The Climate of AI | AI IRL

I'm not the artificial intelligence Revolution is heating up but so is the climate the summer of 2023 was the hottest on record and the race to combat the catastrophic effects of our warming Earth is on in some ways artificial intelligence may offer a glimmer of hope AI powered climate models can provide more accurate and granular predictions helping lawmakers make better and faster decisions about climate policies but technology is also a carbon culprit the world's data centers account for about 3% of global greenhouse gas emissions surpassing even those of the aviation industry machine learning had a significant carbon footprint even before generative AI took off as more companies Barrel into the AI Gold Rush that could get worse so in this episode of AI IRL we'll discuss whether AI can help slow global warming or whether it might speed it up and if the companies who have long been part of the problem can actually help unlock a solution [Music] Sims Witherspoon thanks so much for joining us thank you so much for having me so you're an applied climate AI scientist and you're also the head of a team at Google's deep mind that's actually working on climate change applications using artificial intelligence talk to us about some of the work that you're doing oh my gosh um I think I have the best job in the world I have to say um I think that uh what Google deepmind is trying to do right now um just to give kind of a broad overview of how we're tackling applications of AI for the climate crisis is that we approach how we can use AI in Lally Three core ways um the first is that we try to use AI to understand climate change and its effects on Earth's ecosystem so we really like to use AI to kind of understand the climate science the second one is that we use AI to try to optimize current systems and infrastructure because you know we can't just start over from scratch um today we have to continue to use the systems that that life depends on um and AI is a software only solution that can help optimize a lot of those and the Third Way is um accelerating breakthrough science you know the technology we need for a more sustainable future these are things like you know Fusion for carbon free energy in the future and I think that one of the reasons why I love that framing this you know understand the problem optimize current systems and infrastructure and accelerate breakthrough science is because when you think about then applying that into any sector that has challenges that are driving climate change you can see what problems are are necessary whether that's are necessary to tackle and um applicable for AI whether that's across electricity systems Transportation industry buildings and cities I was going to ask you about specific challenges that AI can help with compared to traditional methods what would a couple of those specific challenges be that you're you're seeing really being helped yeah so um maybe I'll start with the first one around um understanding climate change we're doing a lot of work understanding in weather and um forecasting we have models that do um rainfall forecasting a few hours into the future and we've been working with a Met Office and those models have been um voted and by over 50 Met Office forecasters as more accurate the Met Office in the UK yes yes over 50 Met Office uh forecasters yes in the UK um as being more accurate um and more reliable than the traditional methods so take something like wind power let's say like how do you how do you approach a uh the owner of a wind farm and say please can we have your data so we can model it in an AI system and predict the weather forecast for the next however long so what we would end up explaining is first like the utility of AI you know if we had access to not only readily available weather forecast um but also your operational data around okay the wind was blowing this much and this much wind uh Power was being produced our models learn the relationship between the weather at that time and how much power was being produced and so we can say you know we can watch those weather patterns um it can learn that relationship and then it can make inferences about what future power availability will be based on what the weather forecasts say that you know wind will look like Etc so um it really is about building um increased efficiency and accuracy and providing these tools to these domain experts who will hopefully you know be interested in using this but we never build in a silo this is why we always work with other partners to understand um like what they need you know what are the problems we can help solve we we can't just be doing that in a silo we need to be plugged into the community and that's where we start all of our conversations what can AI do in regards to climate and animals whether that's you know changes to migration patterns or just making the planet better for species other than our own yeah it's a really great question so in addition into mitigation adaptation and addressing loss and damage and climate science which we mentioned earlier um ecology and biodiversity is a huge area um that we can that we can use AI for one of our projects specifically serengetti is what we what we've called it and it's an ecological monitoring system where effectively we have worked with a local group to place motion activated camera traps across the serengetti that captures um obviously by by motion captures the animals that pass by the cameras and reduce the amount of time between capturing that data um and delivering it to the scientists that are analyzing you know to your point migration patterns indicator species that you know should or maybe shouldn't be in an area and trying to understand how climate change is affecting those those populations so ecological monitoring I think is incredibly important and um is something that we're working on as well so artificial intelligence can play a part