De-anthropocentrism as a Conceptual Alternative to Anthropomorphization in the Epistemology of AI

Abstract

The paper explores whether epistemic concepts – such as belief, knowledge, justification, possession of concepts, and doubt – can be justifiably ascribed to AI models. Various positions on this issue are considered, ranging from skepticism to radical optimism. The position defended in this paper, however, leans more towards the optimistic side of the spectrum. The main claim is that justified ascription of epistemic concepts to AI models is determined not by the system’s intrinsic properties, but by its ability to function as a normative epistemic agent as well as its capability to interact properly with other epistemic agents. This optimistic view aims to counter criticisms of unjustified anthropomorphism in our understanding of the design and functioning of AI systems. This is accomplished by advocating for the thesis of de-anthropocentrism. This thesis argues that there is nothing in human nature or in the nature of epistemic concepts that justifies the belief that only humans, due to their unique nature, are capable of possessing ‘real’ knowledge, understanding, reasoning, and possession of concepts. Remarkably, justifying the de-anthropocentrism thesis does not require a reconsideration of human nature or a transformation of views on AI. Instead, it merely requires a certain stance regarding the semantics of epistemic concepts. Thus, the justification for the de-anthropocentrism thesis primarily pertains to the realm of epistemology and the philosophy of language, rather than the study of cognitive science, or specialized research into artificial intelligence.