AI - Climate Change friend or foe?
well it's a very good afternoon and welcome to this lecture on ai climate change friend or phone uh my name is gary reese and i'll be chairing this session the lecture is part of ucl's climate champion generation one but together we're the new generation taking responsibility for climate action and turning science into actionable ideas we welcome all of you joining generation one in a number of easy ways you can pledge your climate action finding out more about it at ucl dot ac dot uk forward slash generation one and choosing what action your pledge you can inspire others by sharing your pledge on social media please tweet from this event if you're enjoying it using the hashtag hashtag ucl generation1 and you can also check out our new generation 1 podcast series as details on the website if you'd like to take advantage of that now in this uh lecture we've got three very exciting speakers who i'll introduce uh in a short while we're going to have some time for questions at the end so please do get busy submitting those questions you can go to slido that's sli dot do and entering the event code which is ucl climate um we'll look at those questions at the end and i will put them to our three distinguished speakers who i would now like to welcome to introduce them each in turn um first we've got professor kate jones who is professor of ecology and biodiversity in the department of genetics evolution and environment which is part of the faculty of life sciences here at ucl professor jones is a world-leading ecologist and her work focuses on crossing disciplinary boundaries to address critical global challenges especially those that are at the interface of ecological and human health but she's made key advances in monitoring the status and trends in biodiversity and particularly in modelling and forecasting zoonotic disease outbreaks in human uh ebola sars and now of course kovid19 and this kind of modeling breaks down traditional barriers between ecology climate change and public health to inform global policy she's also a scientific advisor to uk's climate change committee which is an independent body that tracks the status of the uk government's progress towards those goals she will be our first speaker but i'll just introduce the other two um sir jeff melgen is our second distinguished uh speaker he is professor of collective intelligence public policy and social innovation here at ucl uh sir jeff was the chief executive officer of nesta which you may know as the uk's innovation foundation from 2011 to 2019. um earlier before that from 1997 to 2004 so jeff also had roles in uk government including being director of the government strategy unit and head of policy in the prime minister's office and then finally um we have aidan o'sullivan who is an associate professor in energy and ai at the ucl energy institute and a turing fellow at the alan turing institute before joining ucl aiden aiden was a postdoc in machine learning at mit and his research focuses on applications of ai that accelerate of the energy system he's programmed chair for the unesco international research center on ai climate change programme and the chief technology officer and co-founder of a very exciting ucl spin out called carbon re which is applying ai to a variety of foundational industries to improve energy efficiency so in part of all of you i'd like to welcome our three distinguished speakers we'll now get going to hear what they have to say and i'd like to introduce professor kate jones as our first speaker kate thanks very much guy and thanks very much for the invitation to to speak i'm just gonna share my slides and so i'm an ecologist really at heart and uh i really you know when i got asked to give this talk i really thought about the question and and and how it uh applied to my area of research and so i guess i'm thinking about it with a kind of ecological lens is um ai for helping climate change is that friend or foe and i would say in summary it's a friend with caveats so let me just talk through my um logic here so one of the ways that we're going to try and approach um the climate change challenge is with nature and not against nature and there's a huge movement of trying to think about how we use nature-based approaches to tackle climate change and that includes mitigation which is trying to lower our emissions or absorb carbon that we're producing but also adaptation so there's at least 1.2 degrees warming already committed possibly 1.5 by 2050
so there's a huge amount of climate change in the system that we cannot do anything about and so there's a huge role for uh nature-based solutions for adapting to climate change and so when i say nature-based solutions these are you know these have been defined i've just written on the slide here but they're they've been defined as as actions which you know are managing and restoring our natural systems to address these big societal challenges so for example they could be for mitigation for example it could be planting forests you could be planting broadleaf forests to absorb the carbon that's being emitted it could be a coniferous forest which is fast growing or it could be a broadleaf forest which it provides more habitats for biodiversity as well as for the carbon for the carbon's sequestration also it could be restoring peatlands which absorb huge absorber of carbon and when they're in bad condition they emit carbon so it's a really um important way of thinking about how to manage our climate goals and it's thought that like nature-based solutions are really key part of even meeting the mitigation targets that we have let alone adapting to the ones that we have so these are really important parts of tackling climate change however really big questions remain about what type of nature-based solution how do we monitor these ecosystems how do we monitor the carbon that's being absorbed or emitted how do we monitor their health and how are these healthy ecosystems are really important for nature for climate change and carbon capture so that's where i think ai can really be a friend is is trying to understand how we're meeting some of these targets and how we're implementing how healthy these ecosystems are oh hello so i'm just going to go through some examples and then just come up with some caveats that i talked about before about you know whether ai is a friend or foe so this is an example from from ai for earth from microsoft and they've got a number of of projects which are using ai enabled observation monitoring so this is a system for monitoring deforestation in the amazon which is a big carbon sink and it's and it's a missing carbon at the moment because it's it's kind of um being cut down or it's on fire so this is an ai enabled earth observation monitoring system where we're using data from satellites uh using some ai to then think about you know deforestation risk map so this is a kind of pipeline where you can then have these uh beautiful wrist maps which their policy makers can can use to monitor some of these targets so that's a kind of satellite-based monitoring system but there are lots