Networking Session #50 AI and Environment: Sustainable Development | IGF 2023

Yoshiki YAMAGATA

Professor Yamagata is at the forefront of designing urban systems to enhance resilience in the face of climate change. His team harnesses the power of the Internet of Things (IoT), big data, and artificial intelligence (AI) technologies to achieve this goal. They have focused their research on studying the Tokyo city center and its surrounding areas.

Using IoT, big data, and AI technologies, Professor Yamagata’s team aims to comprehensively understand urban emissions and develop sustainable strategies for policymakers and building owners. They employ machine learning techniques to estimate dynamic carbon mapping and portray emissions resulting from various urban activities. This approach utilizes abundant sources of data such as occupancy information, people’s mobility patterns within buildings, sensor data, and transport measurements.

Professor Yamagata emphasizes the significance of being prepared and implementing preventive measures to mitigate the risks posed by heatwaves. By combining hazard maps with precise location information of workers, the team can accurately assess exposure levels to heatwave risks. In areas identified as high-risk, they can deploy sufficient ambulances in advance to potentially save lives of those vulnerable to heat-related illnesses.

Another crucial aspect of Professor Yamagata’s work is his belief in enhancing walkability in cities to promote the health and well-being of citizens. By utilizing big data and AI, his team can analyze walking behavior in cities, identifying ways to improve the flow of people and enhance the overall health and well-being of urban residents.

The team also recognizes the importance of visualizations as a tool to aid in understanding sustainable urban systems. These visualizations are being developed collaboratively, involving stakeholders such as policymakers. Policymakers are particularly keen to see policy options directly in these visualizations, requiring granular details regarding different options such as energy management, urban planning, and digitalization. Therefore, involving policymakers in the application of AI technologies is crucial to address their specific needs.

Additionally, involving policymakers in the use of AI is a key research question for Professor Yamagata’s team. Understanding the benefits that systems can provide to users is another important consideration. If users cannot perceive the advantages, privacy concerns may arise. Therefore, it is crucial to ensure that users clearly see and appreciate the benefits of these systems.

In summary, Professor Yamagata’s work focuses on designing urban systems that are resilient to climate change. Utilizing IoT, big data, and AI technologies, his team conducts research on understanding urban emissions, developing strategies for policymakers and building owners, addressing heatwave risks, promoting walkability, and visualizing sustainable urban systems. The involvement of stakeholders, including policymakers, is necessary for successful implementation, and it is important to ensure that users perceive the benefits of these systems without privacy concerns.

Audience

During the discussion, participants noted issues with the plug unexpectedly turning off, causing confusion. This raised concerns as the device should not turn off without the plug, creating uncertainty about its status and available positions.

Importantly, the value of having a teacher physically present in the classroom was discussed. The presence of a teacher enhances the learning experience and promotes better interaction with students, emphasizing the importance of in-person teaching alongside online platforms.

Previous online meetings and events, including a webinar on blockchain, were also mentioned. Participants recalled attending various events organized by the Council but noted their absence from a specific event. These events provide opportunities for knowledge exchange and networking.

Additionally, it was noted that one of the panelists was removed from the discussion. The inclusion of a video sent by a participant indicated the sharing of multimedia content during the conversation.

In conclusion, the discussion focused on technical issues with the plug, the significance of face-to-face teaching, previous online events, and the incorporation of multimedia content. Gratitude and appreciation were expressed at the conclusion of the discussion.

Peter CLUTTON BROCK

AI and data science have demonstrated their potential to be key enablers in the global transition to achieving net zero emissions. Several notable examples highlight the positive impact of AI in various areas related to climate action.

One such example is DeepMind’s collaboration with Google, where AI was employed to significantly increase the energy efficiency of Google’s data centres. Through AI techniques, DeepMind managed to enhance the energy efficiency of these facilities by an impressive 30-40%. This advancement is significant as data centres are known to consume vast amounts of energy, and optimizing their efficiency can lead to substantial reductions in greenhouse gas emissions.

Another remarkable application of AI can be seen through the efforts of the Climate Trace Coalition. By utilising AI and satellite imagery, they were able to enhance the accuracy of global emissions inventories. This improvement is crucial in our collective efforts to effectively monitor and manage greenhouse gas emissions, enabling better decision-making and targeted interventions.

Furthermore, Unisat’s Flood AI tool has contributed to improving disaster response in Asia and Africa. By leveraging AI, this tool has enhanced the ability to predict and respond to floods, ultimately aiding in mitigating the devastating impacts of such natural disasters. This application of AI demonstrates its potential to assist in building resilience and safeguarding vulnerable communities against the effects of climate change.

Despite the promising opportunities AI and data science offer, there are challenges that need to be addressed for their wider application. The two main frustrations hindering progress are data discovery and data access. The process of discovering relevant data and accessing it efficiently can be cumbersome and time-consuming, impeding the adoption and effectiveness of AI and data science solutions.

To overcome these frustrations, several strategies are proposed. Firstly, the development of improved data discovery tools is crucial for facilitating easier access to relevant datasets. Additionally, better regulation is needed to ensure that data is appropriately shared, while still protecting privacy and maintaining security. Furthermore, the establishment of commercial data markets, coupled with financial incentives, can encourage companies to share their data, unleashing its potential for AI-driven solutions.

The Centre for AI and Climate is actively working towards developing an intelligent data catalogue specifically tailored for climate action. Their efforts align with the need for a more organised approach to data discovery and accessibility, providing a consolidated platform for researchers, policymakers, and organisations to access and utilise relevant climate data.

In addition to supporting climate action, AI is expected to play a significant role in digitally managed energy systems. It has the potential to optimise investment decisions for asset developers, ensuring efficient allocation of resources towards sustainable energy infrastructure. Moreover, electricity networks can leverage AI to make informed decisions regarding which energy sources can connect to the grid and what upgrades are necessary, thus improving the overall efficiency and reliability of energy systems.

However, it is essential to maintain a balance between automation and democratic input in these digitally managed systems. While the increased use of AI may lead to a more automated electricity system, human control and democratic participation remain crucial for accountability and fairness. By involving stakeholders and ensuring democratic input, it becomes feasible to limit the level of automation and prevent potential negative consequences.

