Governments should restrict artificial intelligence development to prevent existential risks