The Future of Energy: Powering AI, Quantum Computing, and Green Tech
- Hello and welcome to "Agents of Tech," the show where we explore the minds of the greatest innovators addressing the world's critical science and technology challenges. I'm Stephen Horn. - And I'm Autria Godfrey. And today we are looking at energy and how we're going to need more and more of it to power machine learning and AI. (upbeat music) Welcome, everyone.
So happy to have you with us here on "Agents of Tech" today. I am coming to you from Washington, D.C., also right on the borderline with Virginia.
And my co-host Stephen is across the pond in London. Hi, Stephen. - Hi, Autria, how's it going? - It's wonderful. You know, we are talking about this on both sides of the Atlantic.
Global finance group Goldman Sachs predicts in a recent report that electricity consumption in the US is poised for a major surge for the first time in years. In fact, it's projecting that electricity demand will rise around 2.5% from 2022 to 2030 with data centres representing the largest growth segment, nearly a third of total new demand.
That's a big number. - Now that's such a huge story, isn't it? And 'cause we hear about these data centres being built everywhere, you know, to fuel AI and the whole quantum physics revolution. So it's just a big thing. But nobody's really picking up on the amount of electric power that these take.
And, you know, I saw a study from the Electric Power Research Institute that said by 2030, 9% of all US electric power is going to be by fueling these guys. And you know, right at the moment, the infrastructure that we have, the way to sort, we just can't cope. What we really need to figure out now is what we're going to do about this. 'cause we've got the climate change on one side and the other, we've got, you know, the whole world in AI. So what are we going to do about it, Autria? Down to you.
- You're right, I mean, now the race is on to try to create cleaner, greener energy, right? Because a lot of these companies behind machine learning, quantum computing, AI are the same ones championing green energy and clean energy. So they've kind of got to now put their money where their mouth is. So with that in mind, at the recent Materials Research Society’s Spring Meeting, Katie Brace sat down with Henry Snaith, who is a professor at Oxford University to discuss a new material in solar cells. Now these are revolutionary because they offer a very unique ability to boost performance by delivering twice as much power per square metre as existing technologies take a listen. (upbeat music) - So to start off, why is photovoltaic solar energy emerging as a heavy hitter in the future of clean, sustainable power? - So photovoltaic solar energy is converting sunlight into electricity. They're the panels you see in fields and on houses.
Basically they've got very efficient and very inexpensive. So the cost of producing electricity from solar in the sunniest locations of the world is actually cheaper than any other source of power. So they're already cheaper than producing power from burning coal, for instance. They're completely green and sustainable, and there's a load of headroom to make them even better. So from here on in, they just get less and less expensive at producing power.
- So in that vein, what's exciting about the role that metal halide perovskites could play in this future? - So I mentioned that the technology today is good, it's efficient, but it could actually be a lot more efficient. And perovskites have have a possibility to really boost that performance and ultimately deliver twice as much power per square metre as existing technologies of today. On top of that, there's a big drive towards making all the materials even more sustainable. So for instance, a panel today might take about five months to eight months to pay back the energy you put into manufacturing it, 'cause there's a lot of heat goes into melting the silicon and creating the silicon. Actually perovskites could overcome a lot of that and bring that down to maybe a month or something. So that's also a really important aspect of future technologies.
- And these materials were discovered fairly recently. How have they stacked up against other photovoltaic cells in that short time? You kind of touched just upon that. - Yeah, so with the perovskites, so they were first realised in solar cells in 2009, so just over 10 years ago. And in that time, the efficiency improvements have been staggering. So today just a single layer of perovskite is about the same efficiency as the very best silicon cells, which were first introduced in 1956.
So it's taken 70 years for silicon to get to the same place perovskites, in terms of efficiency, have got to in just over a decade. But what's really quite useful for perovskites and something you can't do with silicon is we can actually basically stack multiple perovskites cells on top of each other. We can tune the band of light, which they absorb, and that means we can absorb different parts of the sunlight in different cells stacked on top of each other.
