I would also bet that sometime in the next 10 years, we'll have a masterpiece of cinema on our hands where the heavy lifting (visuals, sound, even screenwriting) was largely done by an AI, helpfully nudged and curated at important moments by human experts. Or, by just one person.
What I meant by "completely done by AI" is that AI is doing the vast majority of the heavy lifting. Sound, visuals, script and ultimately humans are just acting as the director of that AI.
In otherwords, a masterpiece of cinema created by one person and AI prompts. Masterpiece being judged by the above success criteria. I won't accept some spam film that an AI magazine touts as being a masterpiece.
1) Everything. A single prompt will generate a full-length, high quality movie.
2) One person will be able to spend a few weeks or months to produce a high quality movie using purely AI generated visuals and audio, with at least part of the script written by AI.
3) AI will never replace some aspects of high quality movies, although it's not quite clear yet which aspects. It could be writing, acting, directing, or something else.
4) AI will never replace most aspects of high quality movies.
5) Society will rebel against any form of AI in movies; it doesn't matter how good AI gets, nobody will watch movies touched in any way by AI.
My guess is 2.
My pro-AI director friends tell me this is ultimately what they've been doing with humans all along. Sometimes he humans don't give them what they're looking for, so they ask again. And they have to fit within logistical and budgetary constraints.
David Lynch The Grandmother would be considered a "masterpiece" by this definition.
https://www.youtube.com/watch?v=4y0rYWVcxF4
Anyone could make something along those lines right now with AI tools. The ratings is because of what the fame David Lynch achieved after making this.
AI video is going to stall bad because it is just too expensive and what we have now is complete trash. Sora is such a massive disappointment to anyone who was interested in doing exactly what is being described.
Okay, so that's a risk of false positive success. It doesn't mean the bet is pointless, just that it's not the ultimate metric.
I also agree that AI will stall which is the point of the bet. I also don't think an AI recreation of the Grandmother would see critical acclaim. Lynch already did it.
I don't think LLMs can write nuanced character arcs, so let's not include them.
On the subject of the visuals being completely AI, we need to be able to steer the video with more than just text prompts. Do you remove the possibility of using motion capture performances, compositing, or other techniques?
I think we'll see 100% non-photon, non-CG visuals. I just think those performances will be human and the films will have a very human touch.
If you can make that adjustment, then I think we have a bet.
AI is just a tool. And artists are going to use the tools that can get the job done.
Yes I exclude that, because the primary reason to say "We'll be making more long-form, quality content per month than entire Hollywood production years." is that AI has eliminated or vastly eliminated the need for human actors. I'd accept a model trained on motion data or whatever, but I do not think something that augmenting that visual input data counts towards actually reducing production costs and speeding up the process of creating media.
I'd accept modifications to the bet that would still allow for rapid media production. If the human staffing is virtually identical to what it is today then that's not AI actually reducing costs. Hence, AI needing to do the majority of the labor.
For example of what I'd accept, a 2 person team that creates a 20+ minute ensemble film in less than a month or 2 that meets the success criteria above. I'd reject it if the film is "Watch ted go insane in this room" (I think for obvious reasons).
> I think we'll see 100% non-photon, non-CG visuals. I just think those performances will be human and the films will have a very human touch.
We already have that AFAIK. But again, I don't think that's a huge cost or time savings.
> AI is just a tool. And artists are going to use the tools that can get the job done.
I agree, it is a tool. I disagree with claims of how much content it will ultimately enable to be produced.
So humans steering diffusion is off limits? No Krea, no Invoke, no articulated humans?
It's like you're taking away Premiere or Final Cut here. Text prompts are not the currency of AI film. Controllability levers are essential to this whole endeavor.
> I do not think something that augmenting that input data counts towards actually reducing production costs and speeding up the process of creating media.
You haven't spent much time on set, then. An animator can do a performance capture on their webcam and adjust the IK. That's way different than booking a sound stage, renting an Arri Alexa and lenses, and bringing out a whole cast and crew. Set dec, wardrobe, makeup, lighting versus the moral equivalent of a Kinect and a garage studio.
My 6 AM call times, early mornings climbing up to the top shelf of the prop house to grab random tubas and statues, and signing countless legal forms and insurance paperwork all beg to differ with your claims here.
> AI has eliminated or vastly eliminated the need for human actors.
I don't think it necessitates this at all. Kids are going to be flocking to the media to turn themselves into anime VTubers and Han Solos and furries and whatever they can dream up.
Artists want to art. They're going to flock to this. We're going to have to open up the tech for that reason alone.
I'm sure fast moving marketers and the cottage industry of corporate workplace training videos won't use humans, but the creative side will. ElevenLabs is great, but there's also a reason why they hired Chris Pratt, Anya-Taylor Joy, and Jack Black in the Mario movie.
> For example of what I'd accept, a 2 person team that creates a 20+ minute ensemble film in less than a month or 2 that meets the success criteria above.
I'll posit this: a two person team will make a better Star Wars, a better Lord of the Rings, a better Game of Thrones. An ensemble cast of actors piloting AI diffusion characters (or whatever future techniques emerge) will make a film as well acted as Glengarry Glen Ross. Perhaps even set in some fantasy or sci-fi landscape. I bet that we'll have a thousand Zach Hadels, Vivienne Medranos, and Joel Havers finding massive audiences with their small footprint studios, making anime, cartoons, lifelike fantasy, lifelike science fiction, period dramas, and more. And that AI tools will be the linchpin of this creative explosion.
If that happens in the next 10 years and we judge "as good as starwars" using my above criteria. You would win the bet.
We on?
> we judge "as good as starwars" using my above criteria.
Just to clarify, this would be an AI film or "tv show" winning at traditional awards: Emmys (The National Academy of Television Arts and Sciences), SAG Awards, Oscars (Academy of Motion Picture Arts and Sciences), etc. Or traditional film festivals such as Sundance and Cannes, eg. winning the Palme d'Or. I would even be happy setting a threshold whereby a film or long-format show must win more than one award from several such institutions.
Maybe a preponderance of praise (20 or more) from major film and media critics like Roger Ebert (RIP), Leonard Maltin, Richard Brody, et al. could also be a criteria that must be met. Though perhaps that's a necessary condition anyway.
This all sounds good to me.
No special category, and I'm even willing to bank on it being a category that isn't the moral equivalent of VFX.
Let's remember to check back.
:D Probably the hardest part of this wager.
These things are "not AI".
It doesn’t need to win awards or prestige.
Someone needs to say “play me new episodes of the office or arrested development” and it needs to generate something that resembles the office or arrested development. People can have the noise on in the background, and it won’t matter if it isn’t quite coherent or super funny.
By that, I mean something where either the dialog or the video (or both) is completely done by AI. By successful, let's say something that wins a non-AI award (For example, an Oscar or Emmy) or receives something like a 70% positive review on rotten tomatoes, IMDB, or some other metacritic platform that is not specifically made for reviewing AI art.
I do not believe the AI will live up to the hype of "We'll be making more long-form, quality content per month than entire Hollywood production years."
I think we'll see long form AI, I don't think it will be high quality or even something that most people want to watch. The only people that will want to watch that sort of AI slop are AI enthusiasts who want AI to be amazing.