I'm pretty sure that simple story generators are already a thing. Maybe not full on scripts, but there is stuff to build off of.
I know that there are AI which can "read" a story and extract some of the defining qualities and themes. Look at MIT's Patrick Winston who was/(is?) working on AI symbolic understanding.

Writing a novel is a lot more formulaic than a lot of people think.
Jim Butcher has a pretty fun story about his college professor, who is also a prolific romance novelist. Jim didn't want to follow her advice because he thought it would lead to generic feeling crap. To prove her wrong he followed her advice and wrote the first book of The Dresden Files. He's had a great career so far. For a while he was publishing two books a year.

There's NaNoWriMo, where people try to write 50k words of a novel during November. Pretty much any competent writer can bang out a generic script or story in hours or days. There's a formula, the trick is hiding it and twisting it, and making it less generic. The sheer amount of shit out there that's just Shakespeare with a wig and glasses is overwhelming.

There probably just has to be more guidance in a story writing AI.

Break it down into the elements of a story. You have the seven basic stories. Start there. There are probably character archetypes. There are relationship archetypes. Mix and match.

Central plot, central characters who each have their primary archetype and secondary/tertiary qualities. They have their central motivation. They have roles to fill.
The characters try to achieve their goals using the resources at their disposal, according to their parameters.

Statistical and symbolic analysis would probably tell us if there are reasonable approximations for how long each narrative description is, how long conversations should be, the patterns of conversational back and forth, how to divide the plot structure...

If I sat and thought about it, I could probably list a few dozen or more parameters to analyze, and then it just turns into an ad-libs kind of thing.

It'd probably even help to tie sevel AI together. Like a city/building design AI to get an understanding of a world, and interrogating generated content to feed to a character.

So, maybe in a roundabout way making a compelling story bot might lead to more generalized AI, depending on how it's done.

We may not get deeply philosophical and emotionally complex stuff at first, but it's good enough for a flashy action movie or a heist movie. Good enough to make an episode for the average CW series.

I don't think near-future generalized AI is going to be plausible without structure anyway.
Look at a human. It takes weeks for an infant to be more than an eating pooping machine, months to be functional enough to start engaging and babbling, months to start crawling and walking, potentially years to start talking, and their cause-effect understanding is dubious at best. Children have lots of overfitting and beliefs based on false causation. It takes years for people to learn to read, and some never pick it up well. Some never learn more math than arithmetic. It takes most people decades for the person to reach their full potential. We are expecting AI to do it in days? A few years? With piddling resources and limited run times?

AI is already more functional in many aspects than a typical human. With posits and the new hardware coming down the line, and the big dollars being increasingly spent, it won't take long to see an AI tell a coherent one page story. A full script will come sooner than later, even if it's a boring one.