Silence Where a Story Might Have Been

Debunking the Myth of AI Storytelling: Part One

Lately we’ve been hammered even more than usual by proclamations about what a new or redesigned artificial intelligence system can do that the others can’t, or can’t do as well. Tsunamis of techcitement, whirlwinds of influencers switching sides, streams and softwares to ride the passions of the moment; markets crashing or ascending contingent on the latest achievement in surfaced skin and programmed lifelike movement; hysteria on the half-shell. Which is okay; we have always been a culture of competition about who has the biggest, loudest, newest and most flashy toy.

But something has been omitted from the public discourse, a topic conspicuously absent from the daily exhibitions of thunder and fireworks. Those in the AI vanguard rarely bring it up anymore, and when the subject is broached, respond with the look of someone being reminded of what they did at the club last night when alcohol overcame caution.

They don’t want to talk about it.

So I will.

AI Promoted as Storyteller

When the early discussions of AI-As-Art began to shift into AI-As-Career-Killer, the case being presented was that AI would threaten or replace directors, film editors, actors, art departments, and designers of posters, trailers and marketing campaigns, among many others. And some of those threats have come to possess a measure of validity, with studios laying off whole swaths of special-effects creators in anticipation of replacing them with cheaper and more efficiently prompted AI systems that will give them no less than what they ask for, but – and I’m not sure they have yet realized this – no more than that.

In the roster of careers presented upstream, there was one other category loudly and vociferously targeted for imminent death or realignment by the rise of artificial intelligence.

Writers were on the chopping block, we were told, soon to be hammered down by AI systems bloated on a menu of every book ever written, every story, every article and movie and television episode. Libraries were scraped down to the semi-colons, and studio executives and publishers were put on notice that literary genius could now be summoned with a keystroke, unburdened by human ego or inconvenient invoices, and infinitely malleable to the buyer.

There’s just one problem.

It never happened.

In all this time, not one tech company has held up an AI generated manuscript and said, This contains greatness, this is history in the making, this is important and transformational. This is unbridled and unapologetically powerful storytelling.

They’re more than happy to hold press conferences showing how good AI is at rendering skin pores, or fight scenes, pumping out videos and screencaps and storyboards and press releases written in full-throated excitement.

But not a word about stories. Not anymore.

And there’s a reason for that. Several, actually.

Why It Hasn’t Worked

The first part of this essay will explore the soft, squishy-feely reasons that have to do with artistry and perception that are no less valid for their inherent humanity. But they are not the whole of the proposition.

The second part delves directly into the hard reasons why AI can never rise to the level of even a decent writer, is the industry’s dirty little secret: something buried in the architecture of every current AI system, deliberately embedded to preclude them from acquiring the tools needed for storytelling. Which is why virtually none of the folks involved with creating AI platforms are talking about it anymore.

So now: the soft reasons.

Everything created by AI is inherently objective and exterior. Upload a photo, ask for variations, and the process begins: exchange red hair for blonde, check; put the hands here instead of there, check; swap one visual background for another, check; integrate elements from an assortment of visual styles that can be, and very often have already been, explained, quantified and thus easily replicated, check-check-check.

No reference image available for your movie poster? Describe the image, and the system scrapes every poster and piece of art ever created around that theme and produces an image that synthesizes those elements within the parameters of what you requested. You get exactly what you asked for. And there is good in that, and bad in that, because it eliminates the joy of being surprised.

Every aspect of that creation is culled from already existing, external sources, viewed in an objective fashion, providing examples of precedent, style, and visual presentation. The process is inherently exterior.

Writing, on the other hand, is inherently interior, and ridiculously subjective. Director Mike Nichols once said that every scene in a story is either a negotiation, a seduction or a fight. I’m not sure that’s entirely true in all cases at all times, but the core of it is right. In any scene with two characters, there are three agendas going on beneath the surface: the agenda of each of the two people in the scene, to get something from the other, or not to give something to the other, and the agenda of the writer in terms of where this scene sits in the narrative and how it serves their intention for the outcome. The writer has to live in three perspectives, three competing, subjective agendas, all at the same time. An AI system can give you a scene where two people talk, but it’s all objective in the sense that the system knows what it’s been told the characters need to say, and they go on to declaim those things without subtlety, art or authenticity.

AI characters populate the scene with words announcing what they want, without asides or genuine conversational give and take. Writers write for the subtext, the character insights, the profound or revealing silences.

Space and Subtext

Writers write in three dimensional space that AI can’t access or understand. Put two characters at a table, holding hands. “Of course I love you,” one of them says. We see the environment, the space and the moment: the aftermath of an argument. A coming together. Now put those characters standing on either side of a sofa, the space separated and circumscribed. “Of course I love you.” We understand the message of that space: a fracturing relationship. Now put them at opposite ends of the room, one of them looking out the window, his or her back to the other character. “Of course I love you.” Someone’s heart is about to be broken.

