural step in this long journey of bringing always more efficiency to our customers who was putting AI on top of that. Not all customer are there yet, of course, but enough of them were there. So that was making sense to make that big investment, to move one step forward into this direction. At the risk of disappointing you Tom, we had no clue that ChatGPT was coming 18 months later. At that time, we were all in on analytical AI, forecasting, anomaly detection, which we still do, of course, but we have added to that all the generative AI.

Tom Raftery: 6:17

And how is this AI that you're adding onto your systems, making your customers more efficient?

Philippe Rambach: 6:23

I will say maybe there are three different ways that AI will help our customers. And I'm sorry, Tom, I may be a bit tech technical and a bit theoretical, but then we'll go into practical examples.

Tom Raftery: 6:35

Great.

Philippe Rambach: 6:36

The first thing is the energy, which is the cheapest, the greenest, and the best is obviously the one that you do not need. Okay? If you can do the same heating or cooling your building, powering your data centre, manufacturing your things with less energy, that's good for everything. And here, AI really can help, especially for machine learning, because with AI, you are able to create model of processes that without AI you cannot create and then use these models to optimise. The second topic where AI play a very strong role is the fact that if we want to truly decarbonise, if we really want to use cheaper energy, we have to move to electricity. There is no other option, but electricity is not always decarbonised. And especially when there is a peak demand of electricity on the grid, the grid operator needs to balance production and demand. And when he has to add production, quite often the only thing he can do is to start carbon heavy source of energy, fuel, gas, coal. Because you cannot decide to start a wind generation, you cannot decide to start a solar panel. So long story short, if we can avoid to use electricity at peak demand, we help the operator to produce greener electricity. So here, AI again, by learning the behaviours of our factories, by learning the behaviour of our buildings, by optimising and forecasting the usage of solar panels, batteries, et cetera, can help shift the demand out of the peak demand, and therefore help our customers help the planet to use a greener electricity and additional benefit, a cheaper electricity. Because at peak demand, not only is electricity heavily carbonated, but also quite expensive. So I like those two things because not only we are helping people to be better for the planet and energy that you don't choose do not impact a greener electricity impacts much less, but you also help their wallet definitively because using no energy is cheaper, whatever you use. And using electricity out of peak is good. So that's the two main thing. And the third one, which is probably equally important, is AI can remove barrier to adoption. There are plenty of things that we can do to reduce our energy cost to reduce our impact on the planet, moving to electricity changing our processes, introducing renewable. But for all of that, we need to make things easier to operate and definitively AI can help. So in short, three domains use AI to create models to use less energy, use AI to average the demand and avoid to buy electricity and use electricity at peak demand. And third, use AI to simplify the adoption of all these new technologies that makes our customers more efficient.

Tom Raftery: 9:26

Okay, very good. So, most companies talk about AI as software productivity, but you're talking about grids, buildings, factories, physical infrastructure. Do have concrete examples where AI materially reduced energy demand, or shifted load away from peak times?

Philippe Rambach: 9:44

Yeah, of course. And I will start by probably something which could sound very mundane or, or very, very simple, but is incredibly concrete. In almost every building in the world, in every hotel, in every hospital, in every office building in your room, probably today Tom, you have a room controller. Something that control the temperature of the room. Exactly. What does it bring to you? It brings you comfort, the right temperature, and it brings you energy saving because usually you try to set it at a lower temperature or when you're not in the room to not use energy, we're not using the room. Good, excellent. If I come back to my story of data, we were selling room controller. The first level to save energy was to connect those temperature home controllers to the building management system so that we heat and cool only when used. The next level was using AI to go to the next optimisation. Let's say that the building management system knows that you need this room at 10. To get it warm or cool enough if you live in the south at 10, when do you, when do you need to start to heat or cool?

