Four Pillars of Human-Aligned AI Representations: A talk by Prof. Andreea Bobu (MIT AeroAstro & CSAIL)

​When a robot carries a cup of coffee, it has to figure out what matters: keeping the cup upright, moving smoothly, avoiding obstacles. The way AI systems internally represent these task-relevant concepts is at the core of whether their behavior ends up aligned with human wishes and values, or not. So how do you make sure a robot actually knows what you want?

​Prof. Bobu will lay out four properties that any well-aligned AI system should have: completeness, minimality, efficiency, and interpretability. She will ground them in concrete robot learning work from her CLEAR Lab at MIT. She will argue the framework extends beyond robotics to any AI system that must act in accordance with human values, offering a foundation for meaningful human oversight.

​About the speaker

Andreea Bobu is an Assistant Professor at MIT in AeroAstro and CSAIL, where she leads the Collaborative Learning and Autonomy Research Lab (CLEAR Lab). She earned her Ph.D. in EECS at UC Berkeley in 2023 and her Bachelor's from MIT in 2017. She is an Apple AI/ML Ph.D. fellow, a Rising Star in EECS, an R:SS and HRI Pioneer, and received the best paper award at HRI 2020 and the Emerging Research Award at the International Symposium on the Mathematics of Neuroscience 2023.

​Free food and Q&A to follow.
See you there!