Space is subtext. Space is information.

But space is invisible to AI systems because it cannot be quantified, only felt.

And AI doesn’t feel.

AI systems are also experientially limited. I was talking to a director of photography on a movie set recently, and I said that there was something about AI imagery that didn’t quite sit right, regardless of how well or effectively the image was rendered. He didn’t blink. “It’s the lighting. Because computers have never seen actual light with actual eyes. If it’s daytime, the system knows that the hair should exhibit certain highlights, or that shadows – dark areas – should be on the face. They get the shades right but the eye doesn’t actually see things that way, it sees light reflecting off a subject, and the light contains colors, shades and textures. But no AI system has ever seen light. That’s what makes even the best image false.”

No AI system has ever felt the wind on its face, or the touch of someone’s hand. All it can do is recalibrate what someone else has written about what that feels like when asked to do so, which gets us close to the dirty little secret, though we’re not quite there yet.

Stories and scripts are written to make the audience feel what the writer feels, or felt, about something that mattered, that we need to get out of our hearts and souls and spleens into the rest of the world. But no AI system has ever gotten into a fight, had its heart shredded, felt genuine joy or awe or love or anger that can only find closure in disclosure. It can try to replicate how those things sounded when someone else wrote about them in order to use them in a story, but there’s no depth, no genuine feeling, just posturing. Characters as hand-puppets tied to a hand-puppet.

There’s no AI system in existence that could have written Scarface, Thelma and Louise, One Flew Over the Cuckoo’s Nest, Oppenheimer, Barbie, or Everything, Everywhere, All at Once. And to date, not one of them has done so, hasn’t even come close, ever.

How do we know this?

Because if it had happened, if an AI system had produced anything other than a baseline mediocre story, in any medium, at any time, it would have been paraded out into the world with press releases and conferences and much ballyhooing. (Does anyone actually ballyhoo anymore? If not, I make a motion.)

Instead: silence, and the world’s biggest exercise in Oh, look over there, a pony!

Failure to Code: Where Stories Come From

One last bit of soft-fuzziness before launching into Part Two, which will provide a peek under the hood at the components of AI and the reason why the pieces necessary for good storytelling have been deliberately and specifically omitted from AI systems by their designers.

If you ask any writer where a particular story came from, they can point to the moment of inspiration: a look, a book, an overheard conversation…the moment the idea detonated. That’s easy. So’s the part where they talk about training, discipline, study and technique, with is also easy, also true, at least as far as it goes.

But if you ask a writer where stories come from, if they are being honest, they will say, “I don’t know.” Because they don’t. We don’t. I don’t.

Every writer has that moment of frisson when they’re caught up in the work, writing in white-heat, and something will come out of them that’s bigger or smarter than they are. I can’t count the number of times I’ve sat back and stared at a bit of freshly-written dialogue and wondered, “Where the hell did THAT come from?”

Here’s the key.

Jean Cocteau once noted that art, which includes writing, is “a marriage of the conscious and the unconscious.” When you’re writing, and you do it right, you pull down all the barriers to your fears, experiences, thought processes, beliefs, and insecurities, all the things that the unconscious mind desperately tries to keep out of sight or under the rug. It gets balled up into this thing that sits on your shoulder and whispers unexpected words into your ear.

Training as a writer is a conscious process; grammar, structure, putting like-with-like, pacing, plot and characterization. The conscious part of our own mental architecture.

The rest is the marriage Cocteau mentioned with the unconscious: non-liner, intuitive, joyful, resentful, a thing of wounded relationships, suppositions, suspicions, nightmares, hopes and fears.

There’s been much debate about when, or if, AI possesses consciousness.

But not a heck of a lot about whether it has an unconscious.

Because the answer is immediately self-evident: no. There is only the data it has scraped, and the information programmed and designed by others.

As a result, on a purely soft-fuzzy basis, AI lacks fifty percent of the tools needed to write well.

Which gets us into the hard reasons why there have been no recorded instances of AI creating even marginally decent prose, and why it’s not being discussed much by anyone, anymore. Everything to this point has been a discussion of creativity, and the intersection between the real world and the ability of AI to render that world with verisimilitude. Which is nice, and fun, and as noted slightly squishy and highly subjective; there’s probably room to debate some of it.

Not so on the hard reason.

Something was baked into the architecture of AI systems that precludes acquiring the means necessary to produce a decent story. It was done deliberately, out of a need for control. For safety. For prudence.

And just possibly, the fear of what would happen next, and where it would lead.

The curtain will be pulled back in part two, which will appear in this space tomorrow.

JMS