9:30, 9:45, 9: 10:56

15. No way to know. So either you, compromise on comfort or you compromise on savings. We have embedded in this less than $10 chip in this very not expensive room controller, an AI module that learned the behaviour of the room and is able to decide exactly when to start, when to stop heating and cooling and save 20% energy. When you know that 30% of carbon emission come from buildings, if you can cut 20% energy with something as simple as a room controller, it's impressive. So that's on the probably lower end from a technology, not from a saving. A saving are big to the higher end. Let's say that you are an operator of a distribution grid, so you have to provide electricity to homes, to factories to buildings. In the past, it was fairly easy. You had a big chunk of electricity coming from the tranmission network, and then you were distributing. Not easy, but not that difficult. Now you make it more complex. You add to your distribution grid generation, you add solar panels, you add maybe wind generations. You add a lot of electrical vehicles. That make the grid much more difficult to control. Here again, AI helps by analysing, providing recommendation, helping operators, managing alarms, et cetera. So in short, I mean, I don't think the energy transition can happen without AI. And AI will help both on saving energy, will help on driving the grid, which is becoming much more complicated. And also, I mean, they can finish with this one shifting the demand. Take a building like the one we are sitting here, the one you are sitting. I hope that people, because they are climate conscious and cost conscious, have put solar panels on the roof to generate some electricity. If you use what come from the panel, it's good, it's great. You do good for the planet, but that may not be the most efficient because when the sun is shining, like today in Paris, everybody produces. Electricity cost may even be negative. You may be even a problem with your production. Or you think AI, you learn and learn how to be able to forecast the consumption of your building. You learn to be able to forecast the production of your solar panels. You get from the grid when there will be high demand, low demand, expensive electricity, cheap electricity. And then with the classical optimisation problem, I told you I would be a bit technical, sorry for that. You then decide every 15 minutes if it's better to use what you produce, to sell it to the grid, to store it for tomorrow, to take in what you have stored. So and that are practical solution today, commercially available by us and by our competitors, but that provide this solution that exists today. And we talk double digit impact in term of savings.

Tom Raftery: 13:47

Impressive. Just pushing back against that a little for a second, because a lot of people hear AI and climate and immediately think of massive data centres and emissions because they're all being now powered by big gas turbines. So are we underestimating AI's climate upside, or are we underestimating its footprint?

Philippe Rambach: 14:12

As for many things, I think it'll depend on us. First I would like to put a, a, few numbers in perspective. The total energy consumption of the world today for everything we do is about 120,000 terawatt hours per year. The total data centre consumption is about 500 terawatt hours outta which AI is maybe today 10%. So today we're probably about 30 to 40 terawatt hours for AI compared to 120,000. And tomorrow this 40 may become 600. So yes, there will be a massive increase because we may multiply by 10 or 15 the amount of energy used for aI, but still compared to the 120,000, it's still small. So next the question comes is what do we do? If we were using all our AI compute to optimise energy, reduce cost, et cetera, the balance would be definitively positive. So using AI to support energy transition and reduce energy cost definitively is on a positive balance side. On the other side, if you use AI to generate fake news definitively, you are not helping the world. You are consuming energy. So in short, using AI to reduce energy costs is working well. We have made computation that probably there is a ratio between one to 500 between the carbon emitted to run the model versus energy saved. But of course, if you use a model to do things that don't save energy, of course you don't impact.

Tom Raftery: 15:47

And how do you think about the contradiction that AI both increases electricity demand and helps reduce it?

Philippe Rambach: 15:55

It's interesting. And for me if I start with my very positive eyes, what is good with the data centres and so on is they are electrically powered. So technically there is ways to power them with green electricity and green power, which will never be the case of a diesel engine by definition. So But this is possible to power data centres and that's why I'm a strong advocate to build more data centres in Europe, where there is a lot of green energy is available. So definitively, technically we can do that. Second in Schneider as other companies. But in Schneider we work a lot on what we call energy for AI. So how can we reduce the, the amount of energy for an increased amount of compute? We elevate voltage, we work on density, plenty of technology to try to reduce the amount of energy. But overall, you're right, we'll need more, we'll need more electricity. On the other side, we work also on what we call AI for energy, which is everything I've been trying to describe so far to help reduce. Again, for me, the biggest question is not so much on the absolute value because again, compared to the total energy of the world, it's not that big. But it's more the fact that we have been used so far to grids, which were moving very slowly. We're not adding a lot of new capacity every year. So the biggest challenge for us, for me is not the absolute amount of energy, but how are we able to change the way to accelerate the bringing of the electricity, building new transition lines, building new distribution lines to those data centres. That, that's probably what I, what I would say,

Tom Raftery: 18:04

And do you think the market has become too obsessed with these big ChatGPT, LLMs, Anthropic, et cetera, while ignoring the kind of boring industrial AI that might have a much bigger climate impact?