And this can deliver a much higher efficiency than silicon could ever deliver. - And you're mentioning a lot of great things, but what are some of the criticisms about using perovskites in solar power? - One of the criticisms is, or one of the uncertainties is longevity, how long will they last? Solar modules are warranted for 25 years. In fact, there's a big drive in the PV industry to extend that to 35 and even up to 50 years warranty, 'cause the amount of power you produce, the cost that power depends on how long you produce it for. Because perovskites are so new, we didn't even make any devices much more than 10 years ago. We haven't had the data and the learning experience to get them to that same level of stability and prove it. So that's where the risk is.
But of course where half the effort, so half the effort's gone into making them more efficient, the other half's gone into making them more stable. And now in the research community actually activity towards really understanding and making these materials stable is where there's a growing interest and growing impact from the research. - And you kind of just touched on it once again, but should these critiques warrant scepticism? Or really do the pros it seemed like outweigh the cons? - I'm obviously completely biassed in terms of my view of whether perovskites are a good thing or not.
The pros definitely outweigh the cons. There should be scepticism there because we need to address these things and properly, and not just saying, "Oh, it'll be okay," because it won't. We need to do thorough science research and development to enable that to be okay. And that is going on and that has gone on, and the progress that's been made in efficiency has been matched.
In fact, in stability the improvements are usually in orders of magnitude, so factors of 10 improvement. Time on time on time, this has happened. And they're very closely approaching the stability of silicon modules. - So what does the path toward widespread adoption look like for this particular technology? - Okay, so the path, there's multiple companies, tens of companies around the world working on this.
Some companies sort of declare they've had some products, some early products launched. Other companies are launching products, you know, this year, and other companies will be a few years down the line. What we're seeing is basically the early deployment of real commercial products now. They then spend time being deployed, customers become comfortable with the technology, they know and realise that it does what it says on the tin, so then demand can ramp up. But what needs to happen is then manufacturing capacity needs to ramp. Similarly to silicon, this takes time.
So typically takes three years from specification to actually having a production line working. So we're going to see small-scale deployment over the next few years, and then it'll be sort of three to five years before we start to get, you know, many gigawatts. And then beyond that, once it works and it's proven, there'll just be a very rapid rise up to hundreds of gigawatts of production. That could happen this decade. It might be between 2030 and 2040. (upbeat music) - Autria, you know, I love that interview for lots of reasons.
You've got this very clever guy from Oxford University sitting there with all the answers, right? We just listened to him, and it's all to be fine, right? But what I liked about it was, first of all, he says, you know, "solar power already is really efficient." It's already within five months. It pays back its kind of investment.
That was really cool. But the new technology down the pipe, he said is, you know, going to make it even better, work, you know, even more efficiently. And you know, even Sam Altman, CEO of OpenAI is investing in solar power, isn't he? So it does seem that that might be the way to go. - Yeah, I don't think you can overstate the return on investment when it comes to solar energy.
And it's great that so many of these, you know, AI and machine learning companies are realising that and really trying to harness that power, as, you know, the future of green energy goes. Another company that's doing the same is OpenAI's partner, Microsoft. They've invested in nuclear fusion company, Helion Energy. Government labs and more than 30 companies are already racing to generate power from fusion, which also has the potential to slash emissions linked to climate change and generate that power that we obviously so desperately need without producing those long-lasting radioactive waste.
That's all coming from Reuters. So it's interesting how a lot of these companies are starting to kind of pick up on, "Okay, this is where we need to be putting our money right now so that we can continue to make a profit down the road." - All right, Autria, I have one up on you now because I'm joined by Laila.
Laila is our head of research here at WebsEdge. I'm going to turn to you, Laila. And now what do you make of what we've been talking about so far? - So focusing on nuclear fusion, which is a topic which can cause mass hysteria, causes people to panic when they hear nuclear.
But nuclear fusion has the potential to provide us with an almost limitless source of clean energy. It's basically like having the sun in the palm of our hands. Nuclear fusion is taking kind of lessons from nature. So the sun is pretty much a giant nuclear fusion reactor in space. So if we're able to produce that on Earth, we are pretty much sorted when it comes to the energy problem. - And how unlikely is that? I mean, that just sounds like that's great, isn't it? That's a box, we ticked, that we've sorted the energy problem.
- There's a lot of debate on this. Some people say it's, you know, completely not going to be achieved in our lifetime. But people also say that about lots of technological innovation. You know, it was 60 or so years between the first flight and us being on the Moon. So the rate of kind of advancement that we've seen in the past 100, 150 years couldn't have really been predicted by anyone. And Helion is, you know, maybe closer than you think.
