To calculate or to understand? The ontology of artificial cognition between probabilistic models and embodiment

The evolution of Artificial Intelligence (AI) and the proliferation of Large Language Models are challenging the classical paradigm of symbolic manipulation, shifting the epistemological focus from semantic representation to probabilistic prediction in latent spaces. This paper examines the implications of this ‘subjectless’ cognition, asking whether a high predictive capacity can be distinguished from genuine understanding. It also analyses the concept of agency, arguing that authentic agency requires a form of embodiment and biological vulnerability; in the absence of these, AI exercises merely a derivative form of action. In conclusion, contemporary AI does not constitute a mind endowed with intentionality, but rather a sophisticated predictive mirror of human knowledge, which leaves to humans the exclusive responsibility for attributing meaning and acting in the world.