Should Artificial Intelligence Governance be Centralised? Design Lessons from History