The Hollywood Reporter has a fanciful article up on How Artificial Intelligence Will Make Digital Humans Hollywood’s New Stars. With the poor performance of this summer’s crop of films, I can believe Hollywood would be willing to try.
Script supervisors, editors, CG artists and actors all had better look out: “It’s all over by 2045 — we are no longer running the show.”
Seems like a statement that needs some justification. Who the hell said this, Chicken Little?
Imagine if, following Paul Walker’s death in 2013, a computer could have stepped in, watched all of Walker’s performances in the previous Furious films, learning the minute details of how he walked, talked and even raised an eyebrow. And then imagine that artificial intelligence took over and itself helped to create a digital performance for Walker’s character. This isn’t so far off.
I hope you’ve got some evidence to back up this claim.
Now that the use of computer graphics is commonplace in movies and TV, artificial intelligence may be the most important technology to emerge in Hollywood. Potential AI-driven applications — in which the machine takes over and can learn and think for itself — could function as script supervisors, take a first pass at film editing, even create performances either for digital characters that resemble actual humans or more fantastic CG creatures. Such technology could have a huge impact in the areas Hollywood cares about a lot: schedules and budgets, by shortening production times and bringing down costs.
Making bad movies more cheaply is not a good strategy. We have enough wooden, formulaic actors. Even the best AIs today are terrible at innovating. Scripting, editing, and performing are all things that require creativity. While I can imagine an AI being a great assistant to a creative person, and doing much of the tedious, repetitive parts of the job, I haven’t seen anything to indicate that true creativity is anywhere near for AI.
“Digital humans are increasingly important in films. But if your data is too sparse, it doesn’t look right,” says Chris Nichols, a director at Chaos Group Labs and key member of the Digital Human League, a research and development group. “We are using AI to fill in the gaps. The whole concept of AI is dependent on data training itself. The more data you have, the better the system works.” Such techniques have not yet been used in a mainstream movie. But, theoretically, AI could do the job of creating a digital “Paul Walker” faster and more economically than current methods.
But would such a character still be based on a human performance? Will there be motion capture? How will the character travel from the imagination of the director onto film (well, into digital imagery, at least)?
Right now, AI is best at filling in the blanks after a character already has been sketched out, either from an actual performance or by CGI artists. Another application could come into play in a movie like Guardians of the Galaxy Vol. 2, in which footage of current-day Kurt Russell and reference shots of his younger self were combined to “de-age” the character. AI “does sound magical,” says Nichols, but “it’s not exact data. It’s interpreted data — interpreting the missing parts.”
Which, compared to creating a believable character from scratch, is incredibly simple. When Deep Blue beat Kasparov at chess in 1997, people thought AI that could pass the Turing test was right around the corner. Then it took 20 years merely to create a program that could beat a go master. Compared to creating a believable human from scratch, go is child’s play.
Stephen Regelous, a SciTech Academy Award recipient for the development of Massive, an AI-driven software that was first used to create the huge armies in Peter Jackson’s The Lord of the Rings series, sees potential for AI in the animation world as well. “You can shave off tens of millions of dollars from the budget of an animated feature,” he says. In a matter of seconds, AI could animate a character, a labor-intensive process for a CG animator. ” With Netflix and other streaming services [adding to the volume of production],” adds Regelous, “AI can help produce animation more quickly, and they need ways to improve their efficiency.”
Animation software can be a great servant, creating wire-frame animation, allowing characters to be skinned, and doing the tedious work of in-betweening, creating the frames that lead from one key-frame to another. As more and more kinds of movement are captured, they can be used to create realistic animations. But if films simply replay stock patterns of movement, like the old Spiderman cartoon, overuse will wear thin, like the constantly repeated special effects shots in the original Battlestar Galactica series.
AI also could be used to create individual characters in live-action movies. “In Planet of the Apes, it could be an alternative to performance capture, [which was used so that Andy Serkis could play the hero Caesar],” says Regelous. “At some point, you’ll be able to create an actor that doesn’t know he’s not real.”
What? This is either complete guff, or he meant to say that “the audience doesn’t know is not real”. The idea that AI becomes self aware and doesn’t know its not real is pure science fiction, at this point.
That could pose some thorny issues in the not-too-distant future. “The question of the ethics — whether you should do this or not — is still there,” warns Nichols. Futurist-entrepreneur Elon Musk already has voiced fears of “a fleet of artificial intelligence-enhanced robots capable of destroying mankind.” That possibility was suggested recently when Facebook, which already uses AI for editing news feeds and targeted ads, reportedly shut down an experiment in which a pair of AI-driven