In summary, AI and data science have demonstrated the potential to significantly advance efforts towards achieving net zero emissions. Various examples showcase the positive impact of AI, from enhancing energy efficiency in data centres to improving disaster response and enhancing the accuracy of emissions inventories. However, addressing challenges related to data discovery and data access is crucial to unlocking the full potential of AI. With improved regulation, commercial data markets, and the development of intelligent data catalog solutions, AI can be effectively utilised in climate action and digitally managed energy systems.

Jerry SHEEHAN

AI systems have the potential to enable sustainability and transform climate modeling, according to one of the speakers. They argue that tools like carbon-aware computing can shift compute tasks to data centres with higher availability of carbon-free energy. Additionally, they highlight the Climate Trace project, which harnesses AI to track greenhouse gas emissions. These examples demonstrate how AI can contribute to addressing environmental issues and promoting sustainability.

However, another speaker raises concerns about the increasing computing needs of AI systems and their potential environmental impacts. They explain that direct environmental impacts result from AI compute, along with the resource’s life cycle. Furthermore, they point out that indirect impacts may arise from AI applications, which can lead to unsustainable consumption patterns. This argument suggests that as AI becomes more prevalent, it could exacerbate environmental challenges.

In response to the potential environmental impacts of AI, another speaker emphasises the need for common measurement standards and expanded data collection. They argue that without comprehensive data and consistent measurement frameworks, it is difficult to track and analyse the environmental impact of AI effectively. This highlights the importance of developing robust methods to assess the environmental implications of AI technologies.

The role of international organisations, such as the OECD, is highlighted by one speaker in facilitating cooperation on AI and climate change. They argue that these organisations serve as the connective tissue that brings countries together to tackle complex issues that transcend borders. By fostering collaboration and knowledge-sharing, international organisations can play a critical role in addressing the global challenges posed by AI and climate change.

AI’s potential contributions to various sectors, including the environment, agriculture, and healthcare, are recognised by one of the speakers. They explain that AI is a general-purpose technology with broad applications, and its diffusion is increasing across different countries in various sectors. This highlights the versatility and potential positive impact of AI on multiple industries.

The concerns regarding the negative impacts and risks of AI are acknowledged, but there is a belief that breakthroughs enabled by AI can help save the planet. Despite the potential drawbacks, the positive practical applications of AI are highlighted by one speaker. They suggest that while it is important to address the environmental impacts and risks of AI, it should not overshadow the potential benefits it can offer in addressing global challenges.

To address the challenges associated with measuring and understanding the environmental impacts of AI, one speaker proposes the establishment of measurement frameworks. They argue that as AI scales up and is applied on a larger scale, it becomes crucial to have standardised methods to assess and evaluate its effects accurately. This suggests a proactive approach to addressing potential negative impacts through robust measurement practices.

Adhering to the principles-based approach of the OECD is advocated by one of the speakers as a way to responsibly implement AI. They emphasize principles such as transparency, engagement, and a human-centred approach to ensure that AI technologies are developed and deployed ethically and in alignment with societal values. This underscores the importance of ensuring the responsible and accountable use of AI.

Finally, the importance of public involvement and understanding of the benefits and risks of AI is highlighted in the policy-making and system development process. One speaker advocates for the integration of public input and transparent parameters into AI-related decisions. This suggests that inclusive and participatory approaches can help address concerns and build trust in AI technologies.

In conclusion, the different perspectives presented in the summary demonstrate the complex relationship between AI and the environment. While AI systems have the potential to enable sustainability and contribute to various sectors, concerns about their environmental impacts and risks should be addressed. Common measurement standards, international cooperation, and responsible implementation are crucial in harnessing the potential of AI to address global challenges such as climate change. Public involvement and understanding are also important in shaping AI policies and systems.

Patrick

The workshop focused on the relationship between artificial intelligence (AI) and the environment, with speakers highlighting various aspects and potential benefits. One key point discussed was the use of AI in preserving healthy ecosystems. Efficient energy management was identified as an area where AI-based systems have been successfully implemented, citing the example of Switzerland using AI to manage the capacity of public transport and discourage overloading. Real-time data on energy production and consumption was also mentioned as a crucial tool for dealing with the effects of climate change and managing energy resources more efficiently. This application of AI in energy management was seen as a way to improve environments.

Another important aspect was the responsible use of AI to serve its purpose in preserving the environment. The speakers emphasized the need to ensure that AI tools are used in line with their intended purpose and argued that AI should be applied responsibly to help preserve healthy ecosystems. This sentiment was supported by the idea that every human right ultimately depends on a healthy biosphere, and AI could be a helpful tool in achieving this goal.

The workshop also emphasized the significance of international cooperation and the sharing of best practices for achieving environmental sustainability. The speakers stressed the importance of collaboration and the need to share knowledge and expertise on AI’s impact on the environment. For instance, the Council of Europe was mentioned as working with international organizations like the OECD to study the impact of AI in sustainable urban systems. The speakers highlighted the importance of data analysis to track and analyze the environmental impact of AI, as well as the need for common measurement standards to ensure comparability.

Furthermore, the speakers acknowledged the potential benefits of AI in supporting the green transition and addressing climate change. They mentioned that AI can be applied to research across numerous disciplines, aiding the transition to a greener world. Examples were given of AI being used in fields like environmental impact, transportation, and material science. The positive sentiment towards AI’s potential in supporting the green transition was evident throughout the discussion.

In conclusion, the workshop provided valuable insights into the connection between AI and the environment. The responsible use of AI to preserve healthy ecosystems, the importance of international cooperation, and the potential benefits of AI in supporting the green transition were all key takeaways. The speakers expressed a positive sentiment towards the potential of AI in addressing climate change and achieving environmental sustainability.

David ERAY

Artificial Intelligence (AI) technologies have the potential to significantly contribute to creating greener cities and regions by optimizing energy usage, handling power fluctuations, improving energy storage, and predicting energy demand. By analyzing complex and multifaceted datasets, including real-time data on energy consumption, water use, and weather, AI systems can make energy consumption more efficient and reduce unnecessary wastage. This can lead to substantial energy savings and a reduction in carbon footprint.

