gym-corrigible

As part of the 2024 Fundamentals of Artificial General Intelligence Safety course at University of Buenos Aires (UBA), I reimplemented the OpenAI Gymnasium Dynamic Obstacles including corrigibility parameters as characterized in Soares, N., Fallenstein, B., Yudkowsky, E. Corrigibility. Artificial Intelligence and Ethics: Papers from the 2015 AAAI Workshop.

I presented the final project and results at a workshop at the end of the course. Slides