Philippe Rambach: 18:18

I would not say too obsessed because there's a lot of good things, but definitively generative AI, large grid models do not replace what some people qualify as traditional AI. Even if I can tell you, there is nothing traditional, and there is new things coming every day in this more analytical, predictive AI. So the big mistake would be both to say generative AI has no value, and to say it has replaced the old AI. The reality is we need both. It would make no sense to try to use an LLM to optimise the temperature in the room. At the same time, you need LLM to generate reasoning capacities to help somebody drive an electrical distribution grid. So there is an AI for each application, and we should not start from technical. We need to start from the need and choose the most frugal, the most efficient AI for what we want to do. There are plenty of things that can be done with LLM, but do not require LLM. And there are of course, plenty of things that do require LLM. So my strong message is it's not one or the other. It is the two. And there is a lot of things in the industrial world, in the energy world that will fall long be done with what some people would classify classical AI, analytical AI, traditional AI. While of course a lot of new possibilities are opened with LLM, agentic, and reasoning.

Tom Raftery: 19:45

And we're seeing a huge amount of pilots out there in the AI space now, but a lot of these AI projects fail to scale. Why? Why do you think that might be?

Philippe Rambach: 19:57

Yeah. There used to be a joke that was running around AI officers, which was, I'm not running Air France, I'm not running KLM, I'm not running Delta, but I have more pilot than them. So you're right that, that, that's the biggest risk and, and that risk is even bigger with generative AI because it's extremely easy to make a pilot, but scale it is difficult. So the big challenge for AI is certainly is a technical part, but really the transformation and the scaling if I look at, at Schneider, what we have done is from day one, because I come from the business, because I really wanted to impact our obsession. What was we call doing AI at scale and to avoid to fall in the pilot pitfall or the pilot yeah, risk. The very first thing that everybody should think is always start from the business value. Never start from, Hey, a new agent has been released. A new LLM has been released. What can I do with that? You can do plenty of things, but start from what do my customer needs? And by the way, usually AI did not change the customer need, but brings new way to answers, better ways what do my employee needs? Where do I lose productivity? Where am I inefficient? And then the second thing that we have learned is it's high time to stop to treat AI as an innovation. Which mean that the process for put in place to deploy AI are very similar to the one we have to deploy traditional transformation. Which means we put a team. In that team, we make sure that we have technical people from AI of course, but also business people, transformation people, the people who will do the pricing for the customers, who will do the sales training, who will do all of that. And this team is not in charge to demonstrate in a pilot that it is feasible and then somebody else would industrialise. Because guess what? It is never industrialised. But that is in charge to go till the end and to deploy it in production. And the job is not stopped until we have reached our KPI of adoption sales performance and all of that. And by doing that, we avoid to focus only on the technical side. Because quite often when people do pilot, they only clear the technical visibility, but they forgot or they forget the importance of legal visibility, customer acceptance, market price, customer benefit, all the thing that we know when we do other project, but that we kinda forget when we do AI because we get blindsided by the beauty of technical. So super important to not forget what we have learned, how we make transformation, how we bring a product to the market. All of that applies to AI products. And the biggest mistake that for me explain this number of pilots are two, once again, not starting from business value, but starting from, yeah, it's super exciting from a technical point of view. And second, treating it as innovation. Saying a team will do a pilot, another will industrialise where we should have one team in charge of moving from the ideation till the production, at least in our companies. Maybe a bit different if you are extremely innovative working on things totally new. But for us, where our mission is to take what is very, very new from startups, from innovation and bring it at scale to customers in the industrial world, in the real world, that's the key for me. Business value focusing from day one on what do we need to be in production?

Tom Raftery: 23:24

Okay, so it sounds less like a technology problem and more like an organisational change problem. This is, I, I've heard this so many times in so many different podcast episodes, and I've said it myself many times. Technology is easy. It's people that are hard.

Philippe Rambach: 23:41

Absolutely. And sometime when people come to me with questions my first reaction is, let's make a mind experiment. Let's forget for one second it's AI. Is your question specific to AI or is it a question that has been there for 10 years and that we know it's difficult to solve, but we have done that 10 times already? And do we need to change because it's AI or just apply what we have learned as a company on how we transform, how we train? And that doesn't mean it's easy, it just mean that we don't need to reinvent everything for AI. We have to reinvent a few things. For example, it's extremely important to train our employees, explain our customers, because AI comes with a certain weight and we need people to understand. So there are specific things for AI definitively, but maybe 80% of it is things that we have been doing in the last 20 years of transforming, deploying, selling, explaining, et cetera.

