Panel Discussion on AI & Climate Change
Notes from the Applied Machine Learning Days (AMLD)
I have taken a few notes on a panel discussion at the Applied Machine Learning Days (AMLD). The full discussion can be seen here.
- Felix Creutzig, Mercator Research Institute & Technische Universität Berlin.
- Buffy Price, AI for Good Partnerships Manager, Element AI
- Olivier Corradi, Founder, Tomorrow
- Liam F. Beiser-McGrath, Senior Researcher, ETH Zurich
- Eniko Székely, Senior Data Scientist, Swiss Data Science Center, ETH Zürich and EPFL
- Kristina Orehounig, Head Laboratory of Urban Energy Systems, Empa
The panel debate began with the moderator asking to list one particular application of AI. The answers was a combination of policy, causality and building (operations, room temperature). Training forecasting to reduce energy consumption by 25%. In jail it is possible to compute the footprint of everything, understanding ‘externalities’.
Then they went on to discuss barriers to AI. Earth observation the biggest challenge is satellite imagery, that is in terms of using or buying/selling data (a prohibitive factor). Still disciplinary norms of where you publish, you should be publishing at certain places. The journals publishing interdisciplinary is considered sometimes less important. There is a push for interdisciplinary things, but there is a barrier to research. If you want to go into interesting questions you may need than one ML method (more than one). The expertise is hardly available. Institutionally it is not that easy to get things done. In the Swiss data science centre (bringing people together is important). It is hard to get over the first step of understanding each other — where we would give up first. It is hard to find out where to publish, it must be interesting to consider whether there could be joint journals. Urban scales you need a lot of data, you may have privacy protection issues, it may take a long time to get data. Sharing the information out may have problems with privacy. When you publish it is important that data follows for replication studies. For industry, the vocabulary and the process. There are regulations in regards to grants and funding (for good projects, social benefit) you don’t have access to the same funds. The way structures, the legal structures are, as they stand can be a barrier. If you are for-profit it can be formulating a business model (one of the biggest problems). Collaboration between nonprofit and public is challenging.
Looking at one of the people present Olivier Corradi, Founder of Tomorrow is interesting and their website may be worth checking out.
Next question: given limited expertise is machine learning a distraction from this task and can be negative. What are the characteristics of solutions that need to be paid attention to?
Olivier Corradi is the first to answer the question.
- Rebound effect, when it is more efficient and you use more. Quantify the impact and behind that — what is the carbon footprint of using that electricity, if you are using that model.
Followed by Felix saying that the bottom line is that applications are done by Amazon, Google and Facebook. It works in terms of incentivising digital consumption. What is the utility, different use, and it is worth looking into. Then it is a question of what we might want. Profiling might be useful for targeting to influence in terms of climate change, but democratic government is required to listen to dangerous incentives. We can decide we want systematic incentives for reduction policies. What is okay, and not okay, to reduce run-over (spillover) effect and enable positive things.
Liam then says deep knowledge is important and the idea that technology fixes everything — while sidestepping this is not a solution, it creates new problems.
Eniko talks arguing that PhD students are doing more instead of using a basic practice such as a linear regression.
They open up for questions from the audience.
An audience member echoes the statement about PhDs in regards to BERTs as opposed to LDAs.
Another audience member said that AI is not so prominent. Then someone echoes a question of why Europe cannot build their own data centres — apparently many are lying. Why cannot Europe build great cloud computing?
Olivier said that he has been looking at data centres. Cloud are going away from green certificates. Publicly available reports, the 24/7 Initiative, they look at hour by hour how clean the grid’s electricity was at the time. They are trying to look at more marketing claims towards physical claims, and that is not being done enough.
The next audience member brings back up the PhD question with academics alternative routes or focus points. Felix defends publications with the peer review and quality control saying it should maintain its whole. Liam mentions other areas in political science. Eniko says that there is more access to code in recent time. Swiss data science is developing a platform for data versioning to see what is the processed data, not only publications. Buffy Price mentions the funding agencies and data management plans. There are additional initiatives that forward validation or challenging the results. Olivier mentions interactive data visualisations, one of the best ways getting to know data visualisation, typing text and getting to know how it works getting intuition of the model, validation by the community.
Copernicus satellite is mentioned with sentinel data delivered in an open data license schemes useful for climate applications. Europe Open Science Cloud as well for building climate change applications. There are initiatives.
The relationship between big tech and fossil fuel is then mentioned with talks of transferable solutions working in an unintended way to reduce costs for large fossil fuels companies. Rounding off with discussions of open science.
This is #500daysofAI and you are reading article 277. I am writing one new article about or related to artificial intelligence every day for 500 days. My current focus for 100 days 200–300 is national and international strategies for artificial intelligence. I have decided to spend the last 25 days of my AI strategy writing to focus on the climate crisis.