WILL A.I. BOIL THE PLANET? — With VLAD COROAMA
PREDICTING THE SUSTAINABILITY IMPACTS OF A.I.
October 20 2025—Will A.I. use up the worlds electricity? Or will A.I. contribute more upside by helping solve the world’s sustainability problems?
In an era of ‘exponential AI,’ those questions have huge implications for our future, and the answers are nuanced, because ICTs contribute in many complex ways to energy savings AND use.
I met with IT & sustainability guru Dr Vlad Coroama to analyze both sides of the ledger.
Vlad Coroama heads the Roegen Centre for Sustainability (RC4S) where he assists governments in using digital technologies to save energy and emissions, and in reducing the direct footprint of digitalisation. Much of his work focuses on the rapidly increasing energy/carbon footprint of AI and data centers. He has researched the complex relationships between computing and sustainability for 20 years in Switzerland (ETH Zurich, Empa), Brazil (University of São Paulo), Portugal (Instituto Superior Técnico, Coimbra University), Sweden (KTH Stockholm), and Germany (TU Berlin).
After exploring direct impacts, Vlad took the discussion up a level by exploring indirect secondary effects (detrimental and beneficial) to get me thinking MUCH bigger about the dynamics shaping our future. We had a lot of coffee-fuelled fun going back and forth over his ideas! And of course Vlad drew on his meticulous research to offer some overall predictions of where we are going with AI, data centers and energy.
You will enjoy Vlad's company. He is a BIG thinker, he builds his arguments logically, based on data and careful analysis, and he has a way of cutting through complexity using simple, powerful examples. The world needs 10,000 more like him!
Click below to listen.
As always, scroll down for some further reflections from me, links to some excellent papers kindly shared by Vlad, and a full transcript.
And don’t forget you can listen to more of these insights on the future from leading researchers and future-makers by subscribing to “FutureBites With Dr Bruce McCabe” on Spotify, Apple, or on your favorite podcast platform.
PATHWAYS TO A BETTER FUTURE
I came away from this conversation thinking MUCH more expansively about AI, electricity use, and sustainability. Below are some of my key takeaways and reflections.
BENEFICIAL IMPACTS
On the plus side of the ledger, we started with smart grids and the ‘Internet of electricity.’ Every step towards more accurate forecasting of supply and demand, and balancing of production to consumption, yields significant efficiency dividends and allows grids to operate on much higher proportions of renewables. As I’ve covered elsewhere, AI is making huge strides in this area.
Vlad then raised the energy dividends to be realised in the share economy, and the potential to reduce the manufacturing footprint and thus energy footprint of assets being shared. Transportation was a big example we discussed, with dividends in more efficient journeys and in reduced numbers of assets to deliver the same passenger service levels, and similar dividends for freight journeys. Vlad pointed out that giving up private vehicles may require powerful policy levers to help things along!
‘Dematerialization’ got a mention (energy savings from sending photons instead of atoms, ala Zoom) and then we got into the role ICTs in encouraging persistent energy-saving human behaviours, our principal examples being the shower monitoring sensor Vlad researched, and the electricity monitoring apps deployed to support household solar installations. Vlad pointed to climate modeling (to which I will add attribution modeling for extreme weather events, which I discussed in detail with Prof. Andreas Prein recently) as a growing lever for energy-saving human changes at the societal level.
DETRIMENTAL IMPACTS
Direct demand on electricity by AI operations in data centers was the starting point. To this Vlad unexpectedly added the electricity involved in the production of highly refined chips as a potentially very significant unknown.
Then it got really interesting.
Vlad introduced direct rebound effects, ala Jevon’s Paradox, where energy efficiency improvements in new generations of AI chips do NOT necessarily lead to lower overall energy consumption because more “cheaper and easier” often has the opposite effect by accelerating and broadening usage. Uh, oh …
After that, indirect rebound effects, whereby the AI-driven process/work/industry efficiencies free time and resources to be allocated to OTHER activities, each of which has its own energy costs. Vlad gave the ICT-driven examples of e-commerce or online shopping:- less direct energy use, but MORE impact by increasing total consumption and encouraging practices such as ordering 12 items and sending back 11! Vlad also used the example of autonomous vehicles driving laps around the block while waiting for us to finish something.
HOW MUCH ELECTRICITY?
Vlad’s critical review of over 100 estimates of data center energy consumption found:
Estimates for 2020 ranged from 200 to 1,500 terawatt hours per year globally, a discrepancy factor of seven and a half.
Near-term forecasts through to 2035 varied from 200 to 8000 terawatt hours per year, a discrepancy factor of 40!
The methodologies behind the most pessimistic forecasts were worst.
Vlad’s own estimate for 2025 data center electricity consumption, arrived at using three different methods (global numbers, aggregation of country and continent-wide numbers, aggregation of industry numbers) is 400-500 terawatt hours.
Vlad’s predicted consumption for 2030 (‘with a big grain of salt because of the uncertainties’) is 700 to 900 terrawatt hours, with AI accounting for 200-400 terrawatt hours and likely surpassing 50% of data center electricity consumption shortly thereafter.
BOIL THE PLANET?
Vlad’s verdict in a nutshell: “AI will not boil our planet. That's one of the few clear messages that I have on the direct footprint.” While emphasizing that data centres can have very negative impacts locally, he does not think AI and data centers will become a major contributor to climate change globally.
He points out that even a pessimistic estimate of 1,000 terawatt hours per year by 2030 would represent 3% of worldwide electricity consumption, and 0.5% of primary energy consumption (accounting for all energy at source, including burning oils and fuels in transportation, all the energy lost in generating electricity, etc). Based on those percentages, he believes the doomsayers are way too pessimistic.
DIALLING BACK MY PESSIMISM
My position up until now:
Software methods such as A.I. plasticity, and hardware methods such as more bio-analogous chip designs, will eventually beat down the A.I. electricity “wall,” (good news)
The timetable for those advances is unknown and for the next ten years at least we are in for a very tough time because of exponentially increasing data center electricity demand (bad news)
That demand will produce significant pressures on local grids and electricity pricing (not to mention pressures on water) which will trigger citizen backlash and become increasingly political.
Data center electricity use will likely exceed 6% of worldwide electricity production by 2030.
After talking to Vlad, I’m dialling back my pessimism somewhat and replacing 6% with “less than 3% global electricity by 2030.”
But only somewhat.
Having seen some of the stunning A.I. developments in the pipeline, I predict A.I. usage will keep climbing steeply well beyond 2030. Think: continuous use. Think: application to every function of every job in every industry. Think: millions of A.I.s delegating and querying other A.I.s autonomously in the background on behalf of businesses, individuals, and machines.
