Field: AI safety

Position & Organization: Research Scientist, DeepMind; Cofounder, Future of Life Institute

How did you get started in this field? My undergraduate studies were a mix of math, statistics and computer science, and my PhD was in the intersection of statistics and machine learning. During grad school, I got involved in the rationality community and became convinced of the importance of the long-term future and AI safety, which also led to co-founding FLI. Organizing FLI events was a great opportunity to meet awesome AI researchers and get a sense of the latest progress in the field. This also influenced my thesis work towards machine learning, and I ended up working on the interpretability of machine learning systems. After finishing my PhD, I joined the safety team at DeepMind.

What do you like about your work? I get to work on interesting and potentially highly impactful problems in a supportive environment, without the usual academic pressures of teaching and applying for grants, and collaborate with brilliant colleagues who are highly motivated to ensure a positive future for humanity. DeepMind is a very inspiring environment and a great place to work on AI safety, since observing AI capability progress as it unfolds also informs advances in safety. The safety field is still in its early days, which creates an opportunity to make significant contributions, e.g. by defining and clarifying the problems.

What do you not like about your work? Long-term safety research comes with a lot of uncertainty about whether I’m working on the right problems and how much impact my work will ultimately have. Since I am trying to address issues with general AI before it exists, many of my current assumptions are likely to be wrong. What I’m doing could be very important or it could be absolutely useless. I sometimes envy other AI researchers who are working on more tangible problems with more immediate outcomes.

Do you have any advice for women who want to enter this field? If you are starting a PhD, make sure to choose an advisor who is not only a good researcher but also an effective and supportive mentor (e.g. ask their current students what it’s like to work with them). Doing internships during grad school is a great way to learn about different problems and different research environments.

Give yourself the space and self-trust to do deep work. It’s easy to get overcommitted to non-research activities, especially since female researchers are often asked to do more things due to well-meaning efforts of event organizers to improve diversity. Don’t feel bad about saying no to a lot of activities in order to focus on what is the most important to you.

What makes you hopeful for the future? The field of AI safety is growing and becoming more mainstream within the broader AI field, though still somewhat controversial. It’s great to see more and more talented and motivated people entering the field to work on these interesting and difficult problems.