For a recent story I interviewed Lynn Kaack, one of the co-founders of Climate Change AI, a platform designed by scientists working in the AI field trying to bring together more experts and colleagues to help solve sustainable issues with their tech talent. In my text – which you can read here and here – I wasn’t able to include more of our conversation so I decided to add this bonus interview here for anyone interested in the topic (here’s another one with the journalist Clive Thompson); my research was supported by the US section of the Heinrich Böll Foundation.
The 32-year-old comes from near Lübeck in northern Germany, did her doctorate at the AI stronghold Carnegie Mellon in Pittsburgh and is now a senior researcher at the ETH Zurich. She works on various projects on climate protection in the energy sector and is primarily looking at how machine learning (ML) – such as text analysis – can be used to provide important information to political decision-makers. Most recently she published this paper, which describes how to use ML to estimate building heights; these data are often missing, but are needed for urban planning and calculations in the energy sector.
Lynn, Artificial Intelligence and Machine Learning are currently getting a lot of attention from the media, science, politics and business. Does this also apply to the question of what contribution AI can make to address climate change and global sustainability goals?
Absolutely. My colleagues and I are often contacted by companies, policy makers, and other organizations in the public sector to discuss questions around AI’s relationship with climate change. With Climate Change AI, we also organized a virtual workshop day at NeurIPS, the largest machine learning conference, at the beginning of December, which was very well attended. The topic of climate change is starting to resonate with the machine learning community.
Do you think the hype is justified?
First of all, AI is not a silver bullet against climate change. It is a tool, and one of many methods that people can use to catalyze other approaches to address climate change. The other thing I would like to emphasize is that AI as such is also not inherently good or bad for the climate. AI and machine learning are methods that can be used for climate change mitigation as well as for the production of crude oil or gas. And AI is used very widely in online advertising, which is also not necessarily sustainable as it can unnecessarily increase the consumption of goods. All of these are reasons why we need to understand how to make the technology align with climate strategies, essentially to steer it in the right direction through, e.g., setting the right incentives.
With the support of the AI luminaries Andrew Ng and Yoshua Bengio, you and your colleagues at Climate Change AI published an 80-page paper at the end of 2019. In this paper you almost exclusively emphasized the opportunities the technology offers for climate protection in 13 different sectors.
This paper was primarily aimed at those in the machine learning community who wanted to help address climate change through their work, but didn’t necessarily know how. We wanted to show AI researchers and engineers how they can use their work and talent for good, in this case to help address climate change. Given that purpose, we emphasized that although AI is not a silver bullet, it can still make a beneficial contribution. One fundamental prerequisite is that those machine learning researchers, who are not trained in all the other disciplines relevant for addressing climate change, are collaborating with people who are experts in those areas.
»With AI there is a temptation to rely too much on the data alone and to neglect valuable knowledge from other research areas«
Computer scientists and coders don’t look at sustainability and climate change?
Fortunately, that’s not true at all. So far, however, very few also manage to address these topics directly with their work. Combining an interest in sustainability and a career in tech is still difficult. I saw this during my time at Carnegie Mellon University: Skilled graduates are highly sought-after and can have a job with a huge salary immediately after graduation, for example in the advertising industry or with tech companies. You can choose to pursue a different career – but then you may work more and/or earn significantly less. The sustainability sector is hardly competitive from a career perspective, and appropriate incentives are not yet in place.
So you want to convince colleagues and graduates not to be enticed by industry.
It’s not about working in industry or not. There are also many climate-relevant industries. We want to show that, in addition to the classic tech careers, there are also other areas that involve exciting machine learning applications and are moreover relevant to climate change. This was one of the things we had in mind when we founded Climate Change AI. With this platform, we bring people together that work at the intersection of AI and climate change, and provide them with a space to exchange and discuss ideas. Since our founding, we have received positive feedback from many, who are happy about having a place to submit and present their work, and network with others. This particular area is rapidly evolving and it is important to have a place for finding experts and having in-depth debates.
Developing algorithms can consume a lot of energy. A study by the University of Massachusetts Amherst came to the conclusion that training a language model results in the emission of 284 metric tons of CO2e; an individual living in Germany causes an average of around 11 metric tons a year. Another criticism is that training data contains human-made biases and machines then reinforce them. With this in mind, how great can the benefit of AI be?
It is true that those language models have become very large in recent years. Those are not the only energy-intensive AI models, for example processing large image databases (e.g., satellite images) can also require a lot of computational power. The thing to understand here is that there is a huge diversity among the type of models and their sizes; many AI algorithms can be run on a laptop and use very little energy. What is sustainable and what is not has to be seen on a case-by-case basis. The variability is simply too great to make a blanket statement. Together with Emma Strubell, who authored that paper from UMass Amherst, and other colleagues, I have recently written a policy brief that illustrates how these different impacts and opportunities of AI come together. The discussions about ethical issues in AI that you mentioned are very important as well. Basically, there is a lot that needs to be done to understand the risks and opportunities that come with the practical use of AI and machine learning.
In public discussions, two opposing opinions seem to dominate: one is very optimistic, the other is very pessimistic.
There are many who believe that machine learning will fix things. Once I was even asked whether AI is capable of designing climate strategies for policy makers. I believe that this is misunderstanding what AI is currently capable of, how it is applied in practice, and where the main levers are for addressing climate change. As a matter of fact, many of us at Climate Change AI work in the policy area, me included. We are very well aware of where the powerful levers are, and that artificial intelligence is at best an aid to established processes, not a replacement. On the other hand, many consider AI to be a hype that is not worth looking at. With our paper, we have shown that this is not the case and that AI methods can definitely play a role in addressing climate change.
The big tech companies like Google, Microsoft and AWS position themselves with »AI for Good« programs and make their technologies available to social start-ups and NGOs. What do you think about this?
Of course, it is generally good if a company like Google sets up a project to use AI and digital technologies to predict floods, which now can send out early warnings to citizens, and which would have not existed otherwise. But one also has to see the flipside, which means that like this a lot of expertise is concentrated in large multinational corporations that would have otherwise been with governments. It would be ideal if such a project was developed and continued locally in public institutions or smaller companies. But experts skilled in AI are scarce and sought-after, and perhaps many institutions cannot or do not want to afford the necessary capacity and infrastructure. One idea could be to establish specific programs that fund AI researchers working on climate change for a few years in such an institution.
One criticism of such projects is that those with the money and the possible technical solutions live and work in the West – and therefore often do not have an eye on what the local people actually need.
This is indeed a problem—that many scientific institutes with the necessary funding are located in North America and Europe—and it is not only constrained to the field of AI. Something that is particular to AI is the temptation to rely too much on the data alone and to neglect valuable knowledge from other research areas, or from experts. Local expertise is often overlooked, which can cause a lot of damage.
And if they join forces, do they meet as equals?
I hope so, but in practice such interdisciplinary work is difficult. I can mostly speak to how it works in university, and one thing I noticed is that interdisciplinary work means that everyone involved has to explain a lot over and over again. Social scientists and computer scientists often have different disciplinary cultures that have to be translated into a different language. The respect is there, but people don’t always understand immediately what the other one means.
In December 2020 Lynn discussed how AI could mitigate climate change with her co-founders David Rolnick and Priya L. Donti in this webinar for the Heinrich Böll Foundation. There’s also a paper titled »Artificial Intelligence and Climate Change: Opportunities, considerations, and policy levers to align AI with climate change goals« which can be downloaded here.