And now Vlad’s got me thinking about all those indirect rebound effects. If A.I. is the biggest driver of productivity in our time, then what will be the cumulative sustainability impact of all that unleashed productivity on global production and consumption and resource use? In my view, it will be very big indeed.
Thank you Vlad, for a brilliantly thought-provoking, evidence-driven conversation.
MORE TO EXPLORE
Vlad kindly shared links to the following papers relevant to our discussion. I highly recommend them for further reading.
Friedemann’s and Vlad’s paper “Digital Rebound – Why Digitalisation will not Redeem us our Environmental Sins” – section II (and then IV again) discusses autonomous vehicles and how they will substitute public transportation and induce more GHGs; Section III is an easy readable overview of various types of rebound effects. Important rebound literature is referenced.
IEA report AI and energy from April 2025.
George Kamiya’s and Vlad’s critical review of DC energy use.
Gauthier’s preprint on the purity of materials needed by the semiconductor industry.
INTERVIEW TRANSCRIPT
Please note, my transcripts are AI-generated and lightly edited for clarity and will contain minor errors. The true record of the interview is always the audio version.
Bruce McCabe: Welcome to FutureBites, where we explore pathways to a better future. I'm your host, Bruce McCabe, the Global Futurist, and my guest today is Vlad Coroama. Welcome to the podcast, Vlad.
Vlad Coroama: Hi, Bruce. Thanks for having me.
BM: Thanks for making time. And here we are in beautiful Zurich and we're in a lovely cafe having coffee, which is just perfect for good conversation. I'd like to briefly introduce you and just start with your background because I wanted to talk to you so much when I was coming to Zurich because you specialize in computing and sustainability. And it's obviously been a passion of yours for a very long time. Looking at your bio, I think it was about 20 years research on that subject area and in all sorts of countries. I've got Brazil, Portugal, Sweden, Germany, and of course, here in Switzerland. You've really been involved with that for a very long time … A passion from the beginning, or what led you into this space?
VC: It became at some point a passion. And that point, and I'm almost ashamed to admit, it's when I saw Al Gore's movie. I didn't know until 2006 about climate change. I had almost finished my PhD at one of the foremost European universities in a technical field, computing, and I didn't know about climate change. Or perhaps I don't remember exactly. Perhaps I had marginally, peripherally heard about it, but I wasn't truly aware of it. And for me, that was alive, but I was always interested in computing and society. So early within my PhD, so 20 something years ago, 24, 25, it was in the field of ubiquitous computing, which today people would call IoT, Internet of Things. It's basically the same concept. And I early became interested in the societal and economic effects of it. So my PhD were two main projects. One was with the Swiss Association of the Blind, trying to see how we can deploy computing to give blind and visually impaired more power to understand the world better …
BM: Brilliant.
VC: … And the other one was with paper use insurances, which is a concept that today is deployed. So for example... And one of the first prototypes that existed was done by me together with some students and my PhD advisor, just looking what this might mean breaking information asymmetry. So it was a lot about societal and economic effects. And then I saw Al Gore and then all of a sudden, it all became sort of environmental for me. I'm still interested in sustainability generally, but clearly the environmental part is the most interesting. And that's what I've been doing since 2006, since the movie came out.
BM: It was such a powerful movie, wasn't it? It's such a well put together movie, such a powerful message. And unfortunately, that was the politicization moment, wasn't it? In the US, that was when the other side of politics decided to become climate deniers, because it was Al Gore. I mean, it's crazy, but very powerful. I remember it really.
VC: But this is, for me, a reason of optimism. Because if I think that... I mean, now it seems we've always known this, but I personally have known this less than 20 years. So I presume that most of humankind understood about climate change. I mean, even scientifically, we understood it perhaps in the '50s, the first time that someone right out of Russia looking at the ice core things. But it really became mainstream over the past two to three decades. And for that, we have actually at least in the internalizing it, we have come a long way. Now we also need to do something about it.
BM: Yeah, yeah. So it gives you hope because in 30 years, there's been so much change in understanding.
VC: Actually.
BM: So you know what? Maybe we're all moving in the right direction.
VC: Would you agree or?
BM: Yeah, when I think about it. I mean, when I dig into the detail, I get pessimistic about how long it's taking. I get pessimistic, because there's so little time to make so much change. We have all the tools, all the technologies, all the information we need, but we need to act a little faster. So I guess that's where I get my pessimism hat on. But when I look at the big sweep, I agree, you can't ignore it. And everywhere in the world I travel, everywhere, you can see with your own eyes, the climate change, you can see it. When you go to the Maldives, you see the sea level rise, you can talk to people about it. You go to different parts of the world, you go to Japan, or here in Europe, and you look at the snow, and the patterns of snow each year, you can see it, and then the storm frequency. So yes, I guess we'll all know about it, no matter what.
VC: I've seen a video of you done in the Maldives, by the way, very nice one.
BM: You saw that one, standing in the water and talking about the future sea level rise? Thank you for watching it. So now, nowadays, you have a new center or recently you established the Roegen Center for Sustainability, which you abbreviate to RC4S. And I understand you are now doing work supporting a lot of the government agencies and not-for-profits or NGOs in Europe and elsewhere on this topic.
VC: Yes. NGOs, it's not been so far. It's been government and inter-government agencies.
BM: Okay. Can you give us some examples of the types of work you're doing?
VC: So, it's always or it is... At some point, I started in the field by looking at a direct impact, so the footprint of ICT. And I think we're going to discuss about it. But then, at some point, it became clear that the indirect impact, because digitalization, generally AI now, of course, a lot, but digitalization generally, it's such a wide technology that affects basically all walks of our societies and economies. It's a very powerful lever for bringing change in other domains, and in many of them. And these changes are economic, are societal, but are also environmental. And so, this very powerful lever, at some point, you start understanding that these indirect effects, as they are called or often called, are also very important. So, for me, it's looking at both direct and indirect effects.
BM: Yep. And good and bad.
VC: And the indirect good and bad, yes. Unfortunately, it's also bad, not all good.
BM: I know. AI excites me so much for the good. In so many domains, it's just so revolutionary and so positive. And yet, we have to also weigh up the bad. So where should we start? I'm very interested in all sorts of aspects of this question, and I know that people listening to this are very interested in just sort of the future impacts of ICTs on energy. Now, there's a bunch of positive impacts. Maybe we should start with some of those. I just think of smarter grids. There is endless opportunity to make grids smarter, more localized decision-making at the device level, at the household level, at the factory level, which is more simpatico with the grid and the supply and demand fluctuations. Is that the main area that we can sort of benefit, or are you seeing more than that? Where are the big benefits to come?