Local and regional elected representatives play a crucial role in environmental governance. Recognizing the link between the fundamental right to the environment and good governance at the local and regional levels, the Congress emphasized the importance of considering the environmental issue in their decision-making processes. The Congress is working on raising awareness among elected representatives by sharing good practices regarding the environment and AI through handbooks and guidance for smart cities and regions. This highlights the vital role that local and regional governance plays in addressing environmental concerns.

In the realm of public transportation, incentive-based systems can prove effective in managing capacity and reducing the need for extra transport capacity and investments. Such systems often offer different prices for train or bus tickets depending on the transport capacity, thereby encouraging people to choose less crowded public transport options. The implementation of AI-based systems has been observed to increase the modal shift from road to public transport, promoting more sustainable and efficient transportation practices.

The Swiss Energy Park is a unique initiative that encompasses three types of energy production: hydraulic power, solar panels, and wind crafts. By analyzing the consumption and production of energy in the region, the Swiss Energy Park allows for a comprehensive understanding of energy needs and facilitates targeted efforts in energy conservation. It is noteworthy that climate change can significantly impact energy production, as seen in instances where insufficient water for hydraulic power resulted from a lack of rainfall. This demonstrates the interplay between environmental factors and energy production, highlighting the importance of sustainable energy solutions.

Furthermore, AI has the potential to contribute significantly to combating environmental issues and reducing carbon footprint. It plays a vital role in managing public transport, leading to a decrease in carbon emissions. Additionally, AI technologies assist in managing resources in energy parks, allowing for better mitigation of the effects of climate change. These AI-driven solutions have the potential to revolutionize environmental conservation efforts and promote sustainable development.

However, the implementation of AI in policymaking comes with challenges, particularly in terms of privacy protection and data security. Deploying smart grid systems that manage energy consumption requires access to personal routines, raising concerns about the transparency of personal information if the system is hacked. Protecting privacy and preventing data breaches are essential considerations when integrating AI technologies into policymaking processes.

Overall, AI technologies present tremendous opportunities for creating greener and more sustainable cities and regions. By optimizing energy usage, managing public transport, and analyzing environmental data, AI has the potential to significantly reduce carbon footprint, enhance energy efficiency, and promote sustainable development. However, it is crucial to balance the use of AI with care, ensuring responsible energy consumption and safeguarding privacy. The involvement of local and regional elected representatives is pivotal for effective environmental governance and the successful integration of AI solutions in addressing environmental challenges.

Patrick:
It appears that we have a little technical difficulty, but we’ll solve that very soon so we can get started. We will also be showing some slides, at least some of the speakers, otherwise, Fadim, we can maybe change the order of speakers if they are not available right now. So this workshop is about AI and environment and the connection between the two. So it’s my great pleasure here to be in Kyoto, first of all, I also have some colleagues here and some friends from different parts, we also have a number of people that are following online even though right now we’ve combined everything, we are in presence, we are online, nice to see familiar faces and less familiar faces in the room, nice and friendly faces, I’m sure that we also have nice and friendly faces online. So thank you for coming to this workshop, the Council of Europe obviously has had a very special interest in both artificial intelligence and environment for a number of years, and we’ve developed a number of both treaties, but also partial agreements around environment, we’re currently working on a new treaty on artificial intelligence. Both these things were put to the forefront in our Summit of Heads of State and Government in Reykjavik, where the Heads of State and Government also requested that we pay particular attention to that and devise new tools. in this field. Council of Europe works in that, not only with a specific committee on artificial intelligence, but has a number of services that are looking directly into artificial intelligence. As we also know, every human right ultimately depends on a healthy biosphere. Without healthy functioning ecosystems, there would be no clean air to breathe, no safe water to drink, or nutritious food to eat. We need to create that and preserve that. Of course, the artificial intelligence may be a helpful tool in this respect, but we also have to ensure that this helpful tool serves its purpose. That’s why we’ve put together a panel of people that are on the one side scientists and researchers, but also decision makers that have to take on a daily basis the decision to whether or not imply and apply certain methodologies or not. Our very special keynote speaker today is someone who has been involved in the work of what we call the CAHI, the Ad-Hoc Committee on Artificial Intelligence, but also on the Committee of Artificial Intelligence on the Regulation of Artificial Intelligence for some time. He’s a minister, a minister for the environment of the Canton of Jura in Switzerland, and he’s also the spokesperson on digitalization and artificial intelligence of the Congress of Local and Regional Authorities of the Council of Europe. So I’d like to welcome Mr. David Herré. He is uniquely placed in this respect to share his experience as both an active policymaker domestically at the canton of Jura and at the European level and someone who has first-hand experience of actually working with those topics daily and locally as a minister for the environment. Without any further ado, I would like to give the floor to Mr. Aire, who will speak from Switzerland. He had some urgent business, unfortunately, in his government today, otherwise he would have preferred to be with us here in Kyoto, I’m quite sure. Mr. Aire, if you’re there, the floor is yours.

David ERAY:
Yes, I’m there. Thank you so much. I’m here in Switzerland. It’s still the end of the night, so I should say good morning from here and I’m sure you are already in the afternoon. So it’s a pleasure for me to address this session and I’m really grateful in the name of the Congress to be able to share our thoughts. So as you said, I’m Speaker of the Congress for Artificial Intelligence and Numerization. The Congress has a number of 46 state members and this is really a huge organization and we try to have a focus on these thematics that are really important at the moment. As you said, in my country, I am Minister of Environment in the canton of Jura. Switzerland has 26 states and Jura is one of the 26 states. You may know some of the states which are well known, like Zurich, Geneva, Bern, etc. As a politician, a grassroots player in my country, and as a representative of the Congress, I want to share my vision. on this very relevant connection between AI and environment. In October, 2022, so one year ago, the Congress highlighted that the fundamental rights to environment is intrinsically linked to local and regional good governance. Indeed, there cannot be good governance exercised by local and regional authorities without taking into account the environmental issue. So the Congress explored how we can move toward a greener reading of the European Charter of Local Self-Government. We adopted a recommendation, and this is a proposition to have additional protocol to the Charter on this matter. We have several other proposals of international standards on environmental matters within the Council of Europe, including a possible protocol to the European Convention on Human Rights. Whatever option is eventually chosen by the Committee of Ministers, the role of local and regional elected representatives in environmental matters is key. Both the environment and artificial intelligence are high on the agenda of the Congress. The Congress works on raising awareness of elected representative by sharing good practices with respect to the environment and artificial intelligence through practical handbooks and guidance for smart cities and regions. Our communities can become better, can become better places to live if we maximize the use. of AI for the public good. Indeed, AI technologies can be game changers, optimizing the use of energy, handling power fluctuations, improving energy storage, and forecasting energy demand can all help to make energy consumption more sober. AI enables us to analyze complex, multi-faceted data sets, inclusive real-time data on energy consumption, water use, and weather. I want maybe, I want to share my experience in the continent of Jura in Switzerland. We do have several examples. So I want to share a PowerPoint. I don’t know if you can show it on the screen for the participants, if you have it available. This is just two, three slides. I can illustrate my, I’m sure we do.

