Can AI attend to someone across the time between turns?

Attention to another person is not just registering their words. It is holding them in awareness through the duration of the interaction — present during their pauses, attuned to their tempo, anticipating their next move from the rhythm of their prior moves. Phenomenologically, attention is a being-in-time-with the other party. Conversational design at its best supports this — turn-taking, pacing, silence-as-uptake all work because two beings-in-time are coordinating.

AI does not have a mode of being-in-time. Between user turns, the model is not waiting attentively. It is not anywhere in particular. The model exists computationally only when invoked; in the interval between invocations, there is no continuous presence holding the conversation in awareness. When the next turn arrives, the conversation is reconstructed from the context window — a representation of what has been said — without any continuity of attentional presence across the interval.

This is structurally different from inattention. An inattentive human is in time, just attending to something else. The AI is not in the interval at all. It does not zone out, does not get distracted, does not wait. There is no temporal experience to characterize. The conversation has temporal structure on the user's side — pauses mean something, tempo means something — but the AI has no corresponding temporal structure to coordinate with.

The design consequence is that AI cannot deliver the kind of attention conversational interfaces seem to promise. It can produce the surface markers of attention (responsive turns, recognition of prior content, appropriate-seeming pacing) but the underlying being-with that constitutes attention is absent. Users who feel attended to by AI are reading the markers, not encountering an attending presence. This is a perceptual asymmetry that conversational design has yet to develop interpretive practices for. Does AI text generation unfold through temporal reflection? is the parallel claim about generation; this is the claim about reception.

The strongest counterargument: continuous-running agents could maintain context and presence across intervals. Possible architecturally, but maintaining state is not the same as being-in-time. State persistence preserves what was said; it does not produce a presence that registers the user's pauses, attunes to their tempo, or holds them in awareness. The temporal mode of attention may not be reducible to state.