Pat Hayess nave physics papers were highly in uential in the 1980s and 90s, inaugurating the eld of qualitative reasoning, inspiring the CYC project, and laying the foundations of the semantic web. Back then, the underlying motive for studying common sense physics was the development of human-level AI. But this grandiose aim slowly faded into the background of mainstream AI research, and has only recently been revived, under the new moniker of arti cial general intelligence (AGI). Nowadays, AGI is being pursued through the methodology of neural networks, an approach that was anathama to the logic-oriented common sense reasoning community that arose in the 1980s. In this talk I will examine the importance of common sense physics for contemporary AGI research, highlighting a number of insights from AIs past that are still relevant today.