Ramana Kumar

Discovering Agents

2022 · arXiv (Cornell University)

Causal models of agents have been used to analyse the safety aspects of machine learning systems. But identifying agents is non-trivial -- often the causal model is just assumed by the modeler without much justification …