Apparently AI Idols Don't Work Because They Have No Narrative

Before we start

The Goyumyum below is fiction, dramatized into a hardliner for entertainment value.

In reality it was a few short questions and answers, and most of what follows is me taking those and rebutting myself, then fixing myself, over and over. But written as it actually happened it's just a monologue, and monologues don't read well. So every counter-argument that ran inside my own head got shoved into Goyumyum's mouth.

Treat it as a sparring partner in an argument, not a real person's real tone. Nobody actually talks in those clean guillotine lines.

---

What set this off

An argument broke out this morning while I was getting ready for work.

About half of our household conversation these days is RESCENE. Last time, watching RESCENE videos rolled me all the way to the med school rush. Today it went a different direction.

What I threw out was this. If AI moves into the idol industry, doesn't it all just become mass-produced? You can generate the face, the voice, the songs.

Goyumyum cut it off immediately.

"AI can't make a narrative. And idols need a narrative. So it doesn't work."

Clean argument. Major premise, minor premise, conclusion. But I couldn't agree with it. I couldn't point at where it snagged — just that feeling where the explanation works and the conviction doesn't. I treat that sensation as my most important signal.

So I kept chewing on it while drying my hair. What follows is in the order it happened.

---

1. Do idols actually need a narrative?

I'll grant that RESCENE is a narrative-type group. The trajectory from mom-and-pop idol to here is itself the content, and that trajectory is genuinely why I got into them. Woni being adored for Karina-tier looks while staying totally down-to-earth. Minami crossing over from Japan, going through all of it and still staying thoughtful. Liv's inexhaustible energy. May's chattering. Zena. Once you know all this, the videos just look different. That's narrative.

But that's evidence that narrative is powerful, not evidence that narrative is required.

So I threw out counterexamples. Jang Wonyoung, Karina, Sullyoon.

To be clear, I'm not saying these people have no narrative. They all do. What I asked was different: was narrative the path by which the public came in? Mostly not, I'd say. Something overwhelming in the moment they hit the screen comes first, and the narrative attaches after. The order is reversed.

So narrative is closer to a retention condition. Its holding power is real. But it isn't an entry condition.

Which leads here. Is there a reason a virtual idol — pretty, good vocals, sociable, with the producing optimized by AI on top — couldn't clear the entry condition? Hitting what the public likes is exactly what these systems are best at.

Goyumyum said "that's a shell, it won't last." Maybe. But that's a durability claim, not a necessary-condition claim. The proposition quietly changed.

---

2. Can anyone even tell whether they exist?

This is the part I was more curious about.

Most of the public consumes idols only through a screen. Even concertgoers are a minority of total consumers. I watch RESCENE on YouTube and Instagram, full stop. I've never seen them in person.

So how much does it change my consumption experience whether the thing behind the screen is actually breathing?

Honestly, I don't know. The RESCENE I know is the RESCENE inside edited video. I know those are fragments off real people, but I have never verified that. I just know it to be true.

And the answer is already half in. Virtual idols are running. Miku has been running for well over a decade. If physical existence were a necessary condition for consumption, that market would never have opened.

Goyumyum's rebuttal: "you know it's virtual going in, that's different from pretending to be human."

I think that's a fair point. But then the condition shifts from existing to not deceiving. Which is also a different claim.

---

3. If narrative comes from living, then what is living?

This is where it got most interesting.

The root of Goyumyum's logic is: AI isn't alive, so it hasn't undergone anything, so it has no narrative. Fine — then we have to define being alive.

So I threw out a thought experiment.

Say we build something like a quantum-resistant data device. Give it volatile memory, the way a human brain has it. Storage isn't perfect, things blur with time, it forgets. Then, the way humanoids and androids split, give it a physical body and plant a survival drive. Battery, component decay, damage, irreparable failure.

Under those conditions, it lives. It tries something and fails, it forgets, it breaks down, and it keeps doing things anyway.

Is that trajectory a narrative or not?

Goyumyum's answer: "that's a simulated narrative, not a real one."

But the moment you answer that way, the condition for narrative slides from did it undergo something to is it made of carbon. That's not a rebuttal, that's changing the definition. I think that was the single most important point in the whole conversation.

Not that I'm confident here either. Traced back, it comes down to this question:

Is narrative uniquely human, or is it something that occurs once certain conditions are met?

If the former, Goyumyum is right. If the latter, meeting the conditions is a matter of time.

The condition I've tentatively landed on is not being able to go back. Can't recover what you forgot, can't rewind time you lost, can't roll back a failed choice. In a game with save files, a character dying isn't an event. You just reload.

And what AI lacks isn't experience — its logs are more accurate than a human's. What it lacks is the irreversibility. But that reads less like something we can't build and more like something we have no reason to build. (Hypothesis. Nothing verified here.)

---

4. So what about producing? Here it's much simpler

The narrative argument keeps drifting into philosophy, but drop down to the industry side and the math sharpens. It split into two cases.

Case A — the producer gets abstracted away

Hatsune Miku.

Miku has no agency. No head of planning, no fixed narrative. Someone dropped a voice engine out there and collective intelligence wrote the songs, built the character, and attached a narrative on its own. The producer's seat got replaced by something abstract rather than a person, and value as an idol happened anyway.

What makes it interesting is that the production side supplied no narrative at all and narrative appeared regardless. It may be more accurate to say it appeared because none was supplied — nothing blocked anyone from attaching their own.

You can rebut with "Miku is an instrument, not an idol." But keep narrowing like that and you land on "an idol is a thing a human does," and the argument goes circular.

Case B — replace only the backend with AI, keep the human idol as the output stage

This one is more realistic and probably arrives sooner.

Composition, arrangement, choreography design, concept planning, trend analysis, content scheduling, short-form editing, targeted marketing. Replace every element and object inside the entertainment company with AI, and keep only the human output stage that sings and dances on stage.

Is that company competitive?

My claim was this: a 100%-AI producing operation could already outperform a mediocre no-name small agency.

The reason is simple. A small agency's real bottleneck isn't capital, it's number of judgments. How many songs got reviewed, how many concept drafts got thrown out, how many cycles of response data got run. I'd say a large share of the gap with the big agencies comes from that. And AI is precisely a tool for raising that count. What used to be three options built on instinct becomes three hundred, filtered.

Goyumyum didn't push back much here. The response was more "a human still has to make the final pick," which I agree with. Whether that human has to sit inside the agency is a separate question.

---

One side thought

Something else snagged while writing this up, but it came from the write-up rather than the conversation, so I'll keep it short.

Human idols mess up. Scandals, collapsing condition, teams breaking apart. From the company's side that's risk; from the fan's side it might be collateral for immersion. Success only becomes an event if failure was possible.

Made me think what a perfectly engineered AI idol lacks might be this rather than narrative.

But I don't see it as a big axis yet. Miku is close to a counterexample. Just noting it down.

---

Open questions

• Is narrative uniquely human, or does it occur once conditions are met? That was the real crux and it never resolved.
• If the public can't verify existence, does not deceiving become the condition? Then does an AI idol that declares itself AI from day one pass?
• If Case B becomes standard, what's left of the human idol's role? This one goes past structure into where people actually get hurt, so it needs a separate look.
• Has the Miku model succeeded anywhere other than Miku? If it's a sample of one, it might be an accident rather than a model. (Needs fact-checking.)

Started while getting ready for work, ran until I got on the train. I was playing both sides and neither one has folded yet.

Which isn't a bad place to be. If it had tidied itself up this morning there'd be nothing left to talk about tonight.

We'll probably go at it again tomorrow morning — and by then Goyumyum might actually turn into the hardliner. Dumping it here for now.