VC: It is certainly one of the very important areas. It's also certainly not the only one, whether it's the main one or not, I'm not sure, but I think it's also not important. It's certainly one of those very important ones. So we might start there, yes. So, the grid, I mean, you already mentioned a couple of things. One of the main advantages of a smart grid is that it can coordinate production and demand in several ways. We can use algorithms, and AI is getting better and better, to forecast both production and demand. Production, obviously, by weather forecasts and so on. Demand also, because you can predict depending on the season, on the weather, on the time of day, on whether there's a World Cup going on and everyone. It's so many attributes, actually, that have an influence on the demand whether it's holiday season or not, and so on and so on. And these forecasts are getting better and better. And the rule is there is a disbalance between production and demand. And in the traditional grid, it has always been production follows demand.
I think people do not appreciate, as we tend to not appreciate the amazing luxury that we have in our modern lives. One of this is that I can simply get home and switch the light on. And I don't need to think, where will that energy come from? Is there enough? Or I just turn it on and the dishwasher as well and whatever. I can simply charge my electric car. And for this, it's a very fine-tuned balance that has been going on. And those engineers who have built up the grids over decades, they should somehow deserve more recognition, in my opinion, from society. It's an amazing luxury you have, and I think most of us are not aware of it. But now, obviously, this works while you have this one-directional flow from relatively few big producers to a lot of small consumers. And then you just modulate your production within the power plants, but then also switching on or off new power plants as needed and so on. And now, if more and more of the production becomes volatile and depending on the weather, of course, this paradigm is less and less applicable. So we need to start looking at the demand side as well and to do demand-side management. And for this, this is one of the main reasons. The smart grid is not an aim in itself. Smart anything is a buzzword. Anything is smart today.
BM: Of course. Yeah, of course.
VC: Our phones are smart, and they are. And so on. But the grid is perhaps... It's a necessity to make the grid smart if we want to decarbonize it or to be able at least to have the option of decarbonizing it by using as many renewables as possible.
BM: Yeah. Because things are fluctuating and all the rest of it. It's a little complex management problem of switching things around at the right time and all of that. I don't know whether... I think of that and I also think of interconnections. The bigger we can make our interconnections between grids. And I look at these big interconnectors that are crossing the Atlantic now between Canada and Europe to ship DC electricity. As you get geographic spread, you get also more options and the reliability goes up, and you can ship electricity one way or the other. And the same thing, for example, in Australia or Tasmania, shipping electricity to the mainland through a DC cable, and back it can go the other way if they need it. So, smarter but also more interconnected. It's like a big internet for the planet.
VC: Interesting. I thought it's a mostly one-way thing from Australia to Tasmania also. I always thought, but no. It's...
BM: I think it's both. Yeah. And in Europe, there's like 20 of these interconnectors going, especially from the Scandinavian countries down. But yeah, it just seems to me that it's a little bit like the internet. We're heading slowly towards a system where we can switch packets around, like information packets on the internet, but switch electrons around and move them around in a similar way to the internet, where there are optional pathways. And the bigger it is, the more nodes it has, the more resilient it gets. That's kind of how I imagine it. I don't know if you're the same.
VC: No. I would certainly agree. We switch the magnetic field around and it carries. And we see when these connections do not exist. I don't know if you are aware of the blackout we had in the Iberian Peninsula this spring in May. I think it was right...
BM: Yes. I'm very aware, yeah.
VC: Okay. And that's part of the reason, and maybe we get to speak about it, there is also a NIMBY effect there going on and not in my backyard, because it's very difficult to build new power grids, especially across the Pyrenees. So Spain had very, if I'm not much mistaken, only a 2.5 gigawatt connection to France and all the rest to Portugal. And I mean, Portugal is small and can only absorb so much. So that's why they took Portugal down as well. And now, because it was impossible for decades to build across the Pyrenees, now they are actually building in the Gulf of Biscay, so a submarine, a five gigawatt connection. If that had been up already, probably that France and continental Europe could have stabilized Spain.
BM: Yeah. Yeah. Yeah. And it was huge. That outage lasted for a long time.
VC: I think 36 hours or something.
BM: Luckily, it was Spain because, I don't know, everyone has a nice attitude down there and it didn't lead to total chaos for some reason. So you said there's other opportunities outside of grids?
VC: Yes.
BM: Can we get into those a little bit?
VC: Sure. And let me first stick with the coordinating supply and demand paradigm, because this obviously applies to the smart grids, but it can also apply in other domains. For example, when you think of sharing economy. We power tools. So there are these sort of artifacts that we use seldomly that also tend to be large and expensive. So it's also financial incentive there.
BM: And an energy incentive goes with that.
VC: Of course, especially production.
BM: Let's talk about this.
VC: I mean, let's talk about the production that gets into them. So if you could produce less of them and have sharing platforms and coordinate, it's also supply and demand coordination. Or for used goods, refurbished goods and so on. So I see lots of potential there that it's not as much used so far. Of course, human factors and cheating and wanting to own objects. So a lot of human desires and limitations need to be taken into account as well. But I think that potential, it is still a fairly large potential that needs still to be tapped.
BM: I've got an example of that, which might deepen this conversation. And I got very excited when I first saw it. So, the head of computer science and AI at MIT is a lady by the name of Daniela Rus. One of the absolute smartest people I've ever met in my life. And they did a study, she co-wrote the paper, where they modeled with real data for taxis in Manhattan. Let's consider taxis in Manhattan, taxi cabs, an asset, which uses a lot of energy. The purpose of this was not necessarily energy, but you can see the energy implications. They looked at how many taxi cabs there are and what the service levels were with real trips, pickup times, wait times. When you want a yellow cab, you put your hand up. And they modeled that for the fleet. Now, the fleet in Manhattan at the time she did that model, which was a few years ago now, 14,000 taxi cabs.
VC: 14,000?
BM: I know. I know. And of course, a lot of them are waiting in the wrong places and this sort of thing. So she then did another model where she just used a little bit of machine learning and said, “how many taxi cabs would it take to achieve the same service levels if they were smart about where they waited and distributed themselves, knowing the patterns of when people want rides, what time of day, night time, end of work hours, and it wasn't driven just by cab drivers hoping for business?” You don't have to take the cab driver out, they could just follow, if they were just following AI, telling them where to go. You know how many taxi cabs it would take for the same service levels as today? 3,000.
VC: Okay.