in solving the climate crisis but it also requires is a vast amount of energy and the data centers that are powering this technology are very energy intensive so how can we expect the net effect to be positive when it comes to climate change it is absolutely true that AI is an energy intensive technology and like any energy in technology um until we have a grid that is run completely on clean energy those technologies will have a carbon footprint we approach this at Google deepmind in three different ways um the first is that we try to make our research as carbon efficient as possible so we look into how we can reduce the amount of compute that is required to train our models especially over time when we learn when we learn more about them um we also have developed a you know Carbon equivalent emissions dashboard that we give our researchers access to with levers that they can control that will reduce the impact um of their experiment so we find that you know offering that that knowledge um is incredibly important to driving action so is this like literally it looks like you're about to process such and such did you know this will use x number like when you book a flight it will tell you the carbon emissions of the flight and and say you know do you want to proceed it gives them yeah it gives them options around you know you hey you can move compute in space and time to a Greener grid you could choose to run this experiment in Norway right now because the grid is incredibly green versus in the UK where at the moment it is carbon intensive um and you know before having that sort of knowledge you know you just kind of go through your daily setup and you know you click go you run you know you run your experiment um and you just might not know that that's a lever that you can control and so we have surfaced that um and made it um available and and obvious to our researchers which is great because they all care about that and are choosing to adopt those Le levers for carbon efficiency um so that's kind of how one of the ways we look at our own house and our um when it comes to the research U we also work on tools that help others improve their Energy Efficiency so tools that can be used by external third parties almost every technology company out there has some kind of Net Zero commitment to kind of play their own part in combating climate change and clearly AI is going to come in provide all these Solutions do you think that opens up an opportunity to make some of those goals a little bit more ambitious I think we can always you know be more ambitious we're constantly improving uh learning what this technology can do and and raising the benchmarks one of the things that Google deepmind is devoted to right now is helping alphabet and Google itself achieve its one gigaton uh per year um reduction by 2030 um and we are working on things that will contribute to to that commitment which I think is you know an incredibly ambitious one um that we're really excited to be a part of so you mentioned reducing the amount of energy it takes to cool some of these data centers by 30% so with my Skeptics hat on I would say well that's that basically means Google spending 30% less cooling its facilities that's a net benefit to Google so how do you balance that and avoid the risk of greenwashing well I think that um with a lot of business when you look at this it is good to tie uh sustainable operations to the bottom line because businesses are then incentivized to adopt more sustainable practices so with energy this is an easy one because yes if you're saving carbon it means you're using less energy which means you pay less for it it's just kind of a win-win in that situation um to me it's not greenwashing because you are actually reducing the carbon equivalent emissions of what you're using and you are using your resources more efficiently how quickly do you expect some of these applications in climate change to actually make a difference we will move more quickly if we are more engaged in the conversation with domain experts if we are working with groups like CCI to open up data sets safely and responsibly um if we are talking with um deployment Partners who are willing to let us test in Real World Systems and I think going sector specific you know whether it's a forecasting tool that might take you know a couple of years to get ready or Fusion which the big joke is that it's always you know 10 50 years out um it that depends on the actual application but the way we move faster is continuing to work with all of these groups so we understand the problem we get access to the data um we're building tools that people will use and will find Value out of and that we have clear benchmarks for success for AI you know how much better do we need to be um to replace you know traditional systems and um help with efficiency and accuracy 10 20 30 years it depends on the application right um some of them it's some of them will you know be ready in a couple of months some of them are a couple of years some of them are decades it just depends on how big of a challenge you know you're going after and I think one of the things that we've really been surprised by is um the speed of of AI sometimes we expect it to take us a decade to solve a challenge and a couple years in we're like oh oh this is great we're making so much more progress so um it's really really hard to predict and I think my researchers would be very upset with me if I tried to put a timeline on it Sims with the spoon thank you so much for being here it's been absolutely fascinating no thank you so much for having me it's been fantastic Pria dant thank you so much for joining us so you're a professor at MIT and you're also the co-founder of climate change AI can you tell us a little bit about what that organization does yeah so climate change AI aims to kind of Empower a global community of innovators uh decision makers and practitioners to impactfully