Jane: [00:01] Hello and a very warm welcome to the LSEG Sustainable Growth podcast, where we talk to leading experts about sustainability and finance and pretty much everything in between. I'm Jane Goodland, and this week I feel very lucky to have spent time with Sally Radwan, who's the Chief Digital Officer at the United Nations Environment Programme. We talked about UNEP’s take on AI, including the risks, impacts and opportunities with respect to sustainability. But before we hear from Sally, a quick reminder to follow us and rate us on Spotify, Apple Podcasts or any other platform you use. Enjoy the show.
Jane: [00:36] Well hello there, Sally. Thank you so much for joining us on the podcast today. I can't wait for our conversation because I think you've got probably one of the coolest jobs out there. You're the Chief Digital Officer at UNEP. So let's start with, tell me a bit more about UNEP and also what you get up to there.
Sally: [00:56] Thank you. Jane, thank you for having me. It's great to be here. So, UNEP or the UN Environment Programme is the part of the UN that deals with the environment as a whole. So we call ourselves or people call us the global authority on the environment. So we help set the global environmental agenda. We have, of course, all of the UN member states as our constituents. And we promote the environmental dimension of sustainable development. So when it comes to the UN Sustainable Development Goals we're actually custodians of 26 Six of the indicators across six of the actual SDGs. And we also have the mandate to keep the environment under review, and we can go into what that means exactly. We're headquartered in Nairobi, in lovely Kenya and it's actually the only UN headquarters in the Global South. So quite proud of that.
Jane: [01:52] And how long have you been there, Sally? Out of interest?
Sally: [01:54] Just over two years.
Jane: [01:55] Okay. And how's it going? You're settling in?
Sally: [01:56] Yeah. It's quite a lovely place to be. If you're going to have an environment program based somewhere, I think Nairobi must be at the top of the list. The biodiversity and closeness to nature here is just unbelievable. And I get to work at the intersection between that and my passion in life, which is technology and computers and all these lovely things.
Jane: [02:24] Because your background is more on the tech digital side, right? Am I right in thinking that's your kind of that's where your career started?
Sally: [02:33] You're right. Yes. I'm a computer engineer by training. And after that I also did an MBA. So started kind of going into the business side a little bit, but more really to understand the business aspects of technology rather than move completely to the business side. And that's where I've been all my life in various capacities, whether it's the private sector, government or now the UN.
Jane: [02:57] Excellent. Okay. So you get to blend all of these fascinating worlds together. And that's why I said you've got a really exciting job, because I think, you know, you're very much at the nexus of these big, big themes in the world which are which are kind of shaping the world we live in. So tell me a bit more about UNEP's key objectives when it comes to data, and particularly AI as well, because I guess it might not be the first thing you think about when you look at kind of the UN Environment programme. Oh, yes. Well, we've got to think about AI. So help us understand what the objective is of your work, as it were.
Sally: [03:34] That is very true. And indeed the role is quite new. So I'm the first chief digital officer that UNEP had. And UNEP also created the digital transformation Subprogramme, as we call it, which is. So we organise our work along seven sub programs and digital transformation is now one of them. So it's been elevated to become one of the priorities of the organisation. Now you're quite right that technology isn't necessarily the first thing you'd associate with the environment. However, it is growing very strongly in importance in that domain, mainly because of the data as you mentioned. So there are vast amounts of data that we can and we do collect, whether it's environmental data, just monitoring and reviewing and tracking or whether it's now with the different multilateral environmental agreements like the Paris Agreement and the Montreal Protocol and all these things, they all have components that require member states to collect data, report on their commitments, keep them under review, etc. and so data and consequently digital technologies that are able to process this data, including, of course, AI, are becoming a really prominent part of this.
Jane: [04:50 ] So in a word, I mean, or sort of in a sentence, what's your what do you wake up thinking about? Like what's kind of you, your primary goal really as the, as the digital officer for UNEP?