BM: So what that tells me, to your point, and I wasn't even thinking about it before this conversation, but it connects with it. I always tell people, well, that's 80% fewer assets required. Imagine the capital savings, but what about the energy savings? You're talking about moving around only 3,000 vehicles versus 14,000. And if you take that at a city scale in a Singapore or London or here in Zurich, although Zurich's very efficient with its transportation, That's an energy saving. For electric vehicles, that's how AI applied with electric vehicles would drop a huge amount of energy usage in another domain. Because that share economy, that's almost that share economy, just a different way of looking at it.
VC: And of course you could... I mean, that's very interesting and I didn't know the numbers. Thanks for sharing them. Of course, you could take this also then to private car and question whether we need private cars. But then it gets very tricky, unfortunately, because we like to own cars.
BM: Sure.
VC: And in most of the world, and more and more parts of the world, we can afford them as well.
BM: Yeah.
VC: And we like the luxury and knowing it smells like us and it doesn't have any remote viruses and so on. So unfortunately, we will come later to that, I guess. But one needs to think when making, I mean, when it comes to taxis, I guess all you need is policy...
BM: Sure. Or even public assets, buses. Yeah.
VC: Or public, yeah. But some assets, and I guess a power drill is also something I don't like to, or I don't need to, perhaps I don't know, much enough, but I don't feel the need to own one. But a car, I think for many people, is different. It's a bit of also sentimental relation.
BM: But I wonder if…
VC: And we need to take these sort of psychological and societal effects into account as well when we plan for this, to avoid being overly optimistic about the potential.
BM: Yeah. I definitely take your point. Yeah. Definitely. We can't sit there and go, we can do this tomorrow, because socially, not going to be acceptable. But I can see enormous potential for energy saving, I guess, in making other systems more efficient. That's the point. Other energy-intensive systems, besides just electricity production and consumption, if we make manufacturing more efficient using AI, or shipping, or whatever, it all adds up, doesn't it? Yeah. Is there anything else we've missed, or is on your list of things that are positive?
VC: On the sharing part, I will perhaps go to one, two more such paradigms or mechanisms that might lead to energy savings and greenhouse gas savings as a consequence. One of them is, you said already, efficiency gains. Think of planning a fleet deployment. Not necessarily of taxis, but of one, like it's someone who transports goods. And to plan those routes and so on. Obviously, there is already many algorithms going on, but they are getting better with AI now.
BM: For Freight, for example.
VC: For Freight, yeah. Particularly for Freight. Planning new public transportation by having smartphone sensor data, knowing where the need is, and then putting those public transportation lines where people need them. So, in terms of optimizing any sorts of production or transportation or whatever, that's where digitalization is obviously very good and where it has direct impacts on energy. Another one is dematerialization. Sending bits or photons around the world is much cheaper than sending atoms around the world. A photon doesn't weigh anything. So, anything where we can substitute photons for something else, it's usually a good substitution, environmentally speaking, whether it is streaming video instead of printing DVDs or audio for CDs, whether it is sending our images around the world instead of us traveling and so on. Of course, this bears a lot of potential for rebound effects, and we'll come to speak about that later, and that's crucial to also consider that.
BM: Rebound effects. I'm going to make a note. I'll have to ask you. Okay.
VC: And perhaps the final such domain that I would like to mention is, they are related, it's two. Perhaps one is behavioral changes. Sometimes this gets mixed with nudging. Nudging is slightly paternalistic. I think that has its ethical issues. But certainly, without nudging anyone, just digitalization gives us through the sensors, the wireless communication, and the algorithms, and the devices that can present them to us, be they smartphones or whatever, HCI, Human-Computer Interaction paradigms, it can show us the consequences of our behaviors, reveal them. And then if we want, without any nudging, we can then take pro-environmental action just by becoming aware.
BM: Just through transparency.
VC: Exactly. Do I have time for one example?
BM: Of course.
VC: Okay. So, a couple of years ago already, pre-pandemics, so 2017-'18, colleagues of mine from, I was at ETH Zurich back then, and colleagues of mine developed a very nice, it's a shower sensor that you plug it, like you have your shower head that comes out of, and the water flows. And then you plug it just before your shower head. And the device measures a couple of attributes, the water flow, so it can show you how many liters of water you have used while showering, which temperature, knows how cold the cold water is, so can also compute the energy to heat that up. And it also has, because it has a small display, it also has a nice animation with a polar bear on an ice cap, and the longer you shower, the ice cap melts more and more and more, and then in the end, the poor polar bear drowns. So that's the nudging part. And just by putting this into people's homes and showing them this, 20% to 25% reduction in shower, not by showering colder but by showering shorter, and especially turning off the water while putting soap on.
BM: Brilliant.
VC: And consistent. So studies done over two months, over six, because always there is the bias of early phenomenon of volunteers that already want to save. But when you do it for six months, sort of the novelty wears off. And that's why sometimes you have to do studies for a long time to see whether it's a persistent effect. And that was a persistent effect.
BM: Okay. Fascinating. Please don't tell my wife about that device, because my one luxury in life is those long hot showers. Because I'm, well, it's all solar, so it's okay. But it's... Anyway, it's funny. That's hilarious. Actually, on this visibility thing, in Australia, penetration of household solar is extremely high. It's a sunny country and the soft costs are low. So installation costs, the human costs of getting it installed are very low. There's a very little regulatory barrier unlike the US, where it's got a long way to go. But one of the phenomena I observed is, most of the solar companies will give you an app on your phone so you can watch how much energy is coming off your roof. And it graphs it over the day, and it graphs your demand when you switch on appliances. So you can see it real time or look at the daily graph. And what happens is people start moving their appliances under the curve.
VC: Okay.
BM: Yes. And it really has an absolute change. It's changed my life. It's changed hundreds and hundreds of people that I know just for having the app. Because they go, oh, look, I've been running my dishwasher in the evening when there's no free energy off my rooftop. Well, I can see the spike, but I can move that spike. I can just run it in the morning when the sun comes up. It's fine. And they do. And they run their pool filters under the bell curve for the day's solar. And it's changing behaviors exactly like you say.
VC: So behavioral demand-side management, basically, not automatic demand-side management.
BM: Correct. Yes. All human.
VC: All human. Yeah.
BM: All human. You can automate it, I guess, and put things on timers. But I'm watching the human behavior. People get excited. "Look, look, I've moved this under the curve." And part of it is the glee at not paying an energy company a cent. So it's not just environmental. It's partly just the joy of this is free but it's also environmental. It's a mixture of things. Very interesting. Visibility and transparency. I love that. We can go so much further with that, with your shower device and this rooftop solar app. We can do a thousand of those in workplaces and in homes.