and lots of things which are going on under the canopy or in uh non-forested areas which you really need to understand so how they're kind of communities change in response to degradation is really important and these aren't captured by uh satellites so this is a paper that was recently published by some colleagues of mine that looked at the intactness of the forest underneath the canopy and as species respond differently to intensities of human pressure as you can see on the right here some species love in you know degraded landscapes some hate it some of them tolerate it so that's all happening underneath the canopy and so just by monitoring by satellites it's not gonna it's not gonna help us understand uh how intact and healthy these ecosystems are so there's a huge um development in new technologies new sensor technologies uh for ground-based monitoring for biodiversity and for uh eco ecosystems in general and these are focused on two areas which i just like to spend a bit of time on firstly it's like images so camera trapping so these are cameras which are put out into the environment and then their images are collected and then you can use ai to use face recognition but for species so this is a one that we have for kenya and this is a obviously it's a giraffe but we can go through millions and millions and millions of these images uh automatically to produce the classifications that we need and that's how ai has been helping us do that but also another aspect of the ground-based monitoring system acoustic sensors is acoustic so by listening to the environment you can tell what species are present but also how healthy that ecosystem is and that can kind of go from uh being in terrestrial environments but also in the marine environment so coral corals make sounds and you can listen to corals for example to see how healthy they are and ai has been helping characterize uh the sounds that that's been made from these environments and these species and classify those so there's a real big classification um advantage that you have when you're using these ai systems you can also use them for soundscape whole sounds cable how a whole ecosystem monitoring and this is a project we've been doing in in borneo by looking at various um remnants of forest so either virgin and jungle but also different sizes of of the forest that's been remaining after um a company came in to plant do oil palm plantations and you can use ai to be able to use deep learning to um categorize these soundscapes in and they categorize very well to the kind of size of the remnant patches and the state of these ecosystems so that's a really cool way of trying to understand in a very broad way how healthy those ecosystems are which is actually really interesting so that kind of clustering and unsupervised learning for ai has been very important for us but also just to touch on how this can be used for anomaly detection as well so if you know what the sounds of the forest are you can tell if something bad has happened like a chainsaw so you can pick up uh as an early warning system these anomaly detections like chainsaws or gunshots if you know what the sounds what it normally sounds like so these have been very powerful in um helping us understand how to monitor these ecosystems so there's a huge number of of ai enabled monitoring that's going on at the moment i want to draw your attention to this i naturalist one over here and that's a really big citizen science project where people are taking photos all over the world and then these photos are of enabling us to understand how populations are declining in different habitats across the planet and those images are then being fed into a big ai which um has a in it's called imagenet which is going to be used to um identify species just from the images and so people can take pictures on their phone and then you get a species recognition algorithm identification so those are really interesting and really um it's really expanding and exploding i think that's really brilliant but i guess i've got a few caveats so these are my caveats and that a lot a lot of these systems really require that we have these huge amounts of data in order to work and also not just huge data sets but curated data sets so we need quite a lot of labels so we need these um tests of what things are so that the machines the algorithms can recognize what species they are and often those aren't possible and it's not possible to do that we need lots of cheap computing resources to actually process all of these algorithms and we need them scalable so that we can maybe put them on tiny machines or tiny sensors in the field um and often a lot of ecologists aren't computer scientists and don't know how to use any of these ai programs i'm not trying to do any ecologist down but you know you know you need little expert knowledge in order to deploy these i mean the practitioners will want to deploy these with little knowledge of the systems and so they need to be they need to be easier to use and then i guess one thing is about whether it's just accuracy you want so you just want the classification of the species and you don't really need to understand anything about the system and so if you trying to understand why it's being degraded or how to then make it predictive into the future you and maybe ai will need to be supplemented with further statistical modelling so mathematical modelling and forecast modelling to understand how you pin you put some of those ideas from the ai into a more mathematical predictive forecasting model so ai isn't enough on its own and it and it means that um it does it gets you so far but actually understanding it as a tool understanding it as a tool i think is really important and it's just one tool in your toolbox in order to understand how to monitor and manage ecosystems for these nature-based solutions to climate change so i just wanted to end on example from my own work on disease forecasting so this is just an example of what we were trying to think about how to put all of these pieces together in terms to forecast disease risk and ai plays a really important role in in trying to understand how species are distributed across the world but it unless you feed it into all these other parts of the system socioeconomics the vulnerability the exposure of people you won't be able to understand risk so that's that's what i want to say it's a friend but with caveats thank you kate thank you very much for a fascinating lecture i'm sure that would have sparked off lots of questions in people's minds and please do remember you can go to slido that's sli dot d o the website uh search on hashtag ucl generation one uh and pose your questions that will come on to at the end of uh the each of the talks um so we'll move on now and to hear from uh sir jeff malkin and jeff please take it away thanks very much and thank you kate for a fascinating talk i've learned a lot already i don't have any beautiful pictures i'm afraid to share but i will try and provide some angles on how we think about the friend and foe issues of ai and climate change my background is that i have a