Sally: [04:58] So we describe our work in a simple sentence that says we try to be digitally sustainable and sustainably digital. And what that means is we look at how can we harness digital technologies, including emerging technologies, AI and all the rest to fulfilling UNEP's mandate. But also, while we're doing that, how do we ensure that those technologies themselves are environmentally sustainable, responsible, sound and don't damage the environment? So that's what I do every day.
Jane: [05:40] Excellent. So it's those two lenses, and it's important that you've got both sides of that of that coin I suppose. So I want to touch on AI of course, because it's a very topical subject. And when we think about AI, when we hear AI. Often it's really easy just to zero in on the increased energy consumption. I think that's kind of quite naturally where people go to first. But I know from a recent paper that you guys published, Actually, we need to probably think a bit more broadly about the environmental impacts of AI. And I just wonder if you can share a bit of that with us.
Sally: [06:07] Sure. So we, as you said, we take a balanced approach, especially to things like AI because we do see the huge potential it can offer to keeping the environment under review and to helping tackle all these challenges. But it is true that AI has a significant environmental footprint. One of the issues that keep me personally awake at night is we don't really know what that impact is, exactly. There is not enough research, not enough reliable data out there to help us quantify and therefore decide whether or not we need to mitigate that impact. But it's also bigger than energy consumption, as you mentioned. So in the paper that we published, we talk about the entire value chain of AI eye. Starting from raw materials, so rare earth elements and minerals and metals that are, by the way, also crucial for the energy for the green energy transition. But they are also critical for the manufacture of GPUs, for example, and other types of chips that are so critical to AI. And then we go into the construction of data centres, because now most of the AI work happens in data centres or requires data centre capabilities. So you've got the extraction of those raw materials.
You've got the construction of data centres, which are often not the friendliest to the environment. And then you've got the operation which requires in addition to high energy, it also requires a lot of water for cooling unless you've got air cooling. But if you've got a water driven or water cooled data centre, then that requires a lot of water, which is of course connected now to problems of drought and water scarcity that we're seeing all over the world. And not only can it lead to instability and conflicts and strife, and it already does, in fact. But also if we talk about the digital divide or the AI divide that's now happening between the global North and the global South, that is exacerbated even more because you're automatically disadvantaging those countries that suffer from water scarcity. And then you've got, of course, energy consumption, greenhouse gas emissions. And then finally waste and e-waste. And unfortunately, a lot of the e-waste gets illegally transported to Africa. So again, the global South is disproportionately affected by those environmental effects because, again, the recycling or the dismantling of these devices is usually done in a very environmentally unsustainable way.
Jane: [08:43] Can I just ask you a question about the e-waste bit? Because and this may seem like a stupid question, but why is it that a lot of that e-waste ends up in the global South. Why does it why is it not getting kind of Disposed of in a appropriate manner in the country in which it's generated?
Sally: [08:56] That's a very good question. A lot of it has to do with cost, of course. Many of these elements are not recyclable or only to a certain degree, recyclable. And so it's much easier for companies or for countries to just ship them somewhere else rather than spend the money to dispose of them. I think there's also an awareness problem, and that's where consumer information and consumer awareness really comes to the fore. So how can we make sure that there's pressure from consumers to stop these practices or to have more responsible recycling practices? But of course, this has now created kind of a parallel or a shadow industry of the global South attracting these e-waste components to try and reuse them somewhere. But because it's not an organised industry and because it's not properly monitored and properly funded, it ends up being, first of all, even still wasteful because you're not completely recycling what you could. But then also the practices themselves are not environmentally sustainable.
Jane: [10:10] Yeah. And I think it's probably a bigger problem than we than we currently talk about. Right. And I think I was, I was just looking at this paper that you've drafted, your teams drafted, and I'm sure there was somewhere in here it talked about kind of the growth in numbers of data centres, and I'm not sure. Is that something that, you know, off the top of your head? Because I remember when I read it, I was I was fascinated at the kind of the growth that they were seeing. And it sort of really brings to life the scale of the challenge when we think about data centres.
