"As a hub for the public sector, start-ups and science, Berlin can play a pioneering role."

Artificial intelligence has the potential to play a key role in the fight against climate change: Through data-driven modeling, the technology can help develop and implement innovative solutions for reducing greenhouse gas emissions. For example, AI can help grids use renewable energy sources more efficiently and improve load balancing. In addition, using AI in climate science can help to better understand and predict the effects of climate change.

Exciting approaches that Prof. Dr. Lynn Kaack knows very well. She has been Assistant Professor of Computer Science and Public Policy at the Hertie School in Berlin since 2021, where she heads the AI and Climate Technology Policy Group. Her work focuses on statistical and machine learning methods to support climate and energy policy, as well as climate-related AI policy. She is also co-founder and chair of the organization Climate Change AI. She spoke to #ai_berlin about the intersection of AI and climate change , upskilling for project leaders and arguments for greater commitment to climate change mitigation.

Prof. Dr. Kaack, what originally led you to focus your research on the intersection of computer science, public policy and climate change?

I had already planned to do something in the field of climate change mitigation when I was still in school. I had the idea that the societal challenge of climate change could be tackled using quantitative approaches. I first studied physics in order to work in the field of quantitative methods - always with climate change mitigation in mind. I always viewed politics as a major lever for advancing society and making progress on climate change mitigation. After my master's degree in physics, I went on to get a PhD at Carnegie Mellon University in Engineering and Public Policy, which combines an engineering and technical perspective with public policy.

How did you come to focus on machine learning? Was it the logical next step?

During my PhD, I studied statistics and machine learning on the side and found it very exciting, and ended up getting a master’s in this field. At that time, there were hardly any researchers in the field of public policy who were interested in machine learning. This is how I  came to the big question: How can ML methods be used in the field of climate and energy policy?

Is this approach one that was already cemented in the USA or rather something that was only developed when you returned to Berlin?

My research focus was already established in the USA. I got to know some researchers there who were interested in this intersection. Towards the end of my PhD, we wrote a scientific paper on this topic together. Most of our group of authors came from machine learning and were interested in climate change, but for me it was the other way around: I came more from the fields of climate and energy and was interested in machine learning. I think that many of us felt like outsiders in our specialist areas because this intersection simply didn't exist as a topic. In Germany just as little as in the USA.

Our work developed into something of a movement and we very soon founded an organization called Climate Change AI.

What is Climate Change AI all about? What exactly are the goals?

With Climate Change AI, our mission is to catalyze work at the intersection ofmachine learning and climate change mitigation and we offer a whole range of resources and initiatives.

On the one hand, we offer regular workshops at AI conferences where people can exchange ideas about research. We have also set up an online community for direct communication with each other.

On the other hand, we run educational programs, such as our Summer School, the virtual part of which is a series of online lectures that can be attended at low cost. We also offer research funding through our own grants program, the Innovation Grants, which is now entering its third round.

How exactly is the organization set up?

By “we”, I refer to a large group of volunteers, about 50 people from all over the world, supported by a few permanent staff. We are registered as an NGO in the USA, but we are a global organization because climate change is a global problem.    

You just mentioned the Summer School. What are the main goals of this program and what do you hope participants will take away from it?

Our Summer School actually consists of two different programs - a Virtual and an In-person Summer School.

The Virtual Summer School runs from mid-June to early August and consists of a series of lectures that are accessible at a low cost and are intended to provide a good overview of the major topic of climate change mitigation and artificial intelligence.

In terms of content, we will delve into the various sectors and at the same time deal with important overarching questions about the impact of AI, ethical issues or the implementation of projects.

One goal is also to develop a common language. We want participants from the AI community to take a closer look at the topic of climate change mitigation and understand what issues there are, how to work on them and what the criteria for impact are. And conversely, if you come from climate-relevant areas, you can take a closer look at the topic of AI and understand what the technology is actually useful for and what are successful teams for  implementing AI-projects.

And then there is the In-person Summer School. This year, the Summer School is taking place at the Mila Québec AI Institute in Montreal. There will be around 40 participants from all over the world, who we have selected from many applicants and who then develop projects together. The main aim here is to form an interdisciplinary cohort and initiate projects.

Let's take a look at the topic and take stock: where do we currently stand in terms of using AI and machine learning to combat climate change? In which direction could the topic develop in the near future?

Overall, I believe that in many areas AI and machine learning are not yet established in practice for addressing climate change . However, the topic has become salient in most areas, so there is at least awareness of it. Researchers, in contrast, are clearly already increasingly working with machine learning.

Of course, AI is not a silver bullet. Nor can the methods be applied sensibly in all areas.  And even where there is a proven added value, it will take time before we can benefit from an effect on the climate change . A certain level of market maturity is required before ML-based solitions can really be integrated into existing processes across the board and make a difference in reducing emissions. Some sectors of industry are relatively advanced in this respect, but others are not yet. I see the electricity sector in particular as a pioneer.