phd in telecoms but i spend more of my life working in policy and nearly 20 years ago helped on the uk's first climate change strategies which i like to think contributed to the uk doing slightly less badly than most of the other big industrialized countries and uh becoming a world leader in offshore wind uh came out of some of our work maybe 20 years ago um as kate said almost everything we know about climate change is a result of data and models of increasing uh sophistication and therefore should be part of the uh the solution but i think we're in a very uh ambiguous position at the moment on the one hand as we'll hear i'm sure from aiden in feels like energy there is huge potential for machine learning to optimize systems electricity is about a quarter of greenhouse gas emissions at the moment uh and ai is well tailored to cope with intermittent supply from renewables like wind and solar but also to help on the household site a few years ago as part of a project trying to use uh data and ai to make it easier for people to shift their when they did their washing into the an off-peak hours at 2 a.m so as to reduce loading which in turn greatly reduces the need for energy production capacity in transport in a way the store is even more striking imagine a city or a town with driverless cars and therefore without traffic jams cars idling the huge waste of energy we currently have in our in our our lorries and cars and so on and agriculture is also seeing extraordinary applications for ai to optimize uh planting pest control et cetera et cetera that's the good side the bad side which is i think often a bit invisible to people is the actual sheer carbon emissions associated with digital and ai at the moment about 3.7 of um emissions come from digital internet activity estimates of the carbon footprint of individual machine learning training are mind-boggling uh dozens of times the lifetime emissions of a car 60 times a transatlantic flight for for for some and although there's been progress in trying to optimize the server farms which lie behind our internet and their extraordinary use of ai some forecasts a recent one actually was suggested that by 2040 14 of all global emissions could come from uh computation if action uh isn't taken so that's where ai is both friend and foe of the climate so three sort of things which i think follow from that one is about business so to my mind one of the strange things of the last 10 or 20 years is how slow digital businesses have been to really take climate change seriously uh apple was notoriously uninterested in anything ecological and allowed this extraordinary mountain of e-waste to grow up as people discarded their iphones and ipads and so on but uh facebook and others were not much better amazon only very belatedly taking this stuff seriously so applying the brain power of those companies which employ now thousands and thousands of ai programmers applying that a bit to the climate is long long overdue and a bit of a contrast with the world of investment which um italy has still got a long way to go but for many years its leaders have been advocating shifting uh capital reporting investment reporting to take ecological issues more seriously i think the leaders of the digital world have been remiss and they were almost invisible in glasgow at cop26 yet again the second related point is in the public sector the teams doing climate change work have very little connection to the teams doing digital and data and commissioning ai and if you look at the city strategies for net zero or the national strategies they're still not joined up with the sort of techniques and mindsets which are common sense for people working in this space and finally i think there's a structural policy issue here which is really really important and vital that we get a handle on and this may be where we enter some controversy i think to achieve what is possible in 10 or 20 years particularly in cities of mobilizing data and ai to optimize our mobility our energy the logistics of our supermarkets and so on we can only do that if the data is linked if we can actually see the patterns suitably anonymized and then use all the different array of ai tools to to streamline to optimize to adjust the coordination of large-scale physical systems at the moment however nearly all of that data is proprietary it's held within electricity companies or uber or bus companies and so on and it's almost impossible to get hold of it let alone to link it up as a little experiment try and find out all the relevant data where you're living right now about what is happening to energy or transport or anything else you will find it's almost impossible it's locked up in silos now that was the business model of the 2000s and 2010s it's what made google facebook alibaba tencent and so many other companies very rich but i think it's become an anachronism and we will need new ways of pooling that data i think through various kinds of data trust with shared governance protections on privacy because you don't want your neighbor to know uh when you're not using energy for a week and uh you know a burglar might break into your house and many other privacy issues arise but so we can actually mobilize the full potential of ai as a friend of shifting to a net economy and society now with some way from doing that europe is beginning to move in that direction in very different ways uh china is with its city brain and other initiatives but there is a bit of a sort of mindset block because the companies spending the vast sums on ai r d at the moment obviously feel it's against their interest to pursue the structural policy changes needed for the world as a whole to reap the full rewards of um of ai and uh in all its potential so that's a very political uh aspect of this which we don't talk about very much but i think it's going to be essential to making progress on the as a way the everyday life the urban side of the things which kate was talking about i'll shut up there i think back to you thank you very much jeff lots of food for thought there in that slightly um interesting but controversial political area now i must apologize for misleading all of you in the audience last time around with a hashtag for the questions on slido it is in fact ucl climate i won't get used to all this hashtag stuff unless i read out my brief uh appropriately please do keep the questions coming we'll have some time for some conversation at the end and so that's slido and the hashtag ucl climate but last and by no means least um aidan o'sullivan um um aidan please take it away and tell us about um what you have to say thank you very much grant and uh thank you uh john for preempting me i think you'll probably find you're quite correct in your predictions um but yes i'm going to talk about it from uh i guess a positive perspective of what's you know why i think ai is such a critical enabling technology for dealing with climate change so as a kind of a recap of where we are so we've just completed cop26 um there was a reference to it that i quite liked which was our last best chance to take action on climate change and you know the positives coming from it is is there's definitely a measure of international consensus and the argument has shifted from you know is climate change real to you know what do we do about it which is is progress of a sort uh and if we were to you know again think about where we are in the world we were coming out of the covert crisis and what this showed with the kind of the push for the vaccine was you know what's possible through international collaboration through uh accelerated timelines during a kind of a a kind of a war mode or emergency mode you know we can make these big changes so uh the the kind of reduction in emissions that we saw during covert was about 2.3 gigatons due to essentially the world shutting down and you know that is this that gives a sense of the scale of what we need to do so if we look at the scenarios and the trajectory we're on uh we're on course to hit 59 gigatons of co2 which would push us into a 3.5
