Understanding the Understanding of Understanding in the Age of Machines

Understanding is treated as self-evident in most philosophical, scientific, and educational contexts, and is rarely defined even where it is invoked as the ultimate goal of inquiry. The question of whether generative AI systems, whether large language models or the newer world models, are capable of understanding is currently one of the most consequential debates engaging academics, engineers, economists, and policymakers, with the answer's economic value measured in the billions. Yet participants on both sides of that debate largely compare human and machine performance on more or less arbitrarily selected tasks, rather than asking what understanding is in the first place. This essay argues that this obviousness is itself a philosophical symptom rather than a philosophical fact, and that the recent debate over whether large language models and world models can be said to understand exposes the cost of leaving the concept unexamined. I distinguish three senses in which understanding is invoked: semantic understanding, of meaning within a language with articulable rules; ontic understanding, of components of reality achieved through modelling, including mathematical modelling; and a third, unifying sense in which the epistemic and the ontic are brought together, for which the hermeneutic tradition running from Dilthey through Gadamer offers the most developed vocabulary. I argue that aesthetic judgment, from Hutcheson's account of beauty to Baumgarten's founding of aesthetics as a discipline, functions as a further, underused tool for this unification, extending understanding beyond logic and rational argumentation alone. Applying this three-part distinction to the question of machine understanding shows that most participants in the AI debate argue past one another because they invoke different senses of the term without acknowledging that they are doing so. I conclude with a metaphilosophical proposal: that understanding of understanding is best pursued not by seeking a single definition, but by a pluralistic program of studying the intellectual tools, aesthetic and logical alike. This will support understanding without restricting it to any one domain of application.