use AI to address climate change related issues so we provide educational opportunities networking opportunities grants programs as well as trying to set agendas and landscape analyses for the broader space about how AI can be used for climate action and how different players can can play a role in actually making that happen so it's kind of like a Brain Trust for collecting people who think about climate and AI to kind of come up with some solutions yeah so AI can be used in a bunch of different ways for climate action from forecasting solar and wind power to help us Balance power grids with more Renewables to helping us optimize heating and cooling systems in buildings to make them more efficient to helping us accelerate the discovery of Next Generation clean Technologies like batteries and where's where's The Cutting Edge right now is it coming is it Academia is it some of the big tech companies is it the the nonprofits NGS like who's doing the most impactful stuff here so there's a little coming from everywhere but when people think of AI they do tend to think of you know Academia OR tech companies but a lot of this Innovation is actually coming from you know entities that you wouldn't think of as tech companies like companies in the power sector companies in manufacturing NOS that think about forestry or NOS that think about flooding who are using AI techniques for their own problems and on their own data to actually address these climate change related problems so basically lots of entities that just are the owners of climate change related problems have really been doing a lot of cutting edge AI here can you give us an example just just on that what like a really good solid like this is a this is the example of that you said the forestry works something like that like where where that impact is being felt absolutely so for example the map project maap it's a a coalition of of nonprofits that are using satellite imagery to analyze and add um and identify deforestation in real time in the Amazon and by kind of doing this analysis they can then take actions to understand what does it actually take to stop deforestation because we now actually have a clear picture of where it's happening in real time um you also actually have intergovernmental organizations doing this kind of work so the UN satellite Center for example they also use satellite imagery to do real-time tracking of flooding in in Asia in order to understand on a kind of you know hour by hour basis you know what has happen what is happening with flooding and as a result what do they need to do to actually support regions that are being flooded at particular time so there like predictive analysis in this this is where the AI comes in presumably you can monitor more things at once and then also track the changes and predict where actions needed so it's more than just getting a report it's actually it's a bit more involved and that's where the AI comes in so in those two applications of deforestation tracking and flood tracking it is actually just the monitoring part the idea is that you have satellite imagery and it would be really hard for a human to analyze all of the images on you know very fast scale all the time so basically AI is allowing you to scale up what a human could do um and just kind of analyze more images in order to do that but there are definitely applications where AI itself is actually you know being being predictive so for example um uh open climate fix a nonprofit in the UK they're working with this uh UK system operator National Grid to actually come up with better realtime forecasts of solar power by basically analyzing how productive were the solar panels in the UK in the past based on the weather and time of day and then actually trying to kind of play that forward this flick had a uh a question that Nate and I talk about a lot across topics when we're looking at AI is how much of this has already been in the works because we had the generative AI boom really take off last fall with the onset of chat GPT everyone's excited but do you have a a pulse on what types of applications are new versus those that were pre chat GPT era what can you kind of tell us about the new stuff that's going on yeah so I'll say that actually a lot of the applications I mean all of the ones I talked about are pre-chat GPT era um AI refers to a really broad set of techniques that are not just generative AI or sort of chatbot related techniques and so there's really been a lot going on for a long time there are some you know advancements that generative AI can enable but interestingly in my opinion most of the more convincing ones are not actually based in text so they're not what people think of as you know chat GPT they are things like I gave this example of you know accelerating the experimentation for Batteries to reduce the time it takes to invent a battery you can do things like assessing past experiments to uh suggest new ones using generative AI or kind of assessing like you know past molecule configurations for various kinds of clean Technologies to S Suggest new molecules that's actually a really good use case of generative AI this is like Material Science exactly I wonder I mean we talk a lot about how wonderful AI is and how it's going to solve all the world's problems and give us new medicines and things but there is this underlying issue that it takes a lot of energy and I would love to get your view on whether you think enough good is coming out of AI to offset the fact that it takes a lot of power and energy and often natural resources to keep these machines going and getting bigger yeah data centers and like the global information communication technology Center are in that ballpark of you know two to 3% of global greenhouse gas emissions and so when you're thinking about are those emissions you know we first we have to bring every SE to Net Zero by 2050 that includes computation itself but also then