Sally: [10:40] Yeah. So there are a few numbers floating around and again, there isn't enough data around this. But one number, for example, says that the number of data centres worldwide has surged from 500,000 in 2012 to over 8 million last year.
Jane: [11:00] That's the one. I remember reading that and thinking, my goodness you know, I can understand the scale of the issue right now. And actually, I think it's really interesting what you were saying earlier about the location of data centres as well and the extent to which, you know, placing data centres in areas which are either already kind of water and climate stressed or certainly will be in the future, doesn't seem like a terribly sensible idea in hindsight. But presumably there's limited, limited kind of rules around where anyone can put a data centre. Right?
Sally: [11:33] Exactly. And it's becoming a political problem as well, because we were talking about data earlier and the importance of data. And many countries now see it as a sovereign issue to have their own data centres, to build as many data centres as possible on their own soil under their own control. Even if it ends up being a cloud data centre, but it still has to be within the boundaries of my country. And if you're unlucky and your entire country suffers from water stress, or there aren't particularly good areas for building data centres, then you've just got to build them somewhere. So unfortunately there is this global trend of increased data centre construction. So I think this surge that we've seen from 500,000 to 8 million is only going to get even steeper.
Jane: [12:22] Yeah, that's fascinating, isn't it. We've tackled the direct impacts, environmental impacts. But that's not where it stops is it. There's definitely kind of other issues that we need to think about. And I'm keen to kind of understand some of the more indirect impacts of AI in particular. So, for example, some of the things that you've talked about is around access to information, whether that's, you know, and how credible that information is or changing consumption patterns, help us understand a bit more about the other things from an environmental perspective, more in an indirect sense that we, you know, we might want to think about.
Sally: [12:59] Yes, exactly. So in our paper we kind of categorise the effects into direct, which are the ones we talked about, and there are enough challenges with those, as we said, because we can't even measure them yet. But then we start getting into more speculative territory with the indirect effects, and then we've got higher order effects, as we've called them. So the indirect effects are things like you know, using AI in highly targeted advertising, even more than we've had in the past where you can create content almost tailored to every single user, you know, the market segment size of one that we've been talking about forever in marketing. But then things like the proliferation of autonomous vehicles due to advances of AI leading us to each wanting to own our own personal vehicle instead of moving to more environmentally friendly mass transport and so on. So these are the more direct indirect impacts, but then the higher order impacts are really even more speculative, because here you're talking about longer term shifts that will happen to certain sectors or even to social aspects that can only be seen over a long period of time. Things like. So we've got misinformation, disinformation today due to AI, because the content that is created by generative AI is so credible, and the videos are so well made that it's extremely hard to distinguish what's true and what's false.
But then think about the longer term effects of this. How is society then going to react to online or any kind of digital content? How is public trust going to erode in any kind of content that you see, especially with digitalisation becoming more and more important in our lives? But then there could also be more positive, higher order impacts. Like for example wider public engagement with environmental science through accessible AI generated content. So if I can help us simplify the message and help us really help the public understand the issues and how can they play a role in the environment or indeed promoting more circular economy models and, and having them become mainstream? So there are all these sectoral shifts and social shifts that can happen due to the proliferation of AI that will have an effect on the environment, among other things, that we can't really measure now or even predict. You've got to speculate a little bit. So that's why we're saying we've got to start investing in this now, maybe focus first on the direct impacts because they're the most immediate, but then definitely don't lose sight of the indirect and the higher order impacts. Start thinking about them now and what we can do about them to grow them if they're positive and to benefit from them, or to know how to tackle them if they are negative.
Jane: [15:58] I think that's a really helpful way to think about AI in its entirety. Somewhat a little bit disconcerting that we really don't have anywhere near as much information or research to really get our heads around in a credible way what the longer term impacts could be, though. So more work needed, more research needed is the message I'm hearing. You guys are doing some really fascinating things in the digital space, and you share some of those with me earlier when we met. And I'd love to kind of just touch on a couple of these and it really demonstrates, I suppose, the other side of the coin in terms of how AI can actually provide really interesting opportunities, Innovation and around kind of the positive side of the story. And I wondered if we could start with a piece of work that, that you guys have done on the world Environment Situation Room. Tell me more about that.