So they're building the world's first fusion power plant, which is, you know, actually making quite a bit of headway because of that huge investment by Microsoft. It's the biggest personal investment that Sam Altman himself has made. So, you know, it is accelerating. - Gotcha, now another bit of technology that's copying nature is neuromorphic computing, right? - Yeah. - And my understanding of that is that it's a 3D way of doing computing. And it uses a lot less power to be able to process data.
Because like we've been all talking about, it's all about the power it takes to process data, right? - Mm-hm, yeah, so neuromorphic computing, again, looks to nature to try and solve the challenge of using large amounts of energy to process data. So neuromorphic computing co-localizes memory process, memory and processing within the architecture. - Interesting, so our Katie Brace actually spoke to Mark Hersam of Northwestern University at the Materials Research Society meeting about his work in neuromorphic computing. Let's take a listen. (upbeat music) - So what has happened to the rate of improving performance of computers in recent years? - Yeah, so there's two major issues that have, I would say, limited further improvement in computing performance.
The first is power consumption. This is probably the biggest factor. If we tried to run a digital computer any faster than we do today, the temperatures would become uncomfortably high to the point where ultimately the silicon would melt on which the chips are built. That is, of course, not sustainable or possible, so we need to eliminate that possibility by, we're operating at lower speed. The second issue is that the trajectory for microelectronics has been to make the transistors smaller and smaller with time, thereby increase the number of transistors per unit area.
And we're approaching atomic length scales, which is a fundamental limit. And so the fact that we have a two-dimensional or planar architecture is another fundamental limitation to further improvement in computing performance. - So playing off of that, what is the Von Neumann bottleneck and what role does it play here? - Right, so John Von Neumann is credited with the basic design of a digital computer. And that design consists of a memory block and a microprocessor separated in space.
- Okay. - Which means that you need to move data back and forth between those two blocks. And for the past five decades, the amount of data that's been moved is small enough that that was not a limitation.
But in today's world of big data, that has become a bottleneck. It limits speed. And perhaps, more importantly, the movement of data consumes energy, and that leads to further power consumption limitations. - So how can neuromorphic computing possibly circumvent this? - Right, so neuromorphic computing is brain-like computing.
And if we look at the brain compared to a digital computer, one of the fundamental differences is that memory and information processing are not separated in space. They're co-localized in space. And as a result, you don't need to move the data around as much, and that leads to orders of magnitude lower power consumption compared to a digital computer.
Another fundamental difference of the brain is that it's not a two-dimensional system, it's not a planer architecture, it's a three-dimensional system with a higher level of interconnectivity than we have in a digital computer. - So how could this system end up being more energy efficient, a possible energy efficient option? - Right, so by co-locating memory and information processing, we eliminate the need to move data around. That will give us a huge reduction in power consumption. Additionally, if we can go to a three-dimensional architecture, then we overcome the limitation of being constrained in two dimensions where the only option to increase device density is to go smaller. - Gotcha. - And we can't go any smaller. So we have to now imagine going into the third dimension.
- Wow. So what's promising about using neuromorphic devices for machine learning and AI, which is a lot of data? - You got it, so the way AI gets better is to train it on more data. That's the whole concept of ChatGPT, to train it on as much data as possible, which runs into the power consumption limitation. In fact, it's estimated that, by 2027, the training of AI will consume as much energy as entire countries consume. - Wow.
- Which is an unsustainable path that we are on. So neuromorphic computing, by reducing power consumption, enables AI to proceed in a sustainable manner. And even more lofty ambition would be to acknowledge that AI, as we have it today, falls far short of real intelligence. And if we ultimately want to mimic real intelligence, then what better hardware platform to use than the brain itself, hence neuromorphic computing.
(upbeat music) - That's such an interesting interview, isn't it? But you know, that's the second one we've had today, right? Where this very smart guy has sat in a studio with one of our representatives and has the answer, right? So this is the answer for the issues that we've got with this neuromorphic computing. So my question to you is, you know, when's this actually going to happen? - So that's a difficult question, because conceptually it all seems pretty great, you know, creating something which mimics the brain. But what we've kind of developed so far, which are artificial neural networks, they unfortunately don't, you know, translate a map that well spiking neural networks, which is what a neuromorphic computer uses. So there's actually quite a lot of resistance in the AI community to, you know, transfer to the neuromorphic computing method. Although, you know, conceptually it seems like there's a lot of potential, but the threshold and the transfer kind of energy to get there is quite substantial.