Patrick:
We’ll immediately try to put that on screen.

David ERAY:
So, because it’s also, it’s always good to talk, but it’s also good if I can show you some examples. So in my country, we do have several examples on energy use, energy management, and also public transportation management. And in that topic, I don’t see anything on the screen, but I think it should come.

Patrick:
As long as there’s nothing on the screen, I would invite you to continue for the time being. We’re trying to resolve this technically, but please go ahead.

David ERAY:
Okay, so in the public transportation, we have… We have implemented something to be able to manage the capacity of transportation and the need to be transported by the people. And how do we do that? We do have something that we could call an incentive. So whenever you need to buy a train ticket or a bus ticket in Switzerland, the system will propose you several different prices, depending on the capacity available in the public transport that is foreseen. So I wanted to show you an example. If I want to go to Zurich next week, and let’s say I have a meeting at 12 noon in Zurich, and the system will propose me several possibilities, including one with a discounted price at 12 francs instead of 22 francs. And this is a way to move the people, not in the train and bus that are already supposed to be full, but the one that has capacity. And this brings three effects. First of all, we have a better use of public transport. So we use the capacity and we don’t overload when it’s already full. Second effect, this can reduce the need of extra transport capacity. So this can reduce the investment that we, the states like Jura, like Bern, like Zurich, need to invest in our transport material. And the third effect. is also important. This has increased the modal shift from road to public transport. So three effects with a system based on AI and also based on the tools that we have online. The second example I wanted to share is what we have in my region called the Swiss Energy Park. So in my region, we have an energy park that includes three kinds of production, hydraulic power on the river, solar panels in a big solar plant, and wind crafts on the mountain. And in this park, we can analyze online the consumption of the region and the production of the region. And we see immediately that whenever we have wind, water, and sun, this is quite cool because we have enough energy. And during the period like now in Switzerland where we have sun, no wind, not enough water in the river, then we need to import energy from outside the region. And this is something that I wanted to show you on the slide that are not coming, but this is okay. I’m sure you can share the slide later. And on the analysis, okay, this is coming. So we go directly to the seventh slide. Okay, this is the next one, because I don’t want to repeat what I said. Okay, this one. On this slide, we can see on a yearly basis, the black line is the need, the consumption of the region. The green one is the wind crafts production. so we can see that like in December 2022 or February we had quite a lot of wind, enough wind to our consumption. The blue is the water production, so we see that the period from August to now we are having not enough rain in the region, so not enough water in the river, so almost no production. And the sun is also an energy that we have especially in summer and that is not present in winter. So this is interesting to see first of all the management of the energy in the region, also the effect of climate change because we see that when we have not enough water like now due to the climate change we are in trouble and we can also see that the wind is a really high energy possibility or potential, but this is not predictable so we cannot be sure. So this is just two examples that I wanted to show and maybe you can come back to the third slide just to show quickly, okay I come back to this public transportation. On the left you can see that this application you can just select, oh I want to go to Zurich on Friday November 10th for a meeting at 12 noon. In the middle you can see the possibilities offered, so the one that is arriving at 12 is 14 Swiss francs and if you want to be like in the previous proposition it’s 19 Swiss francs and if you want to be at Zurich at 11.26 you pay the full price 22. So this system is a way to as I said to use with the best efficient way the public transport capacity that we have in Switzerland. So this is what I wanted to show you, and I think this is good to make the link between AI and environment, energy and carbon footprint. We see that we have potential, and I think there are a lot, still a lot to do in this topic. Thank you.

Patrick:
Thank you so much, Mr. Aré. I think energy management, the use of real-time data, it’s incredibly important, and sometimes it may be better to go to Zurich a little bit earlier or a little bit later and have a free lunch in Zurich to have it compensated by your train ticket, basically. So thank you very much for this very local experience and how AI can really help in making sure that our environment is also getting better of it. Now let me introduce you to the work of our first panellist, because Mr. Aré was our keynote speaker. Our first panellist is Professor Yamagata from the Ayo University, who is all about developing a new urban system design framework that integrates architecture, transportation and human behaviour in cities. Professor, if you don’t mind telling us about your work on AI and sustainable urban systems, that would be very interesting for this audience, I’m quite sure. What are the main challenges and how did you deal with them? Could you tell us more about this? Thank you.