VC: Yeah. I love your example. And the very last point, taking this sort of visibility, what we talk now is sort of individual behavioral change. But at a societal level, this revealing, showing us what happens is basically simulation and modeling. This is where computing comes from. I don't know if you know why the first computers appeared. People say military, but that's wrong. And that would have been simulation as well. No, it was weather modeling after World War II. So, it's a system of seven equations that are still used today that are known for, I think it was a Swedish or Norwegian meteorologist who put them up in 1900-something. So beginning of the 20th century. And so we knew how to compute the weather, but it's a system of seven equations with a nonlinear system with seven variables. And we just didn't have the human power. You would need 10,000 human computers just to keep up with the weather. So once electronic computers came up, then we could finally be ahead of the weather and make predictions. So simulation is the very first application of computing. That's where it comes from. And now we can obviously compute the climate and so on and understand what will happen. Unfortunately, again, human factors, sometimes we start not believing, having conspiracy theories and so on.
BM: But we'll get there.
VC: But at least we have the data. And it gets better and better. And this is crucial, of course, for decision-making on a societal level.
BM: And it's getting better and better in quality. Yeah. Fantastic. I love that. And you know what? Yesterday we went to see Professor Andreas Prein in the atmospheric sciences area here in Zurich. So, we were talking exactly how models are getting better. So it links in. It's really nice. It's really nice.
So now let's turn to maybe the negative side of things. The energy footprint has been increasing dramatically, particularly in data centers. We can't measure exactly AI usage. We know it's high. Most of it is hidden behind the firewall for what ChatGPT or Claude or Gemini or these things are actually taking per query. And we know it's high individually. But what we can measure is data center electricity usage. And it's obviously, it's almost exponentially increasing right now. It's gone from 2% of the U.S grid to 4%. U.S particularly. So it is a U.S-centric issue. But that one's quite scary because I'm looking at AI and the exponential increase in usage. People are moving to using it all day. They're using AI continuously and the power requirements are very high. And that one worries me. And I guess the other one has been crypto as a big footprint thing.
BM: And I just wonder whether you're seeing the same thing or would you add to that list, or what are your thoughts on sort of the big drivers of ICT as a negative on energy and carbon?
VC: Before we dive into that, let me just remind that there is two sorts of negative consequences. I prefer the term detrimental because mathematically... And I will come back to this in a second. So we talked about the benefits, and the detrimental effects are this what we'll talk now about the direct footprint. But then there is also the indirect detrimental effect, the rebound effects that often are triggered by the very mechanisms of efficiency that we just discussed.
BM: Okay.
VC: So, not to forget...
BM: Okay. I'll let you drive it. I want to dig into all of that.
VC: Yeah. Let's get to the direct footprint. But then we come back to this, that plugs into what we have just discussed. So direct foot... And by the way, why I say, because when you talk positive, negative, the mathematical positive is environmentally negative. And that makes it in my opinion, a bit confusing or perhaps confusing. Because if something grows, if energy consumption grows, that's mathematically positive, but environmentally negative and vice versa.
BM: And sometimes it's seen as economically positive because it's an industry growing.
VC: As well. Yeah. So positive, negative can be interpreted mathematically or normatively. So when you talk purely normatively, detrimental, beneficial, then in my opinion it's perhaps or it's often clear when you get... It's not a big deal but yeah. So footprint, and especially AI, you said, and data centers, you're right. We do not know exactly, even the companies don't know exactly how much they have estimates, how much AI are within their own footprint. Where we are now, and I've just done with George Kamiya, who's a London-based, he worked for many years with the International Energy Agency in Paris, now he's based in London, and we have just this April or May, we have done a critical review of existing estimates of data center energy consumption, both global and continental, and we also looked at that sort of, on country level, we also looked at company data, and so on. And then obviously, just a couple of weeks later, the IEA came out with their study, AI and Energy, that probably you've seen. We actually have been in contact while developing the two. Their focus was purely on AI, we looked at data centers generally, but obviously the two are strongly related. Okay. So the numbers, and there is a huge discrepancy between numbers, especially, like even for today, today you have a discrepancy of a perhaps factor of... No, let me reformulate it, not today, 2020, you would have a discrepancy of between 200 terawatt hours per year for all data centers globally to 1,200 to 1,500 terawatt hours. So a factor of six to seven and a half discrepancy.
BM: You mean a discrepancy between estimates?
VC: Between estimates. Yes. Yes.
BM: Oh my God, that's huge.
VC: Yes, that's quite large. And what it means globally, we consume about, now it's 30,000 terawatt hours per year, back then it was a little less. But so this 200 would be less than 1% of global electricity, and much less of global energy, of course, because primary energy consumption is mainly not electricity, it's mainly fossil fuels that we burn directly into our furnaces, into our cars, the engines of our cars.
BM: Yeah. And then we lose most of the energy in the combustion inefficiency.
VC: Yeah. Exactly. So we are at almost 200,000 terawatt hours per year primary energy consumption, so 200 is really not much. And even 1,000 wouldn't be that much, it would be more in terms of electricity, of course, 1,000 of, you know. It's a difference between below 1% and 3%.
BM: Yes. Just to clarify my mind, yeah, in one way I look at it as the percentage of all energy, but that includes all combustion in aircraft, shipping, automobiles, and the other way is to say what's the percentage of electricity use?
VC: Yes. Exactly.
BM: Okay.
VC: And again, electricity, about 30,000 per year, energy 180,000, so six times more, roughly. The share of electricity is growing, fortunately, because electricity is the only thing we know how to decarbonize.
BM: Yes.
VC: Not necessarily decarbonize, but it's the one thing we know to do. And of course, we have green hydrogen and some other, but those are niches, like it's mainly electricity that we know to decarbonize. So, there is really this discrepancy, and when you are looking at the forecasts, and relatively near-term forecasts, 2030, 2035, you would have a discrepancy factor of 40. So some would say it will stay at 200 terawatts, others would say 8,000. 8,000 is a third.
BM: Wild difference.
VC: Yeah, it's huge difference.
BM: So everyone's estimates are just all over the place.
VC: Yeah. And that's because there are no established methodologies, or until very recently, there were none. Because the primary data is very insecure, because it's such a dynamic field, everything develops. Because people from different backgrounds with different methods come up with estimates. Some look at the life cycle, like include production, others look only operational. Then, especially the forecasts are so dependent on the assumptions you make and especially people. So what we found out is that, and we looked into over 100 such estimates, and we found out... And we subjectively, we try to do it objectively, but it was impossible to define objective criteria to assess the quality of those studies. So we ended up having expert opinions, considering with a lack of modesty ourselves experts, but trying also to agree on some criteria, but not very formal. We had a couple of criteria, but not a rigid grading scheme, because this has proven impossible. But we did independent assessment and then discussed where we would not agree. And so we scored the existing assessments, the quality of them from a very low to very high. And it turns out that the really pessimistic estimates were the worst. So those...