Tom Raftery: 24:35

Okay. And one of the new things that has come about is the role of the Chief AI officer. Is that a genuinely necessary role or are companies just creating another executive title?'cause everyone else is doing it?

Philippe Rambach: 24:49

First we have to admit that it made my mother very proud. Only for that. Only for that, it was worth it. But more, more seriously. It's a good question. You definitely need it, certainly for a period of time. And I think we will need it as long as the technology is moving that fast because you need somebody who is really dedicated and focused on understanding what it can do, how it translates for your business. And with things changing every month these days, every quarter. I think you need dedication. In the future, if one day technology stabilises a bit, maybe we can remerge that into more traditional role about transformation, role about digital roles and so on. But I would say today, keeping the pace, having the focus probably needs dedication.

Tom Raftery: 25:43

Okay, and you oversee a 350 person AI organisation. What would you say, what skills would you say are hardest to find right now?

Philippe Rambach: 25:53

At first, it's a bit easier than it was four years ago when we started. The academic world has reacted very quickly and quite nicely in providing quite many young, young engineers with lots of AI knowledge and capacity knowledge. So I would first say that honestly, it's a bit easier than four years ago. Definitively. What remains difficult and most impossible to find is people who would've both AI knowledge and domain knowledge, because those people coming out of universities by definition, do not have domain knowledge. And I know in Schneider, some people will tell you that you cannot really understand complex electrical things without 15 years of experience. So I think hoping to find people with domain and AI knowledge is a dream that will not happen at least in the next five to 10 years. So the game is more to find organisational way to merge the domain knowledge with the AI knowledge in multidisciplinary team that are able to deliver value and help each other learn from each other to deliver the value. And the biggest mistake would be to create isolated AI team fully in charge of deploying by themselves solutions, because they will probably have the wrong understanding of the market, the wrong understanding of the technology, of the needs, and all of that. So one of the thing where in mind was to say, my team is on purpose, very, very technical, so that we cannot build any solution without making a team with our line of businesses, with our function, and have a team which is half made of technical people coming from my team and business people or function people coming from our line of businesses or functions. And by doing that, we force the collaboration. One of my fear was to, the easiest way would've been to duplicate, yeah, I'm gonna hire two or three sellers to go fast. I'm gonna hire two or three marketers to go fast. But then you go further if you don't go far. So really, and for each company that's diff it's really related to your own company culture and organisation. But the key challenge is on one side, hiring AI talents getting a bit easier, but still as difficult as before, find the right organisational way to merge your company domain knowledge with this AI knowledge that you bring into the company.

Tom Raftery: 28:16

Okay. And speaking of that domain knowledge, I mean, if we look at the energy grids, for example, Europe's energy system is becoming a lot more distributed, electrified, and renewable heavy. So what complexity problems does that create that humans simply can't manage manually anymore?

Philippe Rambach: 28:37

Someone said, a don't remember whom, and I don't know if it's true, but that the electrical grid is the most complex system ever built by a human. And and that's probably true because when you realise that at the end everything is connected, I mean, it's like a gigantic circuit at the size of a continent, if not a planet. So yes, it's usually complicated and the fact that we add now renewables makes it even more difficult to, to drive. Basically, and, and I don't want to go into a, a technical lecture about grid, but there are two main things that you have to control always is the stability of your voltage, your power, et cetera. And the introduction of renewables make it much more difficult because you lose natural amortisers of difficulties. You had a kind of a mechanical way that was helping us stabilise when you were getting destabilised. And the second thing is you always need to match production and demand. Otherwise, your grid very quickly collapse. You cannot really store. So you have to have same production, same demand. AI help on both. I will not come back too long on the demand side, but a lot of our work in Schneider on AI is doing that. Reducing demand, avoiding peak demand, helping people having a more stabilised power because we have a better maintenance, higher quality maintenance, we avoid disruption. So the first thing where AI will help a lot is optimising demand side to make the life of the grid easier. The second thing that AI can do also is really, but that is today more research, really understanding better the grid and helping the operator optimise and react. We are doing things already. If I look for example, at our software to manage the distribution grid, who are helping the operators with alarm management, how to use, how to react to situation, providing them support and advice. At the end. Decision is always done by a human. Nobody trusts AI enough to autonomously manage a grid, but there's a lot of research to continue to improve how we can optimise and, stabilise the grid with AI. But in short, for the people listening to us, two things to remember. We always need to align production and demand. And AI can help a lot reducing demand and averaging demand and improving the quality of the demand. And on the other side, with more renewable, the grid is more difficult to operate and AI can help the operators make the right decision, and understand better the alarms, et cetera, through reasoning, tools and agent that help them act on the grid.