degree world and where we need to be you know as a kind of a minimum acceptable increase is a 1.5 degree world so this is a 25 gigatons of co2 world where we're emitting that so um the change that we need to increase are the change that we need to uh to implement is of the order of 34 gigatons and if the disruption of covert only produced you know two gigatons this gives a sense of the the scale that we need to move at and the the challenges we face so in order to you know do things at that level we need we need radical technologies and we need uh our solutions are going to have to have two key characteristics speed and scale we need to take action now so we need solutions that are able to deploy now and given that emissions have a cumulative effect action today is better than action tomorrow but we also need uh scale so our solutions need to scale to reduce emissions by gigatons and you know my kind of my view on this is that these are two characteristics that ai uh embodies as a software technology so um as a solution for what we need to achieve it seems to me to be a critical technology in in the mix so i work at the energy institute and one of the things that we're we're deeply interested in that we've been discussing for a long time is rates of technology change these are called learning rates and you know it's a measure of the decrease in price and something uh over the uh the kind of cumulative deployment so as we install more solar energy the price of solar drops as we install more offshore wind the price drops as we install more onshore winds the price drops as well however the learning rates of these technologies extrapolated out over a decade are of the order of kind of 34 16 and and 10 percent you know which is you know fantastic but not uh significant in the at the level that we need in order to you know have 34 gigatons of impact so while these technologies are very much the the core focus of solutions and decarbonizing electricity generation is on the critical path of our solutions for climate change we also need technologies that can scale at a level that exceeds this so if we are to compare um scaling in a.i um so one of the the most state-of-the-arts language models currently available as bert and the scaling of ai as a software technology is completely different so the ai community is generally an open source community we release developments we in order to keep accelerating it's it's heavily influenced by research and academia it's really emerged from there and you'll see a lot of the uh the leading figures in in industry also come from academia people like david silver at deepmind who's from ucl originally so again it has this kind of open source academic kind of mentality and hugging face is an open source community that is supporting state-of-the-art developments in language modeling and their technology has been downloaded 29 million times in a month now compare that to the deployment of solar technology there's no comparison in terms of the scale that you can reach with the software technology and you know to take the analogy a bit further um you know things that uh you know facebook took seven years to hit a billion users uh tick tock has taken four years to hit a billion billion users so software technologies inherently scale and this is why there's such a critical uh aspect of the conversation on climate change but what can you do with them so um in terms of you know accelerating you know deployment the speed of getting those things out is is is big but in terms of impact what's the potential for scale so i'm just going to talk about two uh different applications so again there are there are a multitude of different applications and um you know one of the great things about you know training students and working at universities you you constantly come up with students who have new ideas about ways that ai can be applied in the system to reduce emissions um but whatever we do the power sector is always going to be at the vanguard of decarbonization it's it's on the critical path and deploying renewable technologies there and decarbonizing electricity generation is fundamental to a lot of solutions uh the power sector has been you know the single largest source of emissions in the uk only recently eclipsed by transport um and the grade itself is described as the most complex machine ever built you know it consists of you know connections of miles miles of cables all interacting all operating with a commodity that can neither be stored nor nor rejected it has to be consumed at the point of generation so this is an incredibly challenging scenario and as we add renewables to it we're increasing the complexity of operation we've had 80 years to learn how to operate grids under dispatchable generation and you know we know how to do it very well with coal generation and gas generation we have to learn how to run grids with renewables while we're doing it we have to kind of build the plane while we're flying it so uh in that sense we need new tools that can help us in this and one of the research projects that ucl was involved in helping with was uh learning to run a power network which is a challenge at europe's which is the largest ai conference in the world challenging ai teams around the world to develop an autonomous agent using reinforcement learning to operate a grid under a range of scenarios and the winner was whoever could keep the grid from falling over the longest as we kind of poked and prodded it with different scenarios so it was one by a team from baidu and it shows kind of how uh big tech can kind of interact with you know a power sector to develop new tools that help us uh you know adapt and move forward with climate change so the kind of the level of uh the level of improvement that's available within the grid is is significant you know one percent improvement in operational efficiency correlates to millions of tons of co2 which is what we need so and again if you think back to the hooking face example you know once that's developed for the uk or for germany who's to say it can't be downloaded and scaled for south africa or australia or places like this once the software technology