The following is a C and a podcast. This is the climate conversations, I'm Jamie Hill today. I'm talking about climate change and artificial intelligence. As far as a climate crisis goes, we're all looking for solutions, whether it's the pursuit of carbon neutrality or simply trying to push the boundaries of energy efficiency. So what can I add? Well, on human interventions right now face limitations from monitoring and forecasting to optimizing energy use in datacenters. Buildings and power
grids could be a tool in fighting climate change. Yes, advances in deep learning and in computational power have made A.I. less science fiction, more non-fiction and here and now. But are there also downsides of using A.I. to deal with a problem as complex as climate change? With me today is Marcus Croft, professor of chemical engineering and director of Cambridge Care's the University of Cambridge's Research Centre based in Singapore.
Professor Craft is an expert in using computational models to reduce carbon emissions. And his new book, Intelligent Decarbonisation, explains how I might be able to end climate change. Marcus, welcome.
Hello, everyone. Thank you for giving me a chance to be on the podcast.
Great to have you. I'm going to jump straight into the first question for the average man in the street. I'll go out on a limb, and I'll say that maybe artificial intelligence wouldn't be completely understood fully right? In general, we know it helps us analyze data and make predictions, optimize systems, perhaps before we get into climate change. Give us a quick sort of overview of how EIA is already being used today as part of our daily lives. Let me not you wouldn't be entirely aware of
the use of the AI that impacts on everyone. Bailey is in the search engines and in customer feeds and web pages like Amazon or similar companies. This is where we benefit most of the AI right now. In particular, machine learning has been enormous for about three years ago, there has been a breakthrough in natural language processing using
the Transformer machine learning networks. We have also increased our ability to do automatic translation, something I enjoy using almost every day and all to try to understand what we are looking for when we type a search request into Google, for example.
So I'm going to jump on the term that you use, obviously, is something that we are quite aware of here in our newsroom. And as machine learning to sort of Segway into climate change, I'm sure there are machines out there which are learning about man's involvement and attempts to manage climate change. Give us a broad sense, then, how AI and technology in general has already supported our plans various forms to get to net zero mitigating the impacts of
climate change. What kind of EIA is this and how might it apply to a specific area? See our water usage or electricity generation?
This is a very good question, and it's hard to respond to. One of the problems with the response is that what do you mean when you see different people mean different things when they use the word on the and sort of thing, something magic happens and you get a fantastic solution out of nowhere. And I don't think this is actually the case if you look at today's technology. Of course, there is a trivial way to reduce CO2 emissions,
and that is just to stop using fossil fuels. But that would have enormous impact on our society and is generally regarded as not acceptable for staff. Typekit very difficult to go back to Stone Age. So, OK, what can we do instead? It is clear that we have to replace the fossil fuels with something else, and depending on which sector you're looking at, different solutions are coming out. The more simple one was actually in the electrical power
networks here. Many people may have heard of the term smart grid, so this is the first example where digitalisation, not just A.I., is changing the way we operate. This is partly necessary because if we use different energy sources. So, for example, the sun or wind power, then we have much more fluctuations. The on the grid does not necessarily
match the input of energy that we have. So we have to look into, for example, clever storage technology in all these aspects, which would be handled by mathematical algorithms. And here we go. What do these algorithms look like today? In many cases, because we have so much data, we actually don't need to sit down and do physics. But we could use just the data and sometimes and very
often the a combination of two. And that helps us control the smart grid or make the smart grid smart, which is then helping to secure the electricity issue for the population and industry. I will come into the control of smart grids,
so I'm going to use the example of dual island here in Singapore to drill down a little bit into how grids are smart, how energy systems are smart, and maybe I will help them get even smarter, right? In your book, you've written about use cases that are associated
with your island. What did your research? Find on how iOS could do things from reducing costs to emissions in a major petrochemical hub like Drone Island and what other applications are being tested here and what holds the most promising in your view?
I have to say that if you look at your own island, that is potentially the most difficult spot. So for variety of reasons, I personally believe that you will try to do things in other areas, in particular in the smart city projects much earlier. So you have basically two problems. The first problem is, of course, to supply the energy in a carbon footprint free way. At the moment, the energy is by meaning a gas power stations, so you would have to selectively, but which the company owned.