Tom Raftery: 31:13

Okay, so would you say that we're heading towards a world where AI becomes essential infrastructure for grid stability?

Philippe Rambach: 31:20

Yes, definitely. I don't see how we can manage the complexity of the grids to come tomorrow without AI. But I see also a lot of hopes and opportunities because again, we are losing on one side the natural mechanical amortisation of instability by moving from traditional rotating machine to be a bit technical to renewables. But on other side, we add to the grid all these electrical vehicles. And if one day we go to the dream of all electrician, which is being able to use a bit of the energy stored in all these cars to stabilise the grid, that opens the whole world of possibilities. What we talk of millions of cars, tens of million of cars. So there is no other way than to do it with some artificial intelligence and an automatic system. You will not have an operator deciding for each of the 10 million cars when to power them, when to use them as a solution to, help the grid. But that could be a very strong change in the future. The amount of electrical vehicles is both a difficulty, but also a solution to help stabilise the grid.

Tom Raftery: 32:21

Yeah. Yeah. And if you were advising European policy makers tomorrow, what's one thing that you could tell them about AI and energy systems that you suspect they're unaware of right now?

Philippe Rambach: 32:34

I would not even dare to think they're not aware of of everything. But if I share my, my opinion on Europe I would say first year, find ways to be a bit faster in how we create data centres. We have, in Europe, a very good grid compared to other countries. We have a grade with quite largely, decarbonated electricity, relatively cheap, not too much oil price and gas price dependent a bit, but not too much. So there is a big opportunity here. So there will bring up first advice if I were to give advice, but let's not meet the opportunity for Europe by making too long, too much right type to be able to build some of them. And then my second thing would be, and it's a bit further away from your question, but when I look at Europe and when I look at AI, in AI you have different things. You have the chips, you have the data centres, but you also have on what we work here in Schneider, but other companies do what I call, applied AI. Using AI in the real world, in the physical world, in our game to reduce energy costs and decarbonise energy and improve this automation. But in, in, in Europe, we're lucky enough to have a lot of domain knowledge. We have large energy companies like us. Like industry company like us and Siemens, we have large water company Veolia. We have large service company. We have plenty of machine manufacturers with incredible process knowledge. So all of that to say that in Europe we are sitting on an immense wealth of domain knowledge. We are very well positioned to build the AI in these domains and we should do it. And here we can take a leading position as Europe on being the leading continent on applied AI because we have the engineers and we have this strong domain knowledge through our machine makers, our service providers, our water utilities, et cetera.

Tom Raftery: 34:36

Okay, so if a company wants to start using AI responsibly tomorrow, where should they begin?

Philippe Rambach: 34:43

First, they should start immediately to think responsible AI. Don't start AI without thinking responsible AI. The first thing for me is you need to define what means responsible. Responsible is not only comply with legal, and if I take us for example, we decided that because we wanted to make sure we comply with our commitment. We have published those commitments on our website. So you can go on a website and you have a one pager saying what we'll do with AI, what we will not do with AI. And every company should start with that. You need to give your employees to your customer a very strong and clear and published guideline of this is what we think is right, this is what we think is wrong as a company. And again, that has to go beyond legal. Anyway, you have to comply with legal, so, it's not even a question. Nobody wants to go to jail. But yeah, so defining what is, I would say, what is your AI ethos? And dare to publish it, at least you are buying to it. And then you need to, you really need to put in place the processes, the procedures to make sure you comply with that. So if I take us, we have this one pager on the web, then we have policies, then we have procedures, we have a, we're a large company. And I would encourage people who start on AI to not just start by, Hey, what can we do with that? But also, again, how are we gonna do that? And it can be very different. For example, one of our principles in Schneider is, let's try to avoid any facial recognition, people recognition. I was speaking with some other companies, say, Hey, you're lucky in our domain of business, we cannot avoid it. Okay. But for example, in Schneider we have, we wanted to have a way to know how many people were in a room to then compute energy and so on. We decided to not do it with visual cameras, but to do it with infrared cameras because it's much more difficult identify, who's who from a temperature pattern than from a video. So that's the kind of things, that guide your thinking. It was more difficult to do it with temperature, but we really wanted to make sure on privacy and all of that.