is developed so again the opportunity to have a large impact and then to quickly deploy and accelerate its adoption all over the world are the key characteristics of as a software technology and again focusing on scale so if you were to look at at areas where it's possible to have a big impact on emissions uh the energy-intensive industries you know loom front and center after power generation and transport as one of the areas where you know big innovation is needed but there's also big potential for a reduction of emissions so the scale of opportunity is on the gigatons level so energy-intensive manufacturing of things like steel cement they need high-grade energy from fossil fuels and to reach the temperatures that are needed to melt the kind of materials down and they're responsible for 20 of global emissions so as a single point where innovation can happen they're a really attractive opportunity and they're one that's uh we have focus that's focused on at uh carbon re a startup that's been launched out of ucl uh and cambridge so to give a bit of an example of what we're doing uh the opportunity in this space that we see is down to the variability in energy intensity of manufacturers so plants just like everybody have good days and bad days so one day they're operating at a really high energy intensity level and the next day they find a way to run production at a much lower energy intensity level and the numbers you're looking at here are gigajoules of energy per tonne of cement and a plant typically produces about 4 000 tons of cement a day so the energy consumption is huge now uh if we look at the variability over time and if we look at that ranked we can see if you can move from your worst uh death style performance towards your best decide performance the scale of emissions reductions is huge and what if we could make every day a good day so that's what we're doing at carbonread we've developed a software product that we're using to decarbonize the foundation industries by helping them in operational efficiencies and helping them to reduce the energy consumption and bring it more like what they're doing on a regular basis make their bad days look like they're good days make their good days like like their best days so again this is something that's been spun out of ucl and i'm really proud to be involved in as a kind of application of research uh having impacts we've raised one million in investments and we've got five pilot projects happening around the world um and again we believe the opportunity is in the gigatons for impact with 20 reduction in fuel derived emissions and savings per plant so again the solution is all based around artificial intelligence uh what we do is we have a digital twin of the uh factory in this case a cement factory uh where the data i've shown you is taken from and what we do is develop a recommendation engine so we have an ai that learns what the differences are between the good and bad day performances are and identifies the actions that need to be taken in order to move those bad days closer to the good days and it works with the human operator to give them advice as to how they can achieve this so it's a full human in the loop solution leveraging artificial intelligence to have impacts today so just to conclude with a few ideas and again i would kind of uh you know as you might expect i'm quite bullish on the opportunities for ai and tackling climate change um however we also need a lot more people so again uh at the intersection of people who know a lot about ai and the intersection of people who are working on climate change is very small at the moment we need a lot more people focused on this and you know there are lots of reasons why we talk about responsible ai but we should also talk about responsible ai researchers you know is it a responsible use of kind of your knowledge to go and optimize the deployment of ads or is it a more responsible use of your hard work and energy in you know attacking the biggest problem in the world and again a lot of ai has developed with specific applications in mind as we look to tackle other problems there'll be new directions for ai around causal reasoning and detectable ai model explainability that will help us deploy ai in the energy sector a lot more and again there's also the idea of redeveloping the energy system around ai so we're looking right now to see what we can change and what we can do quickly but what if we were to re-engineer the entire system around the characteristics of a software solution that never gets tired can react at a scale far exceeding uh human operation and you know do things that you know so far we've never been able to do in the past so yeah uh thank you very much i think i ran slightly over time but happy to take any questions now thank you very much aidan that was fascinating as well so thanks to all three speakers for some um really provocative and interesting ideas and thank you for those of you who submitted questions and it's not too late we're going to go into the conversation now if you want to submit a question please go to slido that's sli dot do and search for your hashtag ucl climate um so let me get the conversation rolling and perhaps with you aiden derek young has submitted a question saying is the main contribution of ai towards climate change mitigation just reducing or eliminating wasteful inefficiencies in energy intensive processes so i think we're starting to get into that you know shouldn't we be reducing energy as well as mitigating but by uh reducing inefficiencies what's your kind of thoughts on that that's a that's a really great question and um there's a lot of kind of discussion around the right kind of fuels uh in order to you know meet climate change or to mitigate climate change it definitely needs a mix of technologies but i'm of the view that efficiency is the best fuel you know not hydrogen not uh kind of nuclear uh efficiency so the you know the the watts you don't consume should be counted as kind of megawatts in some sense and reducing this is you know uh you know massively impactful so doing things better and it's it's again it's it's down to the the kind of characteristics of a software solution it doesn't require a capital investment you might say the ideal solution is for everyone to operate on hydrogen well there are three thousand cement plants in the world how long do you think it takes to you know convert them all to hydrogen it would take a a long time years and years so and again where does the source come from so in terms of having action immediately definitely reducing energy consumption is the the main way to have uh quick action thanks and let's come on to k then because you know i couldn't help but noticing in your presentation on the one hand we're monitoring amazon deforestation but on the other hand we're doing it by putting satellites on rockets into space and last time i looked rockets weren't the most carbon neutral technology so how do you think in that biodiversity space about balancing the energy consumption of the benefits yeah i think that's a it's a really interesting question i i really