when you're spending those emissions the right question as you're asking is is it is it worth it um and is it worth it yeah and I'd say that um on specific a specific application basis so when we think about like specific applications of you know helping to integrate Renewables into Power grids or helping optimize heating and cooling systems or you know pinpointing deforestation these kinds of applications I've talked about I would say in a lot of cases these are worth it what I'm more worried about is the fact that most of AI across Society is not that right we use AI for so many things that are potentially not worth the emissions like accelerating oil and gas exploration or targeted advertising or um you know it's used to spread misinformation on social media for the targeted advertising question you know just to extrapolate that forward in my head the problem the climate issue of on that side of things is the better the targeted advertising the more junk we're going to buy which means more shipping and transportation I think transport uses like what a quarter of makes a quarter of the world's Greenhouse house emissions or something like that uh and that means more plastic in the oceans and that's bad for the you know the natural world and and and things is that is that the link there or is it something absolutely that's the link and and and in general the the idea is that AI is an accelerator of the systems in which it's employed and as a result it's being you know employed more so in ways that drive you know consumer behavior that can be paid for by more powerful players in society than otherwise and so we as a society really in the same way we're needing to just realign everything in some sense to deal with climate change our use of AI is has to be captured in that we need to think about how do we align the use of AI like business as usual AI with with our climate change related goals what are some of the limits to AI as far as climate goes yeah so AI you know it's not a silver bullet so a lot of people think oh AI powerful technology so any hard problem I don't know how to solve is solved by AI that's not true for example AI is not going to make certain hard value judgments for us as a society I've definitely been asked by you know organizations and by policy makers can I you know put all the data from my country into Ai and run some kind of you know simulation to then figure out what the optimal policy decision is and in reality right there's no such thing our data encodes various kinds of value judgments we're explicitly making value judgments all the time throwing your data into Ai and hoping that that's not a value Laten out answer and also AI when it is a piece of the solution it's often not the whole solution so again we've talked about forecasting solar and wind power or assessing satellite imagery to track floods and if all you do is give the assessment to somebody and they don't do anything with it you haven't solved the climate problem every now and again somebody says climate change is a Lie the World is Flat aliens exist and are secretly controlling all World governments you know something in that camp do you think there's anything that is coming out of what you're seeing with AI as it relates to climate that might shift the needle slightly more in favor of Science and evidence you know honestly I think AI itself itself is having a very mixed effect on perception of evidence I think it can both be used to serve targeted information to People based on you know where their current state of knowledge is and what they need to know it's also being used to spread and generate misinformation on the internet the classic double-edged sword the classic double-edged sword and um I guess what I'd say is in my own work I think the kind of theory of change I ascribe to is that a lot of societal shifts come from not necessarily the majority right but like a very activated minority of people who kind of do a lot to like address issues that they care about and so in some sense my work is really about how do you get those people who are passionate about addressing climate change who have ai skills or who would like to kind of adopt AI into their own Industries for this purpose how do we activate them and help them work better together and really get those people and organizations to punch above their weight in affecting this kind of societal change I'm a big believer in in supporting the small changes yeah I think like a combination of small and big changes are are really important right you can't just do the small scale changes you have to do the systemic changes in order for this to work um and so people will often ask me right like I work in AI right like what climate problem should I work on and and what I say is you the the kind of best problems to work on are ones where you have a sense of the order of magnitude of impact like there is a big difference between you know like optimizing something for you know paper versus plastic straws versus like you know really like decarbonizing a building rising sea levels yeah like are you an optimist you pessimist what you what's what's the Outlook yeah so again I really am um an optimist but I think the thing that keeps me up at night is so while I've mentioned again that there more and more resources and talent going into the climate space making sure that we're actually um enabling people to do their best work that we're using money in ways that are actually effective is what keeps me up at Night free Dy thank you so much for joining us thanks so much here i r [Music] l [Music] [Applause] [Applause] [Music] oh

Channel: Bloomberg Originals

Views: 9,216

Keywords: News, bloomberg, quicktake, business, bloomberg quicktake, quicktake originals, bloomberg quicktake by bloomberg, documentary, mini documentary, mini doc, doc, us news, world news, finance, science

Id: Z-ZMcZq0DSU

Length: 24min 1sec (1441 seconds)

Published: Wed Dec 13 2023