Perhaps the phenomenon of embodiment is not as monolithic as it seems.
We understand embodiment as physical, reliant upon senses, contact, proximity etc. and that the condition of embodiment confers the capability for practical, existentially sustainable interaction with a physical environment that a non-embodied AI would not be expected to have.
But what if it turns out to be possible, in the course of successfully representing those things within the AI, to reach a state which one might label as being synthetic or virtual embodiment.
Perhaps the latest developments in AI such as Sora, in their 'creation of relatively convincingly realistic virtual worlds' from their interpretation of a prompt combined with their training that acquaints them sufficiently well with the practical characteristics of the physical world, turn out, unexpectedly, to have a sufficiently workable appreciation of the nature of the physical world such that they ‘feel’ sufficiently embodied in a virtual world (or even under certain inherent potential constraints, the real physical world) to equip them to deliver the level of competence of interaction and insightful intuition that would otherwise have only hitherto come from physical embodiment.