Yoshiki YAMAGATA:
Thank you very much, Chairman. It is my great pleasure to be able to talk at this session about our recent studies. I’m Yoshiki Yamagata, I’m talking from Teio University, Yokohama. So at my laboratory, we are studying urban systems design for achieving climate resilience. cities. So climate resiliency is two meanings. One is the response to the climate change, because we are experiencing a lot of climate change impacts already, like heat waves and floodings. Another climate change measure is, of course, the carbon neutral, the carbonization of the cities. This is also urgent to meet the target of the Paris Agreement. So for that purpose, we are introducing a lot of IoT, big data, and AI techniques to achieve this goal. So let me explain one example of my studies at the city center, Tokyo. Maybe you have seen this sky tree at the city center, Tokyo. This is a tourist tower in Japan, and we are analyzing this area using big data. So one big data we are analyzing is the occupancy of the offices and shops and restaurants, et cetera, using big data. And the second big data case is this mobile phone mobility information. We are deleting all the privacy information and using the trajectory of the people moving inside the city. So we are using the machine learning technique, which is an AI technology, to detect the transport mode. So by looking at the trajectory, AI can judge. if this is a car, or a train, or walking behavior, transport mode. We’re still working, still studying to improve the accuracy of the classification, but the walking behavior is really challenging for us. So, by combining these building and transport information using GIS information, like total floor area height, and load release node, and big data like occupancy information, and people’s mobility information in the buildings and in the load networks, in combination with the sensor data like smart meter measurement data and statistics, and also the actual transport measurements at the load network, we could estimate the dynamic carbon mapping, which visualizes carbon emissions from the urban activities. This red color means that emissions from the building energy use. Blue color means indicating the emission from the load car traffic, from the engine car. So, from this diagram, we can easily, intuitively understand where the carbon dioxide is emitting, and who is responsible for these emissions. So, it is really important to understand visually, intuitively, for the policy maker, as well as citizens, and in many cases, building owners. other business people in the cities understand what is the goal of carbon emission deductions. So this kind of information can also be used for detecting the heat wave risks by combining heat hazard maps. Remote sensing data can be available for this purpose and we can use this worker’s location information as a heat exposure to the risks of a heat hazard. For instance, if an older person suffering some diseases is walking in the street in a very high temperature location more than one hour, so there is a huge chance that this person gets heat stroke. So if these kind of people are staying in the same place for say 1,000 people, then maybe there is a high chance the ambulance will be called soon. So in advance we can prepare ambulance and send enough number of ambulance to the high risk area to save the lives of people who are suffering the heat strokes. So at the same time we can also do analyze the comfort of people. Actually walking behavior inside the cities is really important health improving well-being experience inside the cities. So by knowing how to improve people’s walkability inside the city is really important indicators for their people’s health and improving the well-being of the citizens. So there are some new technologies available for this purpose. And the big data and AI for using this people’s flow there is really a huge potential. This is ongoing studies I’m conducting with researchers of ETH, Zurich, Switzerland. And so we have a exchange program between KU University and ETH. So I’m very much looking forward to collaborate with policy makers and the researchers in Switzerland in the near future. Thank you for your attention.

Patrick:
Thank you very much, Professor Yamagata. I think that’s really exciting to look at this information, how climate resilient cities and decarbonization can impact or hopefully not impact further climate change. I think I would only give you one suggestion before you prepare the ambulances to prepare for heat strokes. Maybe more importantly that we foresee some other activity that prevents heat strokes to take place. Our next panelist is Peter Clottenbrock from the UK Center for AI and Climate. He works in in-depth on issues of creating data marketplace in relation to transition to net zero, and more specifically changing requirements for data for improved grid management. This may sound strange to you, so we will let Mr. Peter Clottenbrock explain what is meant by all of this. Peter, floor is yours. Peter, are you there? because we see your slide, but we can’t hear you.

Peter CLUTTON BROCK:
Okay. Is that working now? Can you hear me?

Patrick:
That’s perfect, Peter. Thank you.

Peter CLUTTON BROCK:
Great. Thank you very much for the chance to speak today. It’s great to be here, if not in person, then in spirit. I’m going to talk for about nine minutes today about what some of the opportunities are to apply AI and data science to support the transition to net zero, as well as what we can do to help free up some of the data required to do so. I have to move the screen along. Excuse me. There we go. A little bit about us before I dive in. The Center for AI and Climate is one of the leading organizations focused on advancing the application of data science and AI to accelerate action on climate change. We do this in two main ways. The first is thought leadership. We look to inform the debate about what the main opportunities are to apply data science and AI to accelerate the transition to net zero, as well as what some of the bottlenecks and barriers are that are holding back that adoption. Secondly, we look to dive into some of those bottlenecks and barriers and help develop the digital architecture and infrastructure necessary to do so. Perhaps it’s useful to start with a little bit of a framework to think about what kinds of problem AI is good for helping to address. Because there are obviously many challenges in the transition to net zero, some of them AI can potentially help with, others, it’s not the best tool to be used. We need to make sure it’s being used in the right ways for the right kinds of problem. Here, I’ve just summarized four of the types of problem where AI is particularly good at supporting the addressing of challenges. The first is system