Because people write plays for a variety of reasons, not just to "serve" the audience. We might write plays to intentionally provoke the audience, to make them angry, or sad, or otherwise feel some emotion they might otherwise feel for some surprising reason. We might write a play because we were seized by a "genius" and neither personal nor social goals adequately describe the reason. Or maybe we write plays merely as an interaction in some kind of entertainment "marketplace." In reality, these two different sorts of activities are interrelated in complex ways and the foremost goal of the artist (even the foremost goal of the viewer) might change from day to day or moment to moment.

Socrates talks about this, I think. A good doctor doesn't serve the patient. A good doctor serves the patient's health and this might actively piss the patient off. Sometimes an artist is a kind of social doctor (or they may aspire to be one).

Thinking of the complex social relation between playwright and audience as driven entirely by either the ego of the artist or the ego of the audience member flattens out the roles of both to such an absurd degree that discussion of the thing in question is impossible. Hence, doing so is "weird."

[EDIT]: Thinking about this has crystalized something which has been swimming around in my brain for awhile. I think a fundamental way in which the current internet undermines human beings and produces alienation is that people fundamentally need to be met with a degree of resistance from people and serendipity from the world. When we seek out art we are, in a certain sense, seeking deliberately to be given something we don't want, explicitly. When we forage for novelty, we do not want to be served up something "curated" for us, but something which we could not have anticipated on the basis of our previous habits. Building marketplaces for every conceivable kind of human interaction undermines this basic need on the part of human beings. Recommendation engines and curation algorithms undermine this need. Even an object like ChatGPT, in a way, can't meet it. When I talk to a human I want to be, in some small way, and not always, genuinely surprised by what they say. It is difficult for a machine which is trained to predict the next token to do this (it is obviously not impossible because LLMs (and other algorithms) know much more than a person and can thus surprise us simply by conjuring up that with which we haven't yet made contact).