BM: In quality.
VC: In quality. So especially when it comes to forecasts, because they would ignore, like some forecasts would say by 2030 for all of digitalization, not only data centers, would the consumption by 2030 would be 30,000 terawatts. That would be doubling the entire electricity consumption of the world only for... That's impossible.
BM: Absolutely.
VC: Like you cannot build the power plants, you don't build the grids and so on.
BM: Exactly. And the social pushback is so big before you get anywhere close to that. Yeah. Because electricity pricing would start to rise dramatically for everyone else. So ah yeah. Exactly. Yeah, it's crazy. I understand.
VC: So that's why any number you see, you have to inherently take with a grain of salt because there is so much uncertainty, no one can really know for sure how it will be. We are getting better at knowing how it is now. It has certainly grown. It is not those 200 terawatt hours per year anymore. And this is AI, clearly. Probably now we are, the number that George and I consider most likely, but again, you might disagree and certainly others will disagree with us, is probably around 400 to 500 terawatts per year data centers today. And the growth has been mainly AI from, it was not 200. It was probably 350, 2020. So it was closer to the optimistic numbers, but not at them.
BM: And we know it's AI because it correlates with the timing of neural networks and machine learning, don't we? We can see the spiking around 2015... You see the graph start to grow from there. So it correlates very well with the timing of...
VC: And you have no numbers. You know roughly how much chips NVIDIA has sold and you can make assumptions that if they function 24/7 and they sort of do, because I mean, everyone is looking for them. You can estimate, Google doesn't publish how many TPUs, they build their own hardware for that, but you can estimate that AMD, Intel. So you know sort of the big players, what they do. And then you can scale this up and come up with a rough estimate, how much those produce chips, assuming all of them are still working and they sort of are, at least the post-2018 one since NVIDIA Volta came out. So since they started having the Tensor cores on them. Yeah. And then you can sort of know where we are today.
BM: What's your feeling then on where we might be in say 2030 or 2035? I know it gets more and more uncertain, but what's your gut feel with all of the things you look at as to what that number might look like?
VC: With the big grain of salt of predictions are difficult, particularly about the future.
BM: [laughter] That's my job. Don't worry.
VC: Right!
BM: And of course, that's exactly the truth, but it's at least an informed opinion. And that's the thing about predictions, that I'm always seeking out people like yourself, because of course, we're always gonna be wrong. But if we're gonna take a scenario or think about scenarios, it's best to think about expert informed scenarios as probable futures. So that's why … don't worry, there's a big disclaimer on this. But yeah, where do you think we'll be?
VC: So I believe we will be at, 2030, let's say, AI 200 to 400 terawatt hours per year, and data centers correspondingly 700 to 900. So AI will slowly become almost 50% of data centers, and it will continue growing.
And this is where the IEA, and of course, I don't want to compare their study, obviously has so much more notoriety than ours, but it's funny that through very different means, they have done sort of a primary study, and we have just done a critical review of assessing the quality of others, and then saying, okay, those whom we assess to be best, we guess it's around there. And then we also did it in three different ways, global numbers, aggregation of local numbers, so country and continent-wide numbers, and aggregation of industry numbers. So we have sort of three ways, and they sort of coincide around these numbers. So that's what we think, but again, it might change entirely. One of the things that I'm not, and I think we need to get to discuss this, because this will still not mean, let's say, 1,000 terawatt hours per year of all data centers, probably below that. But that's still like 3% of worldwide electricity. Obviously in Ireland, much more, or Singapore, in the US also, as you said in the beginning, 4%, it will grow 5%, 6%. In Europe, it will also grow. Emirates, Saudi Arabia are also growing. So it will be a very... And that's the point. It can be and it will be local a problem, much more than... So I don't think AI and data centers generally will become very soon a major contributor to climate change. Again, 1,000 terawatt hours per year among the 200,000, that's 0.5% of our primary energy consumption.
BM: Keep things in perspective. Yeah. Because it's so easy to get very pessimistic at a local level. I talk to a lot of people that are in the data center business or building data centers or managing them, and they're all going, “oh my God, you should see our electricity usage. It's gone off the clock, and our water usage as well,” which is another issue because the cooling. Recently, I was talking to someone who's building a five gigawatt data center cluster in one location. Oh, five gigawatts in one location!
VC: With the names corresponding, names like Colossus and Titan or whatever, like they're also... Yeah, yeah.
BM: So you're quite a... There's an optimistic thread to this. You're … just keep it in perspective, I guess, is what you're saying.
VC: Well, yes and no. So yes, there is an optimistic trend that, so both among all human activities and in our personal footprint, AI, even though I'm a heavy user of AI, if I were to stop my red meat consumption, it would still be the far more effective climate thing, climate measure that I could take personally, then giving up AI or Netflix streaming or whatever. I gave up red meat almost. I'm not a vegetarian, but okay. If flying less, heating our homes, moving into smaller homes, there are so many other activities that are both personally and then aggregated, of course, on societal level, that are more impactful. Exactly.
BM: Impactful.
VC: So, AI will not boil our planet. That's my very clear, one of the few perhaps clear messages that I have on the direct footprint. This being said, however, it's not all energy and GHGs, and it's not all global, it's also local. So far we discussed only the operation, the production of the chips. It is a huge unknown. The energy consumption of that, it's not known. No one knows how much energy it really takes into, and there are some hints that it might probably not double the consumption, but still be a substantial contributor. For example, the silicium, the substrate of the wafers, that needs to be at a crazy, crazy purity. And to achieve that purity, I think if I'm not much mistaken, N11, N is the number that defines the number of nines. So 11 means 99 point, and then another nine nines. So in total, 11 nines percent purity. Because the thing is, we are now getting, and this is due to miniaturization, this entire, it's Moore's law of miniaturization, and this has been instrumental to allow AI to scale at a level it does today. So this was crucial, the miniaturization.
But to achieve it, we are now almost at the angstrom level. So we are, one transistor gate is now 12 or 15 atoms. So there if you have one atom of impurity, that has a huge effect. And you need this crazy purity, and to achieve that, that's a very energy, I'm not a specialist at all, but I just know from reading a bit secondhand, so I cannot really talk about it. And I think there are people in the industry, obviously there must exist people who know this, who build them, but the intersection of these persons and environmentally concerned people, people who do global energy modeling, I think that intersection is almost zero. So that's perhaps a test for you, bringing these people together. I'm in a project with UNEP, the United Nations Environment Program, where we think a bit about AI possibilities, and that's one of the key messages probably that this report will bring, that this needs to become clear.