You're allowing to their own proprietary power stations. And we have a lot of oil generators just for safety. So and backup replacing the energy will not be straightforward, although you could argue that hydrogen may be the way forward.
The second problem you have with the chemical industry is that the product itself have a high carbon footprint in terms of logistics transporting them around, and b they are made out of fossil fuels, you know, so that the refineries produce the commodity chemicals that are then going to our product that needs to be replaced. We have a double challenge, so to say. So if you can solve the problem with two or island, you're basically on top of things, which is very exciting. And that's partly why
I thought we should look at that. But clearly, I would be lying if I said, we have a solution next Monday. Okay, this is long term research with particular things you can do now. For example, you use overall network of companies under alignment. So, for example, Company one may produce waste heat, which then can be used by another company. You see cost energy resources by doing so, and we do indeed have the study where we looked
into this. We also did a little study just to highlight how difficult the problem is where we say if you want to be brought to the extreme place nuclear reactors under a. There are many reasons why you wouldn't want to do that. But if you were just looking at the CO2 footprint, then this could be a way for. But it was more an academic study in order to highlight the consequences if you really want to follow through with zero carbon footprint challenge.
I would imagine, therefore you would sound as if in the short term, the very short term things that maybe can be improved on within a very unique situation like dual island would be improvements on the fringe rather than, as you see really huge structural changes. And on the fringe,
you are talking about efficiencies in energy usage. How far do you think the eye has already helped there and what more is out there in terms of research that can actually push the boundaries of energy efficiency in really intensive environments like Jurong Island?
I have two things to say to that. If you look at your items, you will see most of the companies. There are big international companies. So, for example, Exxon being one of them or both aid and others, they have fantastic research labs. They know very well what are the newest methodologies be on the forefront? Developing this methodology so they know about. But equally, you have to realize that the investment cycle in such big plant actually quite long.
There is always a balance between sort of capex and opex, and that will lead to an additional time delay, although sometimes technology may already be known. Implementing them, making the investment, getting a return on base, that is what needs to be looked at if you want to estimate the time it takes to completely change the industry. In my view, it also becomes clear that one of the problems is actually policy making because policy making will have a direct
impact on these costs. Most countries are aware of it, and if you look to Europe, they discuss carbon tax and trading. And I think over the next five to 10 years, it become increasingly the method of choice to push new technology in to make this investment cycle shorter.
I'm going to switch accommodation to another large company, Google. And you make reference to them in your book Intelligent Decarbonization, where you mention an Irish system called DeepMind, helps cool Google's data centers, and they are obviously quite big users of energy. And it's apparently helped cut energy consumption by 30 percent, it said. Explain how that worked. Will this technology be game changers in a way in terms of energy efficiency?
It will definitely help. It is not a Magic Machine learning hospital using controlled strategies to depend on the reaches of your data. And I'm sure this could be rolled out of the areas where digitalisation has been developed fine. So, for example, last year I was trapped in my German hometown in London, and that's not Wow. Okay, what am I going to do? So I decided to contact the
mayor and ask them thanks to a smart city. And you have to know that my little hometown, just 40000 people there are really poor and they don't have money. And I said, Let's do smart city together. And as a consequence, they were actually quite open. And we have a project which basically optimised the distribution network and we could be using a variety of methodologies that will reduce the cost for them by 20 percent. And we could use the CO2 impact into what I'm trying to say
here is what happened with Google. It's not an exemption, something that we see no doubt in many more areas. You said API is terribly expensive, and if you look at the big data centers, you have the carbon footprint, which would be high. And this is true. However, it's a bit like the electric car. If you look at the electric car, they think that solves every problem. But it's not all me if and only if the electricity that goes into either electric cars or in the beauty
sector is carbon footprint free. So it's either from solar or from wind or from fusion within. So it's not a problem. And in fact, one can say, you know, we don't have the energy from such as plenty of energy and our haven't you know what we haven't really managed to do is to get our head around how to use solar in the right manner. And this is a few that has developed enormously quickly over the last few years are now talking about perovskite cells with very
high efficiencies and here. This is also very important feed for air. So one of our activities is to automate chemical laboratory, but not just the laboratory or to automate
the site. Right. So you have robots that perform experiments to find more sustainable synthesis rules for better materials, and we can use machine learning to pick the right materials that will have, in my view, enormous impact on the development and the efficiency of solar cells, which then in turn will be because everything will be connected, can then make computing center just the ones that will just carbon footprint free. And then off we go.
