To build machine learning systems that can infer human preferences from their behaviors, make decisions that align with human values, and interact with humans, at the moment, I focus on three (interleaved) sub-areas:
- Preference Learning: how could machines learn about human preferences?
- Human-aligned Learning: what should be the objectives for machine learning models and how could we learn models that optimize them?
- Interaction: from ethical, legal and algorithmic perspectives, what are some considerations for building learning systems that interact with humans in a non-physical way?
When studying these problems, I commonly use tools from sequential decision-making and causal inference and models suggested by behavioral studies including (behavioral) economics and cognitive psychology. The particular applications driving my research now are recommendation systems and decision-support systems.