- You've been looking at something similar, haven't you, Autria, with quantum computing, right? - Yes, so not to throw, you know, another option on the table, but quantum computing could also offer some greater energy efficiency. And one company that I want to talk about right now is D-Wave because it's the world's first commercial supplier of quantum computers, and it's actually the only company developing both annealing quantum computers and gate model quantum computers. I had the pleasure of talking with Murray Thom, who was the vice president of Quantum Technology Evangelism at D-Wave. Take a listen to what he had to say about energy efficiency.
(upbeat music) So let's get started, let's go back and kind of set the stage and tell me how your quantum computers differ from the more traditional models. - I think a useful way for us to think about quantum computing is that we're building machines that allow us to use quantum effects to accelerate calculations. And so it turns out there's multiple ways that we can think about quantum effects acting as a resource for our computers. In some models, we can use the quantum effects to allow the computers to store more information. And that's really useful for applications in like, you know, molecular simulations and quantum chemistry and things like that. In other models, we can use the quantum effects to allow us to move between solutions more quickly, and that turns out to be best suited for applications in optimization.
- Why was this your focus from the get-go? I feel like you were really getting into a terrain that others were staying away from when it comes to quantum computers. - By choosing, you know, a particular model, the annealing model, we were basically going to be able to impact near-term applications sooner, which was really important to us. It was also much easier to build those machines, so we were able to scale the systems up much more quickly. And some of the largest, most highly-connected quantum computers on the planet are D-Wave quantum processors available now. You know, we are really focused on trying to make something incredibly powerful also really easy to use so that small teams of developers of maybe two to three people can make really high-value features for their applications and drive a really important impact and value in their business.
- So here on "Agents of Tech," we've been discussing the explosion of energy that will be required to power AI. And in fact, just to quote one little estimate, some reports say that within five years, AI will be using more power than the entire country of Iceland. Do you think that that's an accurate assessment? And two, given this prediction, do you feel like companies like D-Wave should be investing in and prioritising the use of clean energy so that that energy is there when AI is ready to use it? - Quantum computing has this really unique feature which incredibly low power. So operating systems, that use like 15 kilowatts of power, which is, you know, probably about as much as we would use on like an office building room, but the chips themselves consume less than a 10th of a millionth of a watt in our quantum computing systems. So basically what that's meant is that our first generation through to our fifth generation quantum computers have had no power growth, if they've been basically consuming 15 kilowatts that whole time.
So if we can offset of these really difficult calculations in machine learning onto quantum computers, we can really transformationally change that power consumption profile, I think, in a positive way. - I want to pivot just a little bit because the upcoming year, 2025, was chosen as the international year of quantum science and technology. And the reason is because it recognises 100 years since the initial development of quantum mechanics. And according to the proclamation by the United Nations, this year-long worldwide initiative will be observed through activities that are aimed at increasing public awareness on the importance of quantum science and their applications. Do you feel like we're kind of at a watershed quantum moment here? - Yeah, I think so. I mean, part of the reason why quantum systems feel so strange to us is because we don't really interact with the world at a quantum level.
You know, it took us until, just like you're saying, you know, 1925, to actually start to be able to have the level of control where we could interact with the world and realise that, like, at its fundamental nature, it behaves differently. By making quantum systems more readily accessible, so for instance, like we have a leap quantum cloud platform that allows people to sign up and get trial access to our systems and begin programming them. And those are our 5,000 qubit systems. That's a mechanism through which people can interact with the quantum world in a programmatic way and start to see how it interacts and how it responds.
And I think that when we're able to take something like quantum physics and turn it into useful work, it allows us to understand it in a new sort of fundamental way. So there's an incredible amount of accessibility available now for people. You know, turning it into useful work I think is a very useful tool for us to understand it, just in the way we learn so much about, you know, information processing with the development of classical computers. And also, you know, the ability to use that in applications has been become incredibly easier. So we now have businesses that have taken their quantum applications into production. And so we're in this phase here where businesses are getting a lot of value using quantum computing.
And so kind of like you're mentioning, it's very important for people to become aware of the fact that tool is available to them, and that, you know, they can begin by making a difference, you know, using it and actually getting utility from it, quantum utility. And then, you know, a lot of developers that I talk to sort of say, like, once I know what I can do with quantum computers, I'll be more interested to learn more about it. So I think it'll be an incredibly valuable avenue through which people can discover a lot of interesting things about our world.
- You set me up perfectly here for my final question, which is, where do you see quantum computing maybe just five years from now? What do you think this technology will have the power to do that maybe the general public hasn't even fathomed? - Let me share something with you. I was just at a supply chain conference in Atlanta earlier on this year. And I was telling people, you know, we have retail grocers in Canada here that are using our systems to optimise their grocery fleet and reduce the time to create their schedules by 80%. Vesti Energies in Europe, they've been using our quantum computers to optimise the development of like, heating, ventilation, and air conditioning systems in new buildings to get systems that cost less with, you know, simpler components that are easier to actually implement. And when I spoke to people at that conference, they were like, "I didn't even know quantum computers existed, let alone that they're being used in production."
So five years from now, I think there's a real opportunity for us to take on some really challenging problems. And I think that, in the information revolution, we got flooded with so much information. The challenge sort of in the 70s and 80s and 90s was how do we find the information that's available? And you know, through search and, yeah, search engines, we've really solved that problem. Now we're in this process where we're, like, generating a lot of information, and what's difficult is validating it, you know, even defining how to validate information. And some of the work I've done with some really talented machine learning and artificial intelligence experts, they've talked about these limitations to these models that we're using, and you can't learn logic.
It's actually incredibly challenging for them to learn logic, which is like, this is true or this is false, and here's the relationship that tells you if something is true or false. And so my hope is that, you know, five years from now we're actually going to be able to use these quantum neural networks to be able to learn logical relationships and then, as a result, train our systems to give us information about, you know, here are the references and here is the logic that tells us whether this information is valid or invalid. And I think that will be a really incredible tool for us in the future.
But, I mean, five years from now, I mean, all sorts of things can happen. So it could be something completely different and fantastic. (upbeat music) - Interesting interview, right? - Absolutely.
- But I have to tell you this, Laila, if I had a pound for every time somebody had told me about quantum computing or about AI, I'd have retired a long time ago a rich and happy man. But unfortunately not, right? So what is it about quantum, and kind of where are we with it? - So when people talk about quantum tech, they primarily talk about quantum computing. And quantum computers do have a lot of potential when it comes to crunching data, which our current computers can't really keep up with. But when it comes to actual consumer use of quantum computing, it's more difficult to see, you know, how the average Joe is going to use their quantum computer. What data are they crunching? What large amounts of things are they going to process with their personal quantum computer that they can't currently process with their personal computer? But when it comes to kind of federal government and larger organisations, there's tremendous potential.
But in addition to quantum computing, there's other kinds of quantum tech. So there's quantum sensors which would allow us to, for example, measure the Earth's gravitational waves, the heart's gravitational waves and help with navigation. So, you know, GPS, which we use at the moment, is incredibly vulnerable and very fragile, as in a solar storm could knock out all kind of navigation tech. So quantum navigation has huge potentials in that respect, and so does quantum communication. So in terms of security and encryption and moving around sensitive information, quantum communication is, you know, head and shoulders above the current system- - And as Autria and I found out earlier this year, quantum dots are in our televisions already, right? - Yeah. - So quantum is everywhere, but maybe not quite yet in a quantum computer.
Well Leila, thanks ever so much indeed for joining us and setting the worlds to right for us. We really appreciate, it's been great to see you, so thank you. - Thank you so much for having me.
- All right, so next time, Autria, what are we looking at? We're looking at mental health and social media and a warning from the US Surgeon General. - Right, I mean, could social media and technology be as detrimental to children as smoking? We're going to get into all of that. There's a lot to unpack there.
So make sure that you join us on our next episode of "Agents of Tech." And be sure to like and subscribe wherever you get your podcasts from. Until next time, goodbye. - Goodbye. (upbeat music) - [Autria] "Agents of Tech" is brought to you by WebsEdge.
2024-09-04 23:25