Tom Raftery: 36:49

Very good, very good. Time now, Philippe, for the lightning round, I'm gonna throw some quick questions at you and one sentence answers if possible.

Philippe Rambach: 36:59

Okay, let's try,

Tom Raftery: 37:01

Okay, here we go. What's the bigger climate win? Smarter grids or smarter buildings?

Philippe Rambach: 37:08

Smarter buildings.

Tom Raftery: 37:10

Smarter billings. Okay. What's the most over-hyped AI use case today?

Philippe Rambach: 37:16

The fact that AI thinks, yes, AI can help us a lot, but it'll not replace human judgement . Human intelligence. So people who talk about AGI everywhere they are overhyping it. We are not even far from there. The technology of today are not going there. It can help us a lot, but we need our own thinking.

Tom Raftery: 37:38

Very good. Okay, and what's the most underrated AI use case today?

Philippe Rambach: 37:43

Energy savings, energy impact.

Tom Raftery: 37:47

That was a softball. Okay. would would you rather cut peak demand or build more generation?

Philippe Rambach: 37:54

Cut, peak demand.

Tom Raftery: 37:55

Hmm. That was an obvious one as well. And one industry still ignoring AI risk.

Philippe Rambach: 38:02

I don't know. I don't think any, I think everybody's up to it now.

Tom Raftery: 38:05

I would say the AI industry, but.

Philippe Rambach: 38:08

I like that. I should have said that. I, I'll keep it for my next interview.

Tom Raftery: 38:14

And the last one from this one, AI myth that you'd ban permanently?

Philippe Rambach: 38:18

AGI.

Tom Raftery: 38:20

Very good. Alright, and now a left field question for you. If you could have any person or character, alive or dead, real or fictional as a champion for using responsible AI in grid management and energy savings, who would it be and why?

Philippe Rambach: 38:41

I will give a name that maybe your people don't know. I will say George Bernanos who is a French thinker from the forties and the fifties. He was very, very scared of what he was calling the civilization of machines. And his big scare was, we will stop thinking because machine will use all of our time. And I would love him to drive transformation because I think obviously that AI will help us tremendously. But the biggest risk is to believe that it thinks for us. So we need to continue to think, we need to continue to have a critical thinking. And I would've loved George Bernanos because he was so scared and so convinced that it was important to keep this free thinking, to keep this ability of the man, what makes us human, honestly, our ability to freely think. That's why probably that the guy I would've picked, which is maybe a bit weird, but he wrote a lot of novels and he was very afraid of the decision of machines. And I think in reality would've loved it. But he would've fought to have a civilization of machines in which human remains human and keeps his ability for critical thinking and free thinking.

Tom Raftery: 39:50

Very good. Okay, we're coming towards the end of the podcast now, Philippe. Is there any question that I didn't ask that you wish I had or any aspect of this we haven't touched on that you think it's important for people to think about?

Philippe Rambach: 40:03

I think you did a good job, Tom. I have to say, but maybe I will rebound on one thing. The one thing I would always say about AI, true for every industry is don't get excited by technology. Always stay cool. What do I need? What do my customer need? What my company needs? What do my employee needs? And then how can technology help? And yes, technology have done tremendous progress, and we can do things that we're not even thinking about six months ago or one year ago. But don't start from tech. Start from the business value, the business needs, the customer needs. That way you have your North Star, you have your direction, you know where to go, and you go safely to the right place,

Tom Raftery: 40:39

Fantastic. Okay. And Philippe, if people would like to know more about yourself or any of the things we talked about on the podcast today, where would you have me direct them?

Philippe Rambach: 40:48

Probably two things. Let's start by the easiest one. LinkedIn page of course. But we also have a nice podcast Tom. Not as good as yours, not as competing as yours, but a nice podcast about AI and a lot talking about AI and energy. So we'd encourage LinkedIn and this podcast.

Tom Raftery: 41:04

Okay. If you send me those links, I'll put them in the show notes so everyone has access to them. So Philippe, thanks a million for coming on the podcast today. It's been really, really interesting.

Philippe Rambach: 41:13

Thank you, Tom. It was a pleasure to be there with you.