NAB
article here
The ability for GenAI to generate a movie from scratch has been in the back of Hollywood’s mind, but remained a future prospect. Executives are being urged to think again by Doug Shapiro, an analyst whose track record includes anticipating the rise of Netflix.
“The future is you’re going to have holistic platform where you can start with an idea and end up with a movie. AI will solve the video problem,” he told a Media, Internet and Communications Conference hosted by research firm and Moffett Nathanson.
“I thought this process was going to happen over the next five to 10 years, and now I think it’s probably more like three to five.”
One reasons why he says he misjudged the pace of AI’s impact on media is the exponential scale at which large language models advance.
“As the size of the training [data] set goes up and the number of parameters goes up, and the amount of compute goes up, the loss rate of these things continues to fall. Basically, regardless of the model architecture itself, scale solves problems,” he says. “I didn’t appreciate that.”
There are huge implications that arise from this and Studios as currently constituted are not best place to survive them, Shapiro warned: “AI is more of a threat than an opportunity.”
Previous disruptive forces, such as the impact of streaming on distribution, took a lot longer to come about.
“You had to build out broadband infrastructure. People had to adopt broadband, people had to adopt streaming,” Shapiro points out.
But now YouTube is the number-one streaming TV service in America, if not the world. That upsets the whole apple cart of studio economics.
Studios are stymied in their ability to move with the pace of AI. That’s because of “an incredibly complex ecosystem in Hollywood,” and a fear at the executive level of rocking the unions by introducing AI too fast and too far.
“All of the studios are taking AI very seriously but they are very leery of doing anything that would be perceived to reduce labor demands.”
In addition, Shapiro says, studios are in no position to force talent, particularly creative heads like directors, to use AI to cut costs.
The result is a piecemeal approach to using AI in areas like localization, concept art or VFX which may create 10-15% savings, in Shapiro’s view, while outsiders adopting AI wholesale “and have the potential to see many more dramatic savings than that.”
He is not saying that efficiencies can’t be made, and picks out animation is a genre likely to be the first to change.
Former Disney and DreamWorks chief Jeffrey Katzenberg has said that AI can eliminate as much as 90% of the cost of making an animated film, and Shapiro doesn’t disagree.
“I think animation is one of the first places you’re going to see AI, both because of the labor demands and because the bar is a lot lower if you don’t have photorealistic humans.
“If you have a talking cat or something, then it’s a lot easier to do. There’s no uncanny valley with cartoon cats.”
But AI will also see slash the cost of blockbuster live-action movies. Shapiro does the math and reckons the 15-25% of average movie budgets are allocated above-the-line costs, 50% is below-the-line, and the remaining 25-35% is post, most of which is VFX.
“If we assume that you still need all of the above-the-line creative talent, that means that the rest of that is up for grabs, seventy-five to eighty-five percent. And right now, if you take the average budget, when you figure out just the below-the-line and post costs, it’s probably like $1 million a minute for most big movies,” he calculates.
“The question is: over time, does that cost converge with the cost of compute? Could that be $100,000 or $10,000 or $10 a minute? I think that’s kind of the trajectory. That’s what happens if you bring Moore’s Law to this, as opposed to rising labor costs.”
Stepping further out even that above-the-line talent — your Tom Hollands and Zendayas — are under threat, if that’s what the audience wants.
“There are four scenarios [for] how much AI consumers will accept. It’s not a technological issue, it’s a consumer acceptance issue.”
Scenario one is that there’s some arbitrary level of AI that people don’t like. Shapiro thinks we can dismiss that because VFX has taken over the film business and people still go see Guardians of the Galaxy.
Scenario two would be that people draw the line at artificial humans. He thinks that’s possible.
Scenario three is that consumers draw the line at artificial ideas — that you need to have a “human in the loop” because the story is supposed to evoke an emotional response.
“And an LLM [large language model] may understand the semantic relationship between tokens, but it doesn’t understand emotions. So, will you always need a human to be able to overlay their own judgment on what is emotive or not?”
The last scenario is that there is no line. “In that case you just have this dystopian doom loop of infinite AI generated content, which does not sound like a fun place to live.
Shapiro then thinks that we’re going to fall somewhere between scenarios two and three.
“You’re still going to need real humans for a lot of stuff and you’re still going to need some kind of creative overlay, some human to apply their judgment.”
Disruption is inevitable and it will come from the bottom up. A loose analogy is Pixar’s impact on the animation industry. He says, “It really took Pixar coming in with the CGI Toy Story to shake up the industry and herald the end of hand-drawn animation. The studios are probably going to be mostly dragged kicking and screaming into this world.”
Studios do still have some aces up their sleeve if they play their cards right.
“There’s going to be such a volume of crap. The vast, vast majority of AI content is going to suck.”
Execs have to work out what is still scarce and therefore has value as content itself becomes infinite. The answer might be IP, content libraries, and the ability to market, to create stars, and the ability to foster fandom.
“If you were to create a fine-tuned AI model that was trained on your own content, how quickly could you spin up new iterations of your existing library?” Shapiro asks.
“The business of making content is getting riskier. Studios manage their capital like a private equity investor, but we’re in a venture capital world. You need more swings at bat, at lower cost.
“You need to give the talent more equity. You need to use the network as a focus group, not as a marketing tool. You have to seed stuff onto the network, see what works, and then kill the stuff that doesn’t work and feed the stuff that works.”
Technology and development should work in lockstep, “not the technology guys in the basement” Shapiro said. “You have to really embrace these tools. I think it’s going to require a lot of cultural change in a very short period of time.”
No-one is immune, not even Netflix. Shapiro thinks the arch disrupter could become a disruptee, although the streamer is more nimble to adopting change than other media companies.
“They changed the model once. They probably have more leverage to try to strongly encourage talent to use these tools. And because they do all this stuff cost-plus and they actually line item budget every single production, they’re better empowered to exert pressure surgically inside the budget.”
But what if by 2028 or 2029 good-quality AI content is appearing free to view on YouTube or TikTok? What if those companies decide to extend out the length of videos?
“Do people at some point query why they’re paying $35 a month for this walled-garden service which I don’t always watch?
Hence Netflix, Amazon, Apple and other are moving into live sports — a genre that can’t as yet replace actual humans with an AI bot.
No comments:
Post a Comment