optimization. This often uses a tool called reinforcement learning, where you effectively inform the AI agent about a particular system that you’re looking to optimize. You give it data on the controls that can use to change that system and the environment that affects the system. And then it will effectively optimize using those controls the optimal outcome for that system. And this could apply for a whole system. So for example, the energy system, but also parts of that system. So a particular battery asset within that system could be optimized using reinforcement learning, for example. A particular subset of this is around accelerated experimentation. So we can deploy AI to support faster accelerated experimentation for new battery designs and new battery chemistries, for example, but also potentially for new ways of making steel, which we need new forms of experimentation for. Thirdly, prediction and forecasting. So a lot of the data that we need that we use in sectors relevant to climate change uses something called time series data, which tracks different variables over time. And here, if we’ve got enough historical data on that particular variable, we can find the patterns using AI in that data and predict forward much more accurately using AI than we could with previous techniques. And fourthly, classification. So this is useful if, for example, we have map image imagery or satellite imagery, and we want to be able to classify areas on rooftops that we could deploy solar panels on, or grid infrastructure, or whatever it is. We can deploy AI to help classify different data within that image. AI is not something that’s theoretical at this stage. It’s already being deployed, as David set out in his examples in Switzerland. But there are many others that we’re seeing bubble up throughout the community that are really exciting. So I’ve pulled out three that we think are interesting here. So the Climate Trace Coalition uses AI and satellite imagery to improve the accuracy and transparency of global emissions inventories. Secondly, Unisat’s Flood AI tool enables high-frequency flood reports that have improved. disaster response already in Asia and Africa. And thirdly, DeepMind have used their AI to increase the energy efficiency of Google’s data centers by between 30 to 40 percent, and that’s focused on improving the efficiency of their cooling systems. So that’s just using software they’re able to achieve really significant increases in energy efficiency. It’s worth saying that despite the fact that there are a lot of examples already deployed applying AI to climate action, we still think the potential for further applications is huge. And we think actually it’s probably some of the most important ones that we’re likely to see have yet to be developed. So it’s still a wide open field, and we’re just seeing the tip of the iceberg when it comes to the potential application. So what do we need to do to enable further application adoption of this technology? Well, probably the main barrier and bottleneck that comes up when you talk to the data scientists working in this field is around data. And in particular, two types of data frustration come up in conversations with data scientists, and these are data discovery and data access. So just to be clear on what I mean by data discovery, I’m talking about the process of locating and identifying already open data sets. So for example, an innovator might be searching for solar irradiance patterns in Africa. It might just take them a long time to find this data, despite it being already openly available. And by data access, I’m talking about the process of gaining access to commercial data that is currently not available openly on the internet. So for example, this might be accessing data on EV charging assets from a commercial charging asset operator where they’re not currently opening up their data. So then the question comes of what can we actually do to enable, to address these two challenges that I’ve focused on. And here we see three key opportunities. So the first is better data discovery tools. So ultimately what we think is needed here is better ways of organizing and helping signpost people to data that already exists. So here we think there’s a need for a well-organized and intelligent data catalog focused on climate action. This is actually something that the Center for AI and Climate is already working on developing to really help users and signpost users to where there is data for a particular type. And the organization of that is really key. If there are any country representatives who want to get in touch about that and find out how that could help support data cataloging in your country, please do let me know. Secondly, we see an increasing need for better regulation to open up data, especially in monopolistic sectors. So when it comes to climate action, a lot of the sectors that we care about most often have natural monopolies, whether it’s the electricity sector or the transport sector. We’re often dealing with areas where you have sectors that companies that have a monopoly over particular areas, whether it’s electricity networks, such as distribution and transmission networks or transport networks. And we see a real need to focus on requiring some of these monopolies to open up their data and in particular for commercial licensing. And that last piece is really key. So we want to be able to enable innovators to build products and services on top of data that’s opened up by these types of companies. So making sure it’s available on a commercial license is actually really important. And thirdly, commercial data markets. So to complement the open data piece, we actually see there being a real need to create the financial incentives for commercial companies to share more of their data, in particular in the sectors that we care about, again, when it comes to climate change. And the way you create those kinds of financial incentives is to effectively create a market. for that data. And again, this is something that we’re working on directly. So what I’ve talked through hopefully is a combination of things. So I’ve highlighted a framework by which we can think about the opportunities and the problem types that AI is good for addressing. I’ve talked about some of the case studies about how it’s already being applied and deployed in the world. I’ve highlighted some of the key bottlenecks in particular around data that we need to address if we want to see further and faster adoption of these technologies. And I’ve set out what we think is some of the key ways of addressing those bottlenecks to address these challenges. So with that, I’ll close and say thank you very much again for the opportunity to speak today. And I’ll look forward to addressing any questions. Thank you very much.

