Interview with Sophia Wilson

Sophia Wilson is a PhD fellow in Machine Learning at the Department of Computer Science, University of Copenhagen

What is your PhD project about, and what aspects of sustainable machine learning are you most excited to work on right now?

My PhD focuses on the environmental sustainability of machine learning and large-scale data. I will work on methods for reducing the energy footprint of data, for example by compressing or curating datasets in more efficient ways, and by training models with less data while still maintaining strong performance. Right now, I am particularly interested in understanding the full cost of large-scale data: not only the environmental impact, but also the social and economic consequences of the data practices that underpin modern machine learning. 

What first inspired you to study physics, and how did your interests evolve as you moved from astrophysics toward climate-aware AI?

I was always drawn to maths and science, and in gymnasium I developed a strong interest in astrophysics. For a long time I even wanted to be an astronaut. An exceptional physics teacher helped spark that interest, but two specific moments stand out: both were talks given by female physicists, and seeing someone who looked like me doing high-level research made the field feel tangible.

During my bachelor’s I worked on several astrophysics projects and enjoyed the scientific process, but over time I felt the field was quite far from everyday concerns. At first I thought that meant academia was not the right path for me. Later I realised that what I was missing was a sense of societal relevance and that the issue was not research itself. When I came across the Sustainable Machine Learning group at the Computer Science Department, the focus on problems with direct environmental and social importance immediately resonated with me.

Your master’s thesis was co-supervised between physics and computer science. How did working across two disciplines influence the project and your way of thinking?

I think the project gained a lot from sitting at the intersection of physics and computer science. By that point I already knew I wanted to work with the sustainable machine learning group, and combining that with climate physics created a natural synergy. At the same time, many physicists are somewhat skeptical of machine learning models for modelling physical systems because they often behave like black boxes and ignore physical laws and structure. Embedding physical principles into machine learning models allowed me to address that scepticism while also improving efficiency. The result was a project that advanced both the sustainability goals in machine learning and the methodological expectations in physics.

Do you still feel connected to physics, and how does your background shape the way you approach machine learning problems?

I still feel closely connected to physics, and it shapes how I approach machine learning. Physics trained me to reason from first principles, reduce complex systems to their essential components, and build solutions step by step. Just as importantly, it taught me to work productively even when the full picture is not yet clear. In physics you often work with phenomena that are impossible to grasp in full, whether it is an infinite universe or a system that is simply too complex to hold in your head. You learn to accept that and still make progress. That mindset transfers directly to other fields: you get used to operating with partial or highly complex information without being paralysed by it, which makes problem solving much more effective. Generally, I feel that physics has made me more systematic, comfortable with uncertainty, and confident in tackling open-ended research problems.

Why is it important to consider the environmental footprint of AI models, and what do you think people often misunderstand about this topic?

I believe that it is important to consider the costs of any new technology, especially one growing as rapidly as AI. Some people are optimistic to the point of being naive about what AI will eventually solve, including the climate crisis, and assume that the current environmental burden is justified by those future benefits.

What impact do you hope your research on sustainable ML will have in the long term, whether in science, industry, or society?

I hope my research helps normalise a more resource-aware approach to machine learning. Scientifically, I want to develop methods that make low-resource modelling both easy to adopt and competitive in practice, so efficiency becomes a first choice. In industry, I hope it encourages more careful consideration of the cost of data and the impact of scaling. For society, the goal is for AI development to become genuinely sustainable, so that technological progress does not come at the expense of environmental or social well-being.

You spent 1.5 years working in consulting during your studies. How has that experience influenced how you approach research and collaboration today?

In consultancy I learned the value of building and maintaining a broad and diverse network. It is something I am actively thinking about and I try to be proactive in reaching out and connecting with others. Both my consulting job and my journey to my PhD position began with me contacting someone directly on LinkedIn or by email. Not being afraid to reach out opens opportunities that would not appear otherwise.

You returned to your old upper-secondary school (gymnasium) to teach math. What motivated you to do that, and what did you learn from engaging with students, especially girls interested in science?

I had just left my consulting job to focus on my master’s thesis, but when the chance to return to my old gymnasium came up, I could not resist it. I think it is crucial for young people to have role models they can relate to, and I wanted to be that for girls interested in science. What I learned from the experience is how much representation matters. When students see someone who has taken a path they are considering, their confidence shifts and their questions change.

You also run workshops on sustainable AI for upper-secondary students. What kinds of conversations or questions from students have stayed with you?

Many students tend to equate AI with ChatGPT. When I show them how often AI appears in everyday life, from recommendation systems to social media feeds and dating apps, they are genuinely surprised. Those conversations stay with me because they show how invisible these systems are. It is important to highlight this, since all algorithms carry biases, and you can only think critically about them if you recognise where they operate.

What made you decide to pursue a PhD, and what have been the most important skills you’ve gained so far?

I chose to pursue a PhD because I enjoy being part of a research community, and I found the topic of this position both scientifically interesting and socially relevant. One of the biggest differences between a master’s thesis and a PhD, which I have learned so far, is that the work is no longer a solo project overlooked by a supervisor. My PhD projects are to a much greater extent collaborations, which require coordinating tasks and expectations, presenting progress in a way that supports joint decisions, and integrating feedback continuously. These have been some of the most important skills I have developed so far.

What activities or hobbies help you maintain a good balance between work and life during your PhD?

Luckily, there is quite a lot of “life” happening at work as well. Our group is quite social, and the mindset is that research should first and foremost be fun. Outside work, I recharge by spending time with my boyfriend, family, and friends. I also make a point of doing things that pull me completely out of work mode, like going to concerts, watching documentaries, and staying active through sport.

Do you have mentors or role models who have inspired you along your journey?

My main mentor and role model has been my mum. She has always been supportive, but also actively involved in helping me think about my next steps. When I was unsure whether to pursue a PhD or move into industry, she was the one who encouraged me to try consultancy to gain experience outside academia. I have always looked up to how she manages an interesting career and a busy social life, while remaining the steady anchor of our family.

Have you been involved in student associations, networks, or initiatives, and what role have they played in your academic life?

I have been involved in several student activities, mainly through social committees like the ski and gala committees, tutoring new bachelor and master’s students, and co-founding a running club. These activities are informal, but they helped build strong networks, create a sense of community, and make it easier to connect with people across year groups. On the more formal side, I am now planning to join FemTech, which organises workshops for female gymnasium students interested in computer science, alongside my own outreach work.

What advice would you give to young people, especially women and minorities, who are considering studying physics or computer science?

Do it! Both fields are far broader than they appear from the outside, and once you start, you realise how much freedom there is to explore different directions. They also open many doors afterwards, both inside and outside academia. It will not always be easy, but it is not easy for anyone, and that means there is a strong culture of helping each other. Also, I cannot emphasize enough how social physics at UCPH is. The year groups mix, people know each other across cohorts, and the many events and traditions create a real sense of community, making the studies a lot easier and more enjoyable.