BM: Fantastic. I never thought of that side of it at all. I've not been thinking about it at all. You're making me think a lot bigger about this subject already, so that's fantastic. And we have to try and get some links in the notes for this podcast for people to some of these papers, especially the one, I don't know if it's possible or it's already public, the other paper you're writing about the IEA, looking at all the different...
VC: That's not yet public, but I can show you one or two of the sources. Again, it's not many. And there's one very good guy in France, Gauthier Roussilhe is his name, who's an independent consultant and researcher. So if you will get to France, talking to him... So he was my main reference for this.
BM: Yeah. Well, after this, we'll take some notes on some names and people to go after because that's huge.
One thing I can add that gives me hope, on the technological side, I spend a lot of time trying to look at the next generation of artificial intelligence, both software and hardware. And unquestionably now we're on a journey, and we've always been on a journey, to make hardware more neural-like. I mean, that's what NVIDIA, that's why it's a $4.5 trillion company. They managed to make chips that are slightly more bio-analogous. But if we go to different institutes around the world that are getting some more neuromorphic chip design, it's spectacular, the energy efficiency improvements. I'm seeing already, and it's done in software and in hardware, it's a mixture of things, but we can do AI with the same power as the AI people are using today. For sure, we can do it with a thousand-fold improvement in electricity efficiency, and I think that's coming in next generation designs. But if we look at the theoretical limit as being biology, we're talking about a 500-million-fold improvement. That's the window. We're never gonna get there, but the more the silicon becomes a little bit more neural-like, and that means analog chips in cameras, for example, there's just completely different architectures to the Von Neumann architecture. That's exciting to me. I see hope coming.
I can't put the timing on when those chips really transform data centers, but you can see companies like NVIDIA, they're gonna be on a roadmap of getting better architectures. There's headroom to improve the electricity consumption of the architecture, and that, to me, is also good news.
VC: Yeah. And this also nicely takes us to rebound effects. I wanted to discuss rebound effects in a different context, in that of indirect effects, so effects of digitalization and AI outside the digital field. But also, of course, rebound effects exist also within digitalization. So maybe it's a good moment to discuss them now. Are you aware of rebound effects, what they are?
BM: No. You have to explain them to me.
VC: So rebound effect is... It's also known as Jevons paradox. William Stanley Jevons was a British economist, 19th century, and he wrote in 1865. I love citing in my articles an 1865 thing, you know in computing where everything is like two years old that you're citing. And his book was called "The Coal Question." And what he noticed is that steam engines were becoming ever more efficient, using coal ever more efficient, and yet the coal consumption was exploding. And this seemed... And that's why Jevons paradox, and that's what the book is about. I haven't read directly the book, but because I'm obviously not a native English speaker, 19th century English, probably even more difficult, but I read several articles that cite the book sort of substantially. And what Jevons saw was this paradox. And the explanation is fairly obvious. The more efficient your steam engine becomes, the more coal you can extract because you can use it deeper. A lot of the steam engines were actually used to get water out of the mines, and thus enabling the mining in the first place. And then you can extract water from deeper if you have a more efficient, and then you can dig deeper for coal. And all the other steam engines used for trains, whatever, industry, those would also be more efficient. So it would become more and more feasible to use them in broader and broader domains. So steam engines were becoming ever more efficient, but the coal consumption was exploding at the same time. And the word, and this is known as direct rebound effect. And we'll get to speak about indirect rebound effects also, but the direct is very important now for AI because it might be, or it is certainly now the same phenomenon, exactly the same phenomenon. AI chips more and more efficient. We can do more and more computations. It is cheaper to do those computation. We can apply them in more and more sectors. And so far we're far from reaching a limit. Probably there is somewhere a limit. But as you said, the annual growth rate of efficiency is above Moore's law for AI chips, just the hardware one. I've done this in another study, and there are Epoch AI has done... Epoch AI is around 50% per year, I am at 80%, so even more. And that's just the hardware one, and then you come with the software.
BM: 80% hardware efficiency improvements for AI per year.
VC: Per year, yeah. And then you have the algorithmic improvements. And despite this, we obviously see AI exploding. So it's exactly the coal question. Basically, it's the GPU question, if you want, of modern times. And so it's clear a direct rebound effect because it is cheaper to apply it, we do it in many more fields. At some point, of course, there is a limit to this, and the question is... But until that limit is reached, I think efficiency gains will just bring us more and more usage and will not lead to...
BM: Yeah. Yeah. I see exactly what you're talking about. It just fuels the exponential. Right. I don't know whether to be inspired or really depressed by that. Now, that was direct rebound effects.
VC: Yes.
BM: And indirect is different.
VC: Yes. It's an entire...
BM: Let's explore it.
VC: So even economists do not agree on one taxonomy of rebound effects. But generally agreed, because it comes from economists. And funny enough, let me just say this, Jevons paradox, he didn't call it Jevons paradox. It was later named after him. It lied dormant for over 100 years. So he wrote this book 1865. And in 1980, the first economist came and rediscovered the same thing. So 115 years later.
BM: Brilliant.
VC: And now it's sort of mainstreaming among economists. The indirect rebound effects are those. So what distinguishes the direct one is that one good becomes more efficient. Hence, it becomes cheaper to produce. And as we know from since Adam Smith, like classic economic theory, when the price for something decreases, demand increases. And this is basically the direct rebound effect. Indirect rebound is through, because something, either other goods will... So not the one that became cheaper. But there you save money. And that is sometimes called the income effect. So you have more disposable income and use the income in somewhere else to do something with it. And that will also be energy consuming somehow. You save money, but then you go to holidays or whatever. Or through entirely different mechanisms that are not financial. And one, perhaps the best known is time rebound. You save time, and digitalization is very good at saving us time. And you do something with that time, something else, and that again will be energy consuming.
BM: Oh God. Yeah.
VC: So basically you have a budget or transaction costs. And now we come to perhaps the last, if we still have the time.
BM: We have the time.
VC: We have the time. Okay.
BM: All good. Thank you.
VC: Thank you. To the indirect rebound, so the sort of what digitalization, unfortunately, can contribute to increase consumption energy of goods and energy consumption and greenhouse gases and all sorts of environmental impacts in other sectors.
BM: I can see it already. I can see this already. Because the more efficient we get with AI, the harder we're working. Because actually you have to use it to compete and stay level with everyone else. The more activity, this is just work before we get to anything else. And therefore the more activity is taking place, there's more GDP growth at the highest level, but we're all on the hamster wheel faster. Right?