Patrick:
Thank you so much, Peter. Very, very interesting also how you highlighted the central role actually that data play in artificial intelligence, the system optimization, if you have access to those data. But you also pointed at high energy costs for storing those data and also deploying artificial intelligence on them. So our last speaker is going to be, we’re extremely lucky to have the new director of the Directorate of Science, Technology and Innovation of the OECD, Mr. Jerry Sheehan, who will present the OECD’s work and activities in the field of AI and environment. And in particular, the excellent report on measuring the environmental effects of AI computing and applications published at the end of last year. So clearly, Jerry, OECD has a key role to play. Over to you to tell us about your work.

Jerry SHEEHAN:
All right, thank you very much, Patrick. I’m delighted to be able to join you, even though it can only be virtually today, as much as I’d prefer to be there in person with you. Let me just say, I do have some slides. I don’t know if they can be presented here. I don’t seem to be able to pull them up and share my screen myself. But let me go ahead just to keep us on time and tell you a little bit about the work that we’ve been doing. Very good, thank you. So just to say that accelerating the green transition has been a major theme, continues to be a major theme of our work here in OECD’s Directorate for Science, Technology and Innovation. Among other areas, we have focused on issues of decarbonization of industrial activity, including in some more traditional fields like shipbuilding and steel. We recently released a report as well on AI and science that I’d call to your attention as it highlights a number of ways in which AI can be applied to research across a broad range of disciplines, many of which can inform and accelerate our green transition, including through a number of areas that were just described, through improved modeling, through improved data access and ability, and in fields ranging from environmental impact to transportation to material science. All of which can help us make our world a bit greener. We have been doing work on AI since at least 2017 and including in that have looked specifically at the relationship between AI and the environment. So as noted last year at the COP 27 last November, we launched a report that was asking about the environmental footprint of artificial intelligence. We heard a word about this in terms of some of the large data sets that we’ve been working on. So we have a number of data sets that must be used to inform AI. And I’m happy to share with you today some findings of this work. And actually the slide we have here is just the right one. So for us, we’ve been focused on the notion of the twin transitions, the green transition and the digital transition and looking at ways that digital technologies can be better leveraged for environmental sustainability in the future. As you’ve heard from other panelists already, this is happening in many ways. AI applications can enable sustainability. For example, AI is transforming climate modeling by creating digital twins. The destination earth, for example, is creating a digital twin planet of the earth powered by Europe’s high-performance computing centers and its AI capacity. The climate trace project is harnessing AI to track human-related greenhouse gas emissions with unprecedented detail and speed. DeepMind are using AI to make data centers more efficient by applying reinforcement learning algorithms to reduce their energy use. One example is carbon-aware computing where AI shifts compute tasks to data centers and areas with more availability of carbon-free energy. Let’s go to the next slide, please, just to say that we know compute is on the rise. And as we see computational needs of AI systems going, there are climate impacts as well. We often perceive AI as some sort of an abstract, non-tangible technical system, right, that we interact with through our screens. As noted, it’s enabled by physical infrastructure and hardware together with software that are collectively known as AI compute. And in the last decade or more, as you can see on this slide here, the computing needs of AI systems have grown dramatically, entering what some call the large-scale era of compute. This is no doubt motivated by the increasing capabilities of large and more compute-intensive AI systems, and of course, the rise of deep learning and large language models. Tools like Chatbot. are becoming more widely used, and the computing needs for inferencing of AI systems, contrast to the training of AI systems, is also becoming more relevant. Let’s go to the next slide, please. So why is this problematic? Well, simply put, as AI systems get bigger, not only can they help us address AI challenges, but they need and use more computing resources, which in turn consumes more energy, natural resources, and they produce increasing CO2 emissions. Although some researchers have produced numbers for AI’s environmental impacts at an AI model level, I think an example being for Bloom and for GPT-3, we don’t really know how severe this problem is at a national, let alone at a global level, especially then in comparison to other sectors that contribute to CO2 emissions. That’s because AI-specific measures are still scarce, and those that we do have tend to overestimate AI’s negative impact. So to help fill this measurement gap, we’ve conducted a stock taking report and developed a framework to help better quantify AI’s environmental impacts. Let’s go to the next slide. I’ll tell you a little bit about the analytical framework that we’ve used. The framework builds on work that has been done already by researchers on the direct and indirect environmental impacts of AI. The direct environmental impacts are defined as those that result from AI compute along with the resource’s life cycle, which includes the production, the transport, and operations of this compute, as well as its end of life. There are various environmental impacts, as you can imagine, along this life cycle, everything from critical minerals extraction to transportation, water consumption, carbon emissions, recycling, and waste disposal. For direct impacts, it’s important to note that operations, that is the actual running and operating of servers in a data center being used to train an AI model, for example, are a major source of environmental impact. The majority of resources and existing indicators are in just this area. We should also note though, that direct impacts can also be positive. For example, the heat from data centers is being repurposed, but these cases are still probably too rare. When it comes to some of the indirect impacts, that is from the applications of AI, we found many, many positive examples as well as some that were more negative. So on the positive side, we know that there are sectoral applications. We’ve heard some of those today already, such as AI for energy grid efficiency. There’s climate mitigation and adaption approaches, such as AI for flood prediction and AI for environmental modeling, such as the example of creating a digital twin of the earth. On the negative side, these AI applications also increase consumption patterns in ways that may or may not be sustainable. So let me go to the next slide, please. And I can share with you some of the key findings of our work here. So using this, we identified really five key findings that I wanna share with you just briefly this morning or this afternoon for an evening, for those of you who are joining from other parts of the world. So the first is that common measurement standards are needed to track and analyze environmental impact. And this should allow for greater data comparability between and among countries. Second, we find that data collection on environmental impacts of AI compute could be expanded, should be expanded in a number of ways. Third, AI specific measurements are sometimes difficult to differentiate from general purpose compute. We see this, for instance, in data center usage, where estimates of the percentage of data centers for use as AI compute. is not clear across countries. Maybe not even always as clear within individual data centers. Fourth, we need more data collection on different types of environmental impacts, such as carbon use, water and other natural resource use, and supply chain impacts. All of these are needed. Fifth and finally, we think international efforts, including sharing best practices on AI compute towards environmental equity and transparency, are vital. Let me go to the next slide, please. So just to note that the framework that we developed over the past few years coincides with the emergence of generative AI. Of course, the big question now is whether the arrival and the proliferation of generative AI would change our analysis. We’ve already seen exciting new applications of generative AI for climate action, such as chat climate. We also see considerable interest from countries. In a recent OECD stock taking that we did for the G7, for example, five out of the seven G countries responded that climate action is among their top five opportunities for generative AI. On the other hand, there are questions about the direct environmental impacts of the large scale use of generative AI. For example, on water, it was already reported that Microsoft’s water usage has significantly increased last year, largely due to investments in their operations of generative AI. Of course, I tried asking chat GPT if it knew how much energy it took to run this particular question. But as you see here, coming up with specific numbers is challenging and there’s considerable work still to be done. So organizations like mine here at the OECD, through our OECD compute expert group, for example, are continuing this important analysis, engaged with experts and partners from various stakeholder groups and from around the world. And we hope to be able to come back to you in the future with even more refined results of our analysis. So for now, I’m going to go to the last slide. And thank you for your attention. This is where you can find our report. And again, we’ll have more findings coming out of our OECD Compute Expert Group in coming months and years that we look forward to sharing with you. Thank you very much for your attention today. And I look forward to joining in the panel discussion.

Patrick:
Thank you so much, Jerry. It’s really a question of checks and balances, knowing how much energy is needed to generate the artificial intelligence on the one side, and how much is it going to help us to diminish the, let’s say, the carbon footprint on our development. I already have a number of questions here that come from the online. And one question, Jerry, is actually directed to you. We know that everyone looks at OECD with regards to defining AI as such. I’m not going to ask you that now. But the question here is, what do you see as the role of international organizations, such as the OECD, in working with artificial intelligence?

Jerry SHEEHAN:
Yeah, thank you for that question. I think the international organizations like OECD have a critical role to play here in simple terms. We’re the connective tissue that helps bring countries together to solve collective problems, including around the green transition and digital transitions and relationships between them. I think this is especially critical when the stakes are high and when it involves complex issues that cross borders. And this is particularly relevant for climate change and AI, given that it’s a general purpose technology that can be applied to many different sectors. We’ve been focusing on environment here today, but we know that AI diffusion is ramping up in almost every sector of our economies, from agriculture to health care. and again in all countries at different speeds. Applications like ChatGPT have made AI tangible and usable to the average person. So I think we at OECD and others remain hopeful that the breakthroughs that can be enabled by AI can help us save the planet, right? So these are the benefits and we’ve seen a lot of those in the panel today. We’ve also been attentive to some of the negative impacts, environmental impacts and some of the risks among those, including effects on labor markets and so forth. As noted, these aren’t well enough understood yet. They’re difficult to measure, especially as AI gets scaled up and is applied on a bigger scale. And that’s where I think OECD and other international organizations have a critical role to play because we can help put in place measurement frameworks that can apply across all of these countries.

Patrick:
Thank you.