Sally: [16:50] Yes. So first of all, we love the name. And it works really well with our mandate to keep the environment under review. So you're looking at and we're trying to make it a little bit like an actual situation room. So the idea behind it is really to have a platform that can host both environmental data and applications, tools that can help us and help the world really keep the environment under review. And we've been kind of building it piece by piece over time and trying different things and failing sometimes and then doing it again. But now, with the strength of AI and the strength of infrastructure technologies and cloud technologies, we really feel that the world needs an integrated platform where you can just combine different types of data to help construct complex scenarios. Because, as you know, we live in a world of poly crisis. It's no longer just enough to track biodiversity loss over 20 years in a certain location. You also have to correlate that with climate change, with weather events, with human migration, with health, with even with security and conflicts and strife and global supply chain security and all these kinds of things. And it's a combination of slow onset events like biodiversity loss, which happens over decades, to things that happen immediately, like extreme weather events, for example, floods, tsunamis, things like that. And so having the ability to combine these different data sets that come from different sources, that have different formats and granularities and even different levels of quality, and being able to use them and mould them into these scenarios. That's why you need a robust platform, and then you can build these different vertical use cases on top of it. So this is pretty much what the World Environment Situation Room is. We're now building that foundation. So the pipes and the basic building blocks. And we're also starting to build some of the use cases on top.
Jane: [19:07] And then who ultimately who would be the users of this type of platform. Would it be governments?
Sally: [19:14] Yes. So ultimately our main customer, so to speak, are the governments specifically policymakers who can benefit in many ways from these data sets. So one of the use cases, for example, is a do it yourself kind of create your own data product, data visualisation. So if you go in as a policymaker and you say, show me the effects of climate change on deforestation in Brazil over the last ten years, and I want this in a I don't know, in a time series chart and also in two paragraphs of text that summarise what happened. And you just enter this query in natural language, the way I just explained it to you, and the system is smart enough to generate this for you. So as a policymaker, you're able to communicate this quickly. You're able to make decisions based on it. You can also construct forward looking scenarios like what happens if I want to change, I don't know, 500 buses in a certain country from diesel to hydrogen fuel. What are all the variables that go into that, including cost, but also the impact on different environmental pollutants? And then what happens if you change that number from 500 to 10,000. What's the additional benefit you get and what's the marginal cost. And this is quite a useful tool for policymakers but also for another group of users which are donors, trying to make a funding decision, you do need all these variables. And you do need these forward looking scenarios to be able to make a sound decision. And then, of course, there's there are various parts of the general public, including students who are interested in doing research on the environment. But there's also the general public who want to inform themselves on the latest in environmental science. So there are different types of user groups, and we're trying to tailor the experience to their needs.
Jane: [21:13] That's fascinating. And I can really see how a policymaker looking at different potential approaches should be or would benefit from saying, well, what happens if I do this? Because it's really about kind of thinking about knock on impacts as well, isn't it? Like you said, that so many of these issues are interconnected, like taking action in one space may well create unintended consequences elsewhere. So yeah, sounds really fascinating. Look forward to seeing that when it's fully fledged and operational. Let's move on to climate and greenhouse gas emissions, because obviously that's a potentially ripe area for AI to help to understand and model and to and to track. Are you working on any innovations in that space?
Sally: [22:00] Yes, a number of them. One that is live now is what we call ion methane which is part of something called the International Methane emissions observatory. So what we do there is we use analysis of satellite images to detect methane emission hotspots. So whenever there's a plume somewhere, we start detecting it, we track it. And then once it reaches a critical level or just before that, actually, we send an early warning signal to the governments concerned so they can start putting their mitigation plans into effect. And this has been trialled successfully with at least 3 or 4 countries now. And they're finding it very useful. Because that's the concept of early warning, right. You kind of need to know things before they get to a critical level. And this is where machine learning comes in. So this the component is called Mars, which is this this early warning and alert system that analyses these plumes before they get to a critical level and sends these early warning signals.