That's sort of a related point that I would have then is related to the cost of the machine learning, right? Using DeepMind, as my example is, showed that just training this, the AI system is also energy intensive in and of itself. And there's been studies that find that training a huge system like that could consume significant amounts of energy and
emissions as well. Is this sort of something that you keep in mind as you look at research in terms of how and whether there is a trade off between his own energy consumption and his purpose in optimizing energy use?
I have three things to say to that. The first thing is that you are right when you say the training of an artificial neural network and thinking of particular about GBP three, look to whether you use a natural language network. It's also based on this principle technology that basically has used unprecedented amounts of data. In China, there is a similar network being trained, but you don't have to do it training every time. So you train such a network and then you have to do all minor
modification to use it so that the benefit multiply. Eventually, if you distributed over all machine learning algorithms that are then based on something that could be or transform the network, then the energy use is actually not that dramatic. Second is, if you look at A.I., it would be a mistake to just. Weighted to the kind of structure of the neural network and the way we trained, this field is
rapidly evolving, in my view. We are just about to ship a new revolution in the way we do this API. So far, these networks are based on data and data, and though there is no understanding, they're going to use it. But if you look at our brains, we have similar performance and use only a fraction of the energy and we can learn much, much faster, which is why and probability that we will get to grips with it is very high. So, for example, we spend a lot of
effort to develop policies to develop the work model. In our book, I call it the word avatar, which is basically nothing else but a representation of the world in terms of knowledge in this type of space. Now, of course, you could not just learn on the data as such, but you learn on stuff that is already known. And that, of course, will have enormous impact. If you ask me, will the training for networks that can do similar things like geometry be always as expensive as it is now?
I don't think so. I'm up set with the next few years. There will be significant progress in that respect. The final point I got already made it is our brain doesn't use that much energy, which basically tells you there is a lot of room for improvement. Frankly, we just haven't quite put our heads around it. I believe although I cannot pinpoint specific areas that will improve, I'm sure that because there is so much room for improvement that will happen.
When people talk about the vast sort of potential that's still out there and then you talk about the revolution, it's still going to come in terms of artificial intelligence is that there is still a lot of uncertainty. It isn't always positive. Much of it revolves around fears around uncontrollable forces, right? In terms of what I can do. And you know, going back to your book, one of the premises is that both A.I. and climate change cause
existential threats to humanity. Right? And I see this quote in front of me from your book and you ask this question and I want you to address that too. And you ask what happens if an artificial general intelligence decides that the best way to protect the Earth is to adjust the human population to sustainable numbers? And what
if this number is below the current world population? Talk about that and sort of the philosophical arguments that you sort of have within yourself as to the potential and how to manage that. Those are artificial intelligence and its potential to help but also addressing the larger concerns that people may have scientist crafting this specific to climate change, for example.
I personally think this is a very important question, just for the record. Okay, I want a happy, happy, happy world for everyone. So whatever we do, we have to make sure that this everybody is catered for. So the question is, what are these dangers and how can we address it? They are actually quite a bit of literature that has already done to an Austrian terms on superintelligence is a perfect example of analyzing the threats that the
Superintelligence may pose. Another good book that I've read that I really love was my pick marks like 3.0 I can. We recommend this is the amazing three days of the whole artificial intelligence aspect, and the key word that is important is Google alive. So we have to make sure that the boards of technology are aligned with what we do. And this was where the problem starts, because what are our own boards? What is your Google may not be seen as mighty Google, so how do we pay for?
I have decided that for our work that we start with the Sustainable Development Goals, which means basically the planet has to be sustainable. People should not be in poverty and they should have enough extreme. We should not have a lot of CO2. We should have been warned to everybody by. If you should make sure that the system tries to follow these goals, then my view is how bad can it be? Do you see the first approximation? We are sort of saying there will be an AI
that starts killing people anything. I think men do that themselves enough. But we have to make sure that the right conditions for everyone, the sort of barrier that will mean that there is sort of a region, not just one point. You could imagine it would be a bit abstract that if you think about the Sustainable Development Goals at seven p.m., each of those have targets that can be quantified about and then the state of the world can be represented by one point one hundred and seventy
dimensional space. Now within that space, you may have some space. The world was in that space to be good enough. But whatever happens at this point in, that space is moving. OK, so for example, if there was a massive problem, okay, we didn't have enough water for people and it would come out of it. I would say comfort zone. So you have to make sure that we constantly keep the world in this comfort zone that is found by the U.N.