do think that um the the scale of the kind of monitoring and well not just monitoring declines but restoring habitats is so huge that it does counteract the kind of carbon that was taken to put it put it into space and those weren't put into space for deforestation monitoring they were put in for more commercial uses we're just using them for that so i guess i guess what i'm seeing is that there's this human in the loop as you know as alien was saying human in the loop smart sensors that kind of optimize can optimize our understanding of the habitats and actually what actions are successful in increasing biodiversity and increasing health or storing carbon and i think that's that's really important so it's not just about monitoring your decline but it's managing your system better and i think we're so behind in the ecology in the agriculture sector in the forestry sector we're like totally way back to compared to just what i've just seen from aidan you know and monitoring you know efficiency of abuses but we have got nothing like that cyber-enabled systems nothing at all across the planet so you're seeing a conceptual similarity there between his energy intensive processes maximizing efficiency and megawatts you're seeing that as a potential future for managing biodiversity and managing ecosystems yeah i wouldn't say biodiversity i would say ecosystems because it's just much more than the function of which you know cute cuddly species you want to to manage it's more about having functional ecosystems which can store carbon are resilient to you know massive climatic effects that can support our health and well-being it's much more about particular species it's more about the the earth system that that we need to be supported by great let let me bring um jeff in then i mean we we're talking essentially about that uh balance of incentives between energy consumption and energy efficiency and i guess one of the surprising things when you told us about your policy initiatives you didn't say carb attacks or anything like that what's your view on something like an incentive like that to help us reduce our carbon emissions well where we have achieved fundamental change as we've done a waste in the last 20 or 30 years it's been a combination of things it is rules and laws and taxes sometimes which have forced industries to recycle 50 60 70 percent of glass paper plastic etc plus behavior change people themselves deciding to change how they lived and what mattered to them and market competition to drive new technologies and it's exactly the same i think on everything in relation to carbon uh yeah 15 20 years ago we worked on carbon taxes and in australia i was part of the team the prime minister who introduced them with a huge backlash so they've been sort of on the agenda for two decades without really yet the political will to make them part of the uh of the package and that's why the sort of things aiden was talking about are so important at least you can get on with them and there's direct bottom line benefit to the electricity company or the cement company and so on can i just raise two other things which i think maybe link all of our all of our commentaries so as i think aiden said you know the best minds of the last generation you know jeff hammerbacher of facebook said were devoted to click through advertising but i think we we could use a whole generation you ai tools to prompt us to think or behave in different ways fast fashion is a good example less than one percent recycled now you know when you're making a purchase can you be prompted towards a less so carbon destructive uh clothing option we're seeing this in science where the productivity in r d has been in a long term decline for decades but ai is being used to prompt new kinds of combinations the the things we know in protein with which deep mind and others have worked on but also for scientists working in other fields to think of where there might be a connection with other knowledge how to speed up the productivity effectiveness of of the the very core activities of science and i think these are quite exciting they're underdeveloped uh as yet but they're the equivalents to what has been done so successfully by the social media companies and then very finally on kate's point one thing which was briefly mentioned in cop26 is the idea that in future capital assets will be linked to objective monitoring of the state of deforestation or below canopy behaviors and that's a very quite an interesting new version of capitalism which actually doesn't go through complicated esg reporting but actually is more objective in rewarding uh incentivizing the right times of actions on biodiversity fascinating really interesting um and thank you very much have any other other two panelists scott's got any responses to jeff on that or shall we move on yeah i i guess i would come in about value chains and supply chains i think that's a really important um point that you met well you made lots of important points but i think i'd like to pick up on that one just because we i think there's a huge market for that because there's a huge appetite for having sustainable goods and now how using kind of new technologies to be able to monitor where your where the supply was how how effective you know local governance is local community involvement how how that's being managed your local resources are being managed and that can be tracked back right to your you know new cardigan or whatever i think that's really important and i think that that that's a really really excellent use of these kind of value chains through the system i mean jeff i can't i can't get away from the nagging suspicion this is all about encouraging humans to do more and there can be negative consequences of that as well so uh your beautiful vision of a land of driverless cars and and and that was going to be great we can equally think i've had people point out to me that rather a handy thing about having a driverless car is you don't need a parking space because you can just have it drive round and round and round while you're in the climate change lecture and pick you up at the front entrance when you're finished which of course would result in lots more congestion lots more energy consumption so i guess my question is isn't the key at the end of the day human behavior and persuading people to do less to not buy the cardigan at all that's definitely part of it and i'm sure we all everyone watching this cool probably buys lots of things we don't need and i you know one one hope is there will be just a profound culture shift away from waste and excess consumption uh including at the very top one percent or top 0.1 who often own 10 homes they