VC: Exactly.
BM: So there's more happening, more activity in work because of AI. It's not less activity, it's more. And there's a consumption effect there. But also, even if it was less, the leisure time potentially gets reallocated to other things and other activities. So I think that's... Yeah, I can see it.
VC: And so much more. Think e-commerce or online shopping. A very simple example. It is so convenient. I can stay after dinner with a glass of wine in my hand at 11 PM on my couch. I don't need to go to a store. It is so convenient. I browse the net. I have reviews. I can compare prices. Of course, this is consumption increasing. I go to my favorite shop in Switzerland where I usually buy my electronics. They have the offer of the day. And then I don't really need it. But come on, it's like 30%. And because I know there is the offer of the day, even when I don't need anything, I sometimes log in to see what's the offer of the day today. Maybe I need it. I get targeted commercials, targeted advertisements. All of this. So online shopping and e-commerce generally increases consumption. The fact that I can online order clothes, don't get me even started with clothes, like in three sizes and four colors each. And then I send 11 of those 12 back or perhaps all 12 of them, right?
BM: Yes.
VC: And autonomous vehicles. We were talking about taxis in the beginning.
BM: Yes.
VC: Autonomous vehicles are again, so just to wrap up e-commerce. So this is about convenience, decreasing of transaction costs, the economic concept of transaction costs of the monetary and non-monetary goods to find out about the product and assess its usage, its usefulness for me. And the same happens with autonomous vehicles. Why am I taking public transportation today? Aside of perhaps environmental concerns, but from a purely hedonistic perspective, I'm taking it because I'm a procrastinator. So I need the time to finish my presentation. I don't want to waste time driving. And all the conveniences that public transportation, that I don't find a parking spot in any big European city or mostly around the world. And while I'm public, I don't need to care about it. But if the car drives me there, of course I can work. I don't need a parking lot. I let the car circle around the block empty for two hours consuming energy or just it goes empty to the border of the city and then it comes back when it knows I finished my thing. And when I get back, I can read a book, have a glass of wine, whatever. Anything I can do in public transportation except perhaps going to the bathroom.
VC: So bad bladder is a good argument for the public transportation. But other than this, all the other conveniences, talking to people, working leisure time, I can do it. So why would I expose myself to the smells, noises, viruses of public transportation? And we could go through many other examples. Take just as the last one perhaps, video conferencing. We said, of course it saves photons, sending photons around the world is so much better than atoms. But of course if you could not have sent your photons to me via email and video conference, we would not have met today.
BM: That's true. Yeah.
VC: So video conferences allow both business and private romantic relationships to endure distance. But that distance will sometimes be bridged by sending atoms and not always photons. And that's also an indirect remodify.
BM: Yeah. And just rewinding, when we talk about e-commerce, Jane and I spent some time inside an Amazon distribution center, one of the big ones, for some hours. And I've never had a more visceral and powerful demonstration of the consumer and energy effect of e-commerce than those hours. It was like watching trees go in one end and come out as cardboard packaging to be thrown away the other end. And the scale of it was so big, you couldn't even see the end of the inside of the building. It was unbelievable. And that was one of many distribution centers. But it's just your point, it just resonates with me so powerfully that the secondary effects here, downstream of convenience, it's a huge footprint on the planet, what's happening in those distribution centers. Wow.
Well, maybe we're coming to a close. We've covered a lot of ground. I'd like to finish by saying, well, is there anything else that you wish more business and government leaders knew about your work at the Roegen Center for Sustainability or your research work? And then I always ask the last question which is related to that. But maybe we can prioritize and also ask you what you think the most important thing is, if you had to pick one that you wish leaders could do to create a better future.
VC: And I will stick to this field. I think the two questions are related. So there is two prevailing narratives today. One is, let's call it the doomsdayers narrative, and that's about AI will boil the planet. And that's wrong. It has some effects that we did not discuss on local electricity grids, local water scarcity and so on. But I'm sure you know about them. So locally, it can have very negative impacts, but from a climate change perspective, AI and digitalization in general will not be the main cause. And also giving it up will not save us, unfortunately. It would be a nice way out, but it's not. So this is one narrative. And the other narrative is that, yeah, yeah, yeah, we have this footprint, but we have all these benefits that we talked about in the beginning. And that's clearly outweighs. It is the much stronger lever. And this is true. It has all this potential. But first, the footprint is real. Small or large as it may be, but it's real. The other is mainly potential. We still need to tap it. So that's one caveat of it.
And the other one is, and that's the main one, there's also this indirect negative or detrimental effects. Again, because negative mathematically is different. So the indirect detrimental effects. So we have these three main effects. We have the footprint, which is the direct one, and the indirect effects that can be beneficial or detrimental. And the second narrative is, of course, pushed a lot by the industry and also by lots of, I'm sure it's a lot of not only mercantilism, it's also idealism. People who want to believe, who say, yeah, we have the footprint, but all this potential.
BM: AI will save us.
VC: Yeah. But no, it has also all the other parts and we need to take good care about them. And again, and that's perhaps also what I think we need to do is, for example, this autonomous vehicles example is my favourite example and has been for some two or three years, because for many, it creates a high effect. Wow, I didn't think about this. And it's not something that technology will solve by itself. This we need to solve somehow through other means.
BM: This is about incentives, social incentives, it's economic, it's policy.
VC: It's policy.
BM: Yeah.
VC: And again, it also of course creates, coming back to my very beginnings when I was working with blind and visually impaired, it creates, of course, autonomous driving allows elderly to lead a more inclusive life. So perhaps for elderly, we want that to happen, but not for the business person who could take the train. So it requires a lot of socio-political discourse to see what are the technical possibilities, what might happen, what is likely to happen, and then what do we want to happen, and how can we perhaps prevent the other. So the main message would be these two levers of digitalization, its indirect effects across our entire society and economy are very true, they exist. It can be both beneficial and detrimental, societally as well as environmentally. There are some trade-offs between societal and environmental, so sometimes you have a societal benefit at an environmental cost or environmental benefit at a societal cost, for example, privacy, you're giving up some privacy for some energy-efficient coordinating things. And all this discourse is not happening, and it should be because these levers are real, and we should see how we can use that one and prevent the other as much as we can.
BM: Well, hopefully we're contributing to the discourse a little bit today. Dr. Vlad Coroama, this has been an absolutely fascinating conversation. And you have helped me think so much bigger and more expansively about this topic, so thank you very much for making time.
VC: Thank you. Thanks a lot, Bruce, for coming and giving me this chance.