Another example would be the use of AI solutions for predictive maintenance. This is not necessarily climate-friendly per se.  But if this happens in areas that are key pillars of climate change mitigation, such as railroads like Deutsche Bahn, and it helps to reduce costs and make processes more reliable, then it can also make a difference for the climate.

AI is already playing a major role in information procurement. It generates knowledge that decision-makers can use, for example, to plan better, create better laws, or monitor changes in the climate and how humans adapt. There are also more opportunities to actually enforce the laws. For example, many tools for monitoring deforestation of rainforests are already being used in a relatively standardized way.

So there are many promising areas of application. What do you see as the biggest challenges or limitations that need to be addressed in order to use AI effectively for climate change mitigation and adaptation?

As already mentioned, we still often see difficulties in practice, especially in smaller organizations or in the public sector.  In many areas, ideas exist and are perhaps already being tested in a pilot project, but the cost-benefit calculation remains uncertain. It is a relatively large financial risk to start using AI from scratch. You need a lot of expertise, digital infrastructure, data management, etc., which is not yet available in many areas.

In my view, acting on climate change  must also be the primary goal. This means that if a project ties up a lot of resources due to AI, this is not necessarily always conducive to climate change mitigation. You have to weigh things up wisely and focus on where an application offers a lot of added value and where the conditions are right.

What policy changes do you think are most important to facilitate the effective integration of AI technologies into climate change strategies? Are there specific areas where policy makers are currently lagging behind?

We often have the problem that you need to build quite a lot of capacity for AI in terms of skills or alternatively understand how you might be able toimplement a project with external experts. For many companies, it is simply not possible to set up their own AI division. According to surveys, many stakeholders in the energy sector don't actually know who they could turn to if they were interested in AI.

There is a massive lack of clarity here and this is where politics could provide support. One idea would be secondment programmes where AI experts could go to companies or organizations without burdening them financially. This would perhaps allow them to experiment, for example to understand where there is potential before certain projects are even launched. In addition, we need an ecosystem of AI providers who have expertise in the various climate-relevant sectors and are also transparent in terms of costs. At the moment, a lot of this AI consulting is provided by big tech and large consulting firms.

You have already mentioned the topic of upskilling. At the Hertie School, you have set up a certificate program for the autumn. What elements are you focusing on and what importance do you attach to the topic, particularly in the political business?

That's right. This is a certificate program from the Hertie School with a focus on AI in public administration, which is funded by the Dieter Schwarz Foundation and was set up jointly with TUM and the Oxford Internet Institute. It is relatively intensive and consists of five modules, each lasting two and a half days.

Our aim is to offer a program in which executives from public administration who want to get to grips with AI can acquire the necessary know-how. Wherever AI projects are set up, a lot of knowledge and expertise is required: Who do you need to recruit? What can be achieved with AI? What is needed in terms of IT infrastructure? In this program, we are not training anyone to actually build AI models. Instead, our approach aims to promote understanding of the technology in all areas of administration.

In addition to capacity building, promoting knowledge and building a network, what other measures are there that can also be incentivized economically? What other arguments are needed for a stronger commitment to climate change mitigation?

One major lever is probably to raise the CO2 price further so that companies feel that climate change mitigation is worthwhile. This is not specific to AI, but we would probably also see more AI applications that help to reduce emissions.         

Another key area is data. The public sector could, for example, support the creation of data pools that promote the exchange of data between companies within a sector. Unfortunately, valuable data are often held proprietary  such that scientists, competitors or potential AI implementers never get to see them. In many organizations, data are also not sufficiently shared internally and there is no awareness of which data exist and what could be done with it.     

Do you think the Berlin ecosystem is well positioned when it comes to AI and climate change?

Berlin is definitely a hub for climate issues, and AI also draws many people to Berlin. I personally know quite a few researchers working at the intersection of AI and climate here. Berlin is excellently positioned in terms of research and start-ups. The city can play a pioneering role in Germany and Europe thanks to the interplay between the public sector, start-ups and science.

What is the appeal of the location and the ecosystem for you?

For me as a researcher working on climate policy, Berlin is the most important place in Germany. There are numerous universities and institutes here, such as the Potsdam Institute for Climate Impact Research, which work on topics related to climate change. At the same time, there are institutions such as the Hertie School, which conduct internationally renowned research in the field of public policy. My work also benefits greatly from the presence of the German government.

What advice would you give to young scientists and political decision-makers who want to use AI to contribute to solutions for climate change?

There are many different ways for young scientists to engage with the topic, and with Climate Change AI, we provide a platform to get started. Regarding what to work on, I would not want to point to a particular area. The types of problems that relate to AI and climate change are so diverse that you should really aim to pursue the topic that matches your own strengths and interests. Key to successful projects is to work in interdisciplinary and cross-sector teams, and to engage with all relevant stakeholders from the get go in order to make sure that solutions are meaningful.