only use a few days a year and probably own 10 cars most of which they don't use at all we have we have a chronic problem of deep waste built into our economy at every level but i think in terms of driverless cars i don't think that is about behavior choices i think that is more about designing the architecture of daily life and it's true and we found this with ubers that they may all drive around waiting for lifts but in that vision of a future city you probably won't need huge multi-story car parks with huge spending on cement to build them you won't have shopping malls in quite the form you would you did in the past you won't have parking spaces all over the place there's all sorts of resources can be liberated by that ai enabled transport future the key is to look at the net effects because they often are as you say a bit surprising and they have been in the past things you save in one part of your consumption patterns you then reallocate the spending to something more carbon intensive and this is where intelligent analysis of net carbon effects is so important because it is quite counterintuitive often uh what really happens and i think nowhere more than in the cloud which we are all part of at the moment so i think most people still think has no real footprint they don't realize it's based on vast server farms enormous mining and minerals and the kind of kate crawford in her brilliant uh recent book on the atlas of ai shows just how much this is a very material technology even it appears immaterial fascinating let me pick up aiden on another aspect of human behavior you touched upon where i think you're exhausting people to not go and necessarily work for solutions that would uh serve more adverts to us so we bought more cardigans um but but you were very much a proponent of working on um presumably socially useful um things how would you encourage that beyond just saying that's a good thing are there actual is there actually anything you've identified that we could do to encourage those kind of behaviors i think the the community is kind of waking up to that i think a little bit as as an option and i think i i guess the challenge is a little bit that um you know tech companies offer such attractive salaries and such big perks you know it was definitely you know the last decade it was the case of if you were working at google that was like the best place to work and i kind of feel like that is kind of changing um more generally now and how things are viewed and their impact on the world you know it was kind of uh that the bigger picture is being considered and hopefully that will uh peter through into kind of the ai community and people you know start to look at what they would do or find more meaningful and i think it's been you know maybe an outcome of the pandemic as well you know of people you know reevaluating uh the choices they've made the jobs they're doing you know thinking about other ways of working and thinking about other things to work on and you know what what gives kind of uh satisfaction and a sense of achievement to them um and again you know you have to provide viable alternatives it's no good just to say um you know this would be good and kind of walk away at carbonread we do want to bring in the best ai and machine learning talent that we can find and give them the opportunity to have gigatons of impact on on emissions so um that's kind of what we're doing i think the investment in climate tech at the moment that we're seeing shows that you know there's a real appetite for you know investors and venture capital firms for these kind of companies and again i i think yeah coming back to give google a hard time um you know this they had this mantra of you know don't be evil um that's no longer good enough you know that's kind of you know now that's like obvious um before it was kind of a groundbreaking thing to say you know what a what a great company you know don't be evil you know do good things but now it's kind of a case that's expected and you know what how what can you do that's positive not just you know don't don't be negative thanks that's that's really full of uh some really interesting points i'm sure will fascinate everyone who's listening um i i want to move on a little bit we've got about 10 minutes to go uh and tackle a couple of areas and one is going back jeff to your um point which which actually resonates what aiden just said about big companies that they lock up a lot of data and we have a question should there be a requirement for ai data to be open source to ensure transparency and as a way to corporate social responsibility for companies that might be a little more reluctant so should should we be mandating that or perhaps to pose it as a question jeff you know how how would we persuade uber and the bus companies and everyone you're talking about to open up their data and create these data trusts i don't think we'll persuade them because i don't think in the current market dynamic it's really in their interest to do so every past industrial revolution required new rules of the game new laws new regulations and so on and on this the one we're entering the fourth industrial revolution is no different my preference would be that any company running essentially essential infrastructures in energy transport but also actually the main platforms should have requirements to open up their data through apis of some kind with anonymization or privacy enabling absolutely crucial in order to enable the kind of coordination and planning at a city level or national level which will be so essential for uh for achieving net zero and there are some precedents versions of this have been done in banking the uk was a pioneer of open banking data allowing us to sort of take our banking data give it to a third party the european commission is looking at something quite similar as part of its ai rules now even five years ago this was all unthinkable i published various things five ten years ago about this and they were deemed utterly you know on the margins but they're not quickly becoming mainstream and i think even well google actually is microsoft both realize i think that is this is the direction of travel and they probably can make plenty of money in a new environment but will probably take us a few years to get there what we're still missing is the good enough designs of the guardians of that data the same is true in health we saw this last year around covet although we moved quite quickly to free up some of the health data we don't lack we lack rather the trusted intermediaries who we can feel confident will protect that data protect our privacy but also maximize the value to us of linking that with other data sets to get to better cures or better solutions to climate change and this is i think one of the great challenges of where political science meets technology interesting i like that idea of open banking i was thinking about open energy where we take our personal carbon footprint with us as we change providers and just to follow up though i mean i i agree that trust is not perhaps at a high level in our societies either with some companies or many companies we might say but also to some extent with governments and there is a lot of frustration among citizens about that so what steps do we need to do to build those trusted intermediaries because it's it's an urgent task if i've understood you correctly i i don't think it can be the government we're never going to trust our trust ministers and it can't be the companies on their own especially after