Multiparty Dynamics and Failure Modes for Machine Learning and Artificial Intelligence

David Manheim

An important challenge for safety in machine learning and artificial intelligence systems is a~set of related failures involving specification gaming, reward hacking, fragility to…