WASP Researcher of the Month - Finn Rietz
Finn Rietz
Researcher of the Month is Finn Rietz, doctoral student in computer science at the Center for Applied Autonomous Sensor Systems. Finn Rietz defended his doctoral thesis on 19 October, entitled "Knowledge Transfer, Alignment, and Interpretability: Modular Architectures and Inductive Biases in Deep Reinforcement Learning."
We asked Finn a few questions to get to know him and his research.
Describe your research in one sentence.
I work on advancing foundational deep reinforcement learning to make AI agents more sample-efficient, interpretable, and aligned with human intent and to ensure that advanced AI remains safe, controllable, and beneficial as capabilities grow.
What do you hope your research will lead to?
I hope my work contributes to safer, more reliable reinforcement learning systems in the real world. Mechanistic interpretability, understanding how and why an agent makes a decision, is a necessary precondition for true alignment and control. Ultimately, I hope that the architectures and techniques I develop serve as building blocks for AI systems that humanity can inspect, trust, and safely deploy for the common good.
What is the most unexpected result you've had so far?
How surprisingly "creative" RL algorithms are at discovering edge-case exploits in human-specified objective functions. Lacking human common sense, agents often solve tasks in ways that are equally hilarious and unsettling. Experiencing reward hacking firsthand makes it undeniably clear why interpretability and constraint architectures aren't optional extras; they're absolutely essential before models can ever be allowed to interact with the real world.
If you could collaborate with anyone in the world, who would it be and why?
It's hard to choose just one. Richard Sutton and David Silver are immediate choices as pioneers of (deep) reinforcement learning whose foundational works shaped the field and first instilled a feeling I can only describe as awe in me. Beyond RL specifically, I deeply admire Yoshua Bengio. Not just for his immense scientific impact, but because he approaches AI research and safety with genuine integrity and a steadfast commitment to the public good, which I find incredibly inspiring.
What do you do when you're not researching?
When stepping away from the screen, I like to spend time outdoors, whether that's walking or hiking with our dog, running, or camping. But tinkering with side IT and AI projects remains my biggest hobby and passion, even outside of working hours.
Thank you Finn!