facebook in the last two or three years so again this has happened many times through history where there's been a need to create intermediaries third-party bodies which have a really strong ethos and competence we have them around central banks we have them in our regulators we have the embodies like the bbc there are lots and lots of precedence of trading institutions designed to be trust building in in is neither the government nor the market and we just need a whole family of those now for data and ai i think good and aiden just turning to you we were talking about open data here and open energy you work in the energy industry is that is that something you see as an opportunity really to be much more open about energy usage and and along the lines jeff says or are there some barriers in your industry or in your industries where you do your academic work the yeah the energy sector has been transformed by digital digitalization over the last few years there's been a number of reports energy data task force things like this instituted by ofgem and it's interesting how they've gone about it i think they've gone about it in quite a nice way in that they've tried to benchmark companies against each other so every district network operator who manage the kind of the local grid they have to publish a digitalization strategy which you know they can then evaluate compare score and say well the best people for in this the best report came from western power and the worst was from you know ukpn and we're going to put pressure on ukpn to do better for example and that's a nice way and i would like to see something like this again um you know regulation needs to have teeth and there needs to be kind of sufficient powers for regulatory bodies to have impact in the space and that can be quite challenging in when technology moves quickly and it's hard for regulators to keep up and i think engaging with academia is is one way for them to really you know even that out and i think you know it's something that we would probably like to see more of rather than academics working for big tech companies have them work with regulators to keep them at the forefront of technology and to design the right policies and right structures around having impact very very interesting now we're moving into the last five minutes or so of our discussion and i just wanted to tackle one final area which is it it strikes me that um uh and maybe we'll come to uk that ai is very much a technology of developed economies and here we are sitting in a developed economies but but we also hear at least i hear uh from cox 26 about marginalized communities about the global south and about the applicability um of these technologies in in that area so i think my question is how do we uh thinking again about slido how do we ensure we include communities on the front line of climate change and how we ensure that the ai isn't just a developed economy technology that is is being used to deliver unsustainable solutions kate i wonder what thoughts you have about that yeah i mean i did see that at the cop and i think that's a really a welcome development that that got some air time in talking about the targets and inclusion especially for the deforestation target including indigenous groups and that was really great um i think there's a an actual positive role i think for um local indigenous communities in ai i think there's a lot of tools which can make their knowledge really useful and so they can use those tools to manage their own resources and i think that's that's really you know there's new technologies that they can use but also i think there's a kind of training gap which needs to be bridged so i think that there's there's there's going to be the need for more investment in those kind of training opportunities for those groups but i also think there's a huge opportunity for them they can leapfrog a lot of these technologies and you know if they can get trained in those those areas there's a huge opportunity for them to be at the forefront of some of those solutions for tackling climate change that's a really interesting possibility reminds me of the growth of mobile banking across many many areas of sub-saharan africa aidan looks like you want to come in here yeah just quickly um i mean if you think of the contribution of the industrial world to uh climate change you know we've we've been the ones driving it you know it's the cumulative emissions from the uk america you know they dwarf the rest of the world and in some sense ai is an opportunity to give back and to kind of you know distribute technology that you know we've been developing and that has been largely developed here and it's easy to hand over to other countries for example you know the best solar power forecasting algorithms should be distributed you know from national grid to colombia to you know south africa to places like this and to let them progress and to let them have benefit from the advantages and the research that we've invested in and reduce their emissions um and yeah i'm afraid i have to jump off now but thank you very much thank you very much and jeff maybe a final thought from you yeah so one of the projects we're doing at ucl at the moment is with the un and ukri is mapping global r d and how well it or badly it aligns with the sustainable development goals and that of course shows enormous imbalances between the poorest countries and the rich between the biggest companies even and governments and the huge imbalances in terms of where the brain power and the money goes we're hoping to turn this into a more permanent kind of observatory so at least we can know where ai research money is going at the moment that is very opaque very little discuss very little accountability but also to help promote exactly the sort of things kate was talking about these flowering fields where citizens science collective intelligence plus sensors plus ai come together to transform the patterns of power in feels like biodiversity and ipcc is always also beginning to tap into indigenous input so things are moving in the right direction but the imbalances are enormous when you actually look at the numbers very sobering right i think sadly our our time has come to an end in this lecture but i hope um you've got a sense from our three panelists of some of the things we're thinking about at ucl some of the things we're doing at ucl and i would like to thank kate jeff and aiden for their wide-ranging knowledge and expertise they applied in this discussion and also i think for the sense of optimism combined with realism um i think in this very difficult and challenging but extremely important issue if you liked today's lecture you may be interested in the next lecture on tuesday the 23rd of november also 1 to 2 pm and that will be on health and climate change a very important topic and we'll be addressing how can tackling climate change improve your health question mark but for now thank you again to kate jeff and aiden our speakers thank you very much to the audience for listening and we'll see you again on a lunch hour lecture near you soon goodbye from me