With the rapid pace of advancement in the field of artificial intelligence (AI), this essay purports to accentuate the importance of corrigibility in AI in order to stimulate and catalyze more effort and focus in this research area. We will first introduce the idea of corrigibility with its properties and describe the expected behavior for a corrigible AI. Afterwards, based on the established meaning of corrigibility, we will showcase the importance of corrigibility by going over some modern

more »

... near-futuristic examples that are specifically selected to be relatable and foreseeable. Then, we will explore existing methods of establishing corrigibility in agents and their respective limitations, using the reinforcement learning (RL) framework as a proxy framework to artificial general intelligence (AGI). At last, we will identify the central themes of potential research frontiers that we believe would be crucial to boost quality research output in corrigibility.