Jane: [23:04] And so you've got satellites looking across the world, working out where there's discharges or releases of methane, which of course has a kind of massively huge global warming potential over and above carbon dioxide. So a really, really important greenhouse gas to get under management. How can the model determine what's an accidental release versus a natural kind of or expected. Is that all in terms of the learning that the AI model kind of puts in place? I so don't know what I'm talking about when I talk about technology. So apologies.
Sally: [23:36] No, you're absolutely right. So in in most of these systems what happens is a combination of the learning. So you take historic patterns, you take events that have happened before and how they've been evaluated, including by human experts. You also have your evaluation, which is done after the training phase where, again, human experts look at this and make sure that the results that they would come up with are the same that the system would come up with. But then the strongest type of AI models are a hybrid between that and rule based systems. So you'd have your knowledge, your science expertise built into the model as well, so it can augment that knowledge that comes from the training. Which is why the longer you operate these models, for the better they get, because the learning effect and the different types of events that could lead to a leak, for example, or the different patterns or the different timings at which you can issue an alert or all these things they get better. As you know, the more data you have feeding into the system.
Jane: [24:40] Brilliant. Sounds great. Earlier in the conversation, we did touch briefly on information, access to information and misinformation. On sustainability and environmental issues. How is AI helping to sort of tackle this area specifically? And is there are there anything that you guys are working on to focus on that element?
Sally: [25:02] So we're working on something called environment GPT, which is a working title. It's currently in a closed beta. But what it is our own implementation of GPT. It's based on GPT four for now, but we're also experimenting with the newer models. Including so-called reasoning models, which offer even bigger promises when it comes to what AI models can actually do. But what it does is it's a chatbot, kind of like a ChatGPT. So you interact with it the same way and you can ask it questions about the environment. What it does, however, is it always draws from a corpus of UNEP approved and UN approved scientific reports so that unlike the normal ChatGPT, it's not trained on the whole internet where it could hallucinate its way through what looks like a very reasonable and scientifically sound answer, when actually it's a load of nonsense. It only draws from these sources, and when it gives you an answer, it also gives you the link to the reference so you can verify it yourself. And if it gives you two different opinions or if there are two differing opinions as an answer to the question, which happens a lot in environmental science, then it will give you both sides of the argument, as it were. And that's how we minimise hallucinations. So we haven't spotted any cases of hallucination so far. But also this is how you make sure that you're also able to substantiate your answers with the references and then draw your own conclusions.
Jane: [26:37] That sounds amazing. When's that going to be available? Because you said it's in the test phase now.
Sally: [26:41] It is we're hoping later this year. So we've got our UN environment assembly in December and we're hoping to launch it then, along with the World Environment Situation Room.
Jane: [26:55] Oh, brilliant. So folks need to watch out for that. You're doing so many interesting things, and I think that I do definitely think that you've got one of the coolest jobs out there at the moment. I think, you know, combining the power of digital and AI with the kind of sustainability challenge and the environment. Very, very cool, Sally. Very cool indeed. It's been great talking to you and learning and hearing about what your work. So thanks very much for spending time with us and sharing your experience. And also sharing some of these really amazing innovations that you guys are working on. And we will watch with bated breath for the next innovations to come out. Thanks very much.
Sally: [27:32] Thank you. I enjoyed this very much.
Jane: [26:35] Well, I hope you enjoyed that episode with Sally from UNEP. They are doing some incredible things there. And if you want to check out some of those innovations. Then go to UNEP and you can find out more if you've got questions, comments, or someone you want us to talk to, then do get in touch by email at fmt@lseg.com. That's all from me but watch out for the next episode very soon.
