Towards an Understanding of Default Policies in Multitask Policy Optimization