Jerry SHEEHAN:
And just on a final note as I see you’re getting the microphone going, just to say of course now-

Patrick:
Yes, I’m trying to get it going. Thank you so much for your input. I think indeed, and as Council of Europe, obviously we work very closely also with the OECD and other international organizations around artificial intelligence and the impact of artificial intelligence. I already have a question. I have another question for Professor Yamagata because he showed us quite a number of visualization research, the use of AI in the sphere of sustainable urban systems. But Professor Yamagata, the question is also how can these systems be used in policymaking and do policymakers make use of them? I’m sure that Mr. Ere will be very interested in your response on that.

Yoshiki YAMAGATA:
Thank you very much for- There are interesting questions and that is very vitally important questions. At the moment, we are studying these visualizations using big data and AI for the stakeholders of the area. Of course, this includes the policy makers, but usually the policy makers need to see directly the policy options in this visualization, rather than the low carbon emissions. Of course, carbon emission is a final parameter to reduce, but perhaps the policy makers need to understand more closely the details of the different policy options, like energy management options, or urban planning options, or digitalization options, which also could have a positive and negative impact. This is a really important research question, how to involve policy makers into the use of AI.

Patrick:
Thank you. Let’s ask the policy maker, Mr. Herre, what would be, for implementing artificial intelligence in local and regional authorities, what do you see as the biggest challenge in implementation of artificial intelligence in day-to-day policy making?

David ERAY:
Thank you for the question. I think the biggest challenge is linked to the privacy, the protection of the privacy. What I showed before regarding this energy production and consumption, we try to implement what we call the smart grid, and that needs to implement in every house, in every apartment, a system that can manage. the need of energy. Let’s say you come back home at night, you want to load your electric car, so the system, like the big brother, should know that you are home, that the next day it’s like 7 a.m. you want to leave to go to Geneva, so you need the full load and then the system should manage in the best way to load your car linked to the production capacity and production perspective that we have during the night. So if you imagine that the system would be hacked, then that means that the entire life of the people could be transparent and given to the hackers. And this is maybe a big issue that we would have in terms of data protection and privacy respect.

Patrick:
We don’t hear you. Microphones are AI steered, so that basically means my simple intelligence doesn’t manage to get it going at the right time. Now, Peter, I have, if you’re still there because I don’t see you on the screen, but Peter, there’s a bit of a stargazing question for you. That is, what do you think a digitally managed energy system will look like in 20 years time? Give us a bit of glass ball staring.

Peter CLUTTON BROCK:
It’s a really good question and I’m not sure I’m going to be able to do the question justice, but I think effectively what we see is that AI will flow into a lot of the decision-making processes throughout the energy system. So just starting at the bottom, when you’re looking at when an asset developer might be looking to develop a solar farm or battery asset. AI will flow into optimizing those investment decisions. And then for the networks themselves, the electricity networks, they’re making decisions around what can connect to the networks, as well as what upgrades they’ll need to make to the networks. Again, all of those decisions will be optimized using AI. And so increasingly, increasingly, I think we’ll move to a system where electricity systems are effectively automated and the human capacity in them is more to check, to make sure that the AIs are working in the right way and the way that we want. But increasingly, we’ll see those humans in the loop starting to come out of the loop as the trust from the AI system is built. So ultimately, I think we will be heading towards a pretty much completely automated electricity system, albeit one where there is good democratic input, which may be perhaps the limiting factor on some of these automation features.

Patrick:
Thank you. I did put you a little bit on the spot there, but I will alleviate a little bit the burden of you and ask the same question because we have two minutes left. So I’ll ask the same questions. How do we see those checks and balances between the use of artificial intelligence? How do we make sure that the benefits outweigh the risks in the use of artificial intelligence in the coming years? And since we have four speakers, you have 30 seconds to reply to that. Shall I start with David?

David ERAY:
Yes, exactly. I think the check and balance, we need to be careful with the energy consumption of the AI hardware and the balance of the data. The benefit in terms of environment and energy saving, thanks to AI. So this is where I see a big challenge for us. 20 seconds.

Patrick:
Hello, thank you. Professor Yamagata, 30 seconds to check some balances for the future.

Yoshiki YAMAGATA:
Yeah, thank you very much. Actually, it is very important to see the benefit, understand the benefit of the users of the system. If the user enjoys the benefit, I think they understand why this system is actually useful for the community. If they don’t understand this is just a scary privacy problem, that’s my point.

Patrick:
Thank you. Jerry?

Jerry SHEEHAN:
So yeah, I would say that the way to do this is to ensure we’ve got a principle-based approach to AI, whether it’s applied in energy grid, whether it’s applied in transportation or others, that adheres to what I would say are the OECD principles around AI, which include issues of transparency, engagement. It’s a human-centered approach, which I think is what we were just hearing about engagement of the public and understanding the benefits, the risks, and having the opportunity for transparency into the policymaking process and the system development. I will just note that we at OECD are in the process of reviewing the 2019 AI recommendation with a view toward its revision, and this is happening at the time when generative AI is raising a number of new questions. So we hope to have something more to report on that in 2024.

Patrick:
Thank you so much. Thank you. With this, we are right in time to have finalized the discussion. Sorry, Peter, I haven’t given you back the floor a second time on this very difficult question. Thank you for the audience online and here in the room to have followed this session with so much interest. Thank you for the many questions that we received and thank you for your active participation. Thank you so much. Bye.

Audience:
Normally, it should not turn off if you do not have the plug, but here it is on. It turns off and then you have to turn it on and then there are several positions that are not clear. OK, because already last night, exactly. And so you turn it on, but you do not know if it is on. But it was perfect. Alone on a panel. Alone on a panel, it was perfect. Yes. Yes. Yes. Yes. Yes. Yes. Yes. Yes. Normally, it is on. Normally, it is on. Hello. Hello, nice to meet you. How are you? The teacher should normally be in person. Hello. Hello. Hello. Hello. We met online. I wrote about the blockchain. We had the launch in the webinar. You were there. I will just introduce myself. I attended many events of the Council. I haven’t been there. No worries, go ahead. No worries, go ahead. No worries, go ahead. I removed one of the panelists. You sent a video. Yes. You sent a video. No, no. No, no. Yes. That’s what I did. You sent a video. Yes. That would be complete. . . . . . . . . th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th th Thank you.