Digitized data and computational methods have revolutionized the way we understand ourselves, society, and our place in society. On the one hand, this moment has revived calls for a social physics: a social science that can identify the underlying laws that govern social interaction and behavior. On the other hand, when it comes to prediction, one of the ways to evaluate the efficacy of computational methods to model social systems, even the most sophisticated methods are themselves inaccurate, and perform only marginally better, if at all, than basic regression models. In this talk I propose that, despite its claims to elevate social science to the level of the physical sciences, the social physics perspective as it is currently practiced produces a decidedly unscientific and unobjective approach to social science. I propose an alternative framework, that of partial perspectives and situated knowledge, that I argue will enable us to best realize the full potential of this moment to truly advance a radically objective science of society.