Development Goals. This is the way I think about too much, and maybe that was a bit too mathematical in terms of the way I described it. It is a very important question, and what we are trying to do is we not only develop the world model, we ought to develop a way to classify the work model in terms of these goal. So only then we can do forward align.
The purpose of our artificial intelligence is actually to make sure that the living conditions for us are good enough to have a happy and fulfilled.
I'm going to start closing off our conversation and ask really large, even larger questions, maybe to get a sense of your take on it. It seems that obviously experts in the field will say that it's is not going to be a silver bullet in dealing with climate change. But there will also be optimists, the likes of Bill Gates, for example, who believe that technology will potentially evolve to
help overcome such large problems. Where do you stand in the sort of the realist versus optimist sort of spectrum? How confident are you in his potential as seen in that context, for example, in 2050, if we were to look forward next? Yes. How much of the gains in our efforts at decarbonization could actually come from gains in technology and AI? Is it something that should be part of the thinking for people and governments out there?
In my opinion, absolutely. But the time period that you have mentioned is so enormous that it's very hard to think about the technological changes. I have already indicated there may be potential breakthroughs in both the efficiency, but also in the performance of these systems. And it just will help us to solve the problems. And you know, they can be used for maintenance prediction and energy optimization or basically in the classical way. We don't need to use
a machine learning algorithm. You can use a classical, argumentative, similar results, maybe not as good as you can do now with the AI, but there is more than that because if you look at society at large, it's a complex system. There is a lot of information going back and forth. There is a lot of information loss you
see from one people to another. If you just look at how science has been done in the past, how science is done now in students from something of the life of the ordinary person to not be, nobody else knows about it. That's over very soon. Every finding is a finding that is available to other instantaneously. Imagine the acceleration of the process of finding new things and new
solutions to things, and that is all about government. If I look at how governments have to do policy, they are often in the dark and they have a report here, a report there, and they can't really base their opinions on proper fact. It's very hard for them to get it started with the smart cities and smart states and
completely change. Not only they will know at any point in time, you know what the state of this, but they can induce progress in the areas we basically call these parallel worlds in your system, you can work out what if scenarios and you can implement it right away very fast. So the time for implementation time for planning all that shrinks enormously and that will help us to solve the challenges that we have. It's not just global warming,
it's living in peace together. It's making sure that we build our environment, and I believe that this will help to stop same.
It certainly sounds really optimistic. It sounds Huebel. But as a last question, do you therefore also in your work, have some concern that in using AI, in looking forward to the potential for A.I., humans like us may also be tempted to maybe subcontract decisions using an artificial intelligence instead of, you know, I would say, hard human choices, hard human decisions that have to be made separately from what I have to deliver.
Well, I subcontract decisions in my life with software.
That's always the smart thing to do.
Yes, they do. But what I'm trying to say is this, of course, we will subcontract decisions to our own benefit. Every decision has consequences. You have to be aware of what they are and be willing to see them through. You know, if you look at this palest time machine and you have the Eloise and the Warlocks and the Eloise living in this fantasy world, where they don't have to care for anything and sort of slowly develop into vegetables, is that going to happen with humans? I hope not.
Even if there was a super intelligent, you knew everything that can, you know, it would be not relevant to us because we still enjoy each other's company. Our thoughts, of course, would be in order. You can just look things up, but that doesn't mean that you have fully understood it and you can still think about it. Or you can think about what the system was a great agency. So I don't think it'll happen impact on our meaning as a human being. And I think this is very important. Well,
Marcus Kraft, thank you very much. Thank you for being with opportunity to talk to.
And thanks for listening to the climate conversation, stay up to date on CNN's coverage of climate change on CNN Asia. You can also find this and other senior podcasts on our website and on iTunes and Spotify. The team behind this podcast, Christina Robert Insulating and Erin Low. I'm Jamie Hoh again to next week.