Automatism of the articulation of things an individual interprets at a much higher level is intelligent behaviour, and a high proportion of synchronous communication is based around establishing/clarifying things the other party is already supposed to have in mind. Someone isn't demonstrating greater intelligence because they can't parse the basic greeting as a single concept so have to retrospectively analyse each individual word of the sentence to infer meaning, because they don't have an existing opinion on an issue (or haven't started thinking about it when it first became the subject of the conversation a minutes earlier) or because they have to iterate through number of possibilities to guess at something somebody else simply recalls or already figured out.

GPT-3 determining the most probable of millions of possible followup word sequences each time the other party emits a new word by comparing it with a vast corpus of mostly irrelevant data isn't reasoning at as high a level as the person who grasps as soon as the other person references $subject that what the other person wants is agreement, and thats as simple as human "automated" responses get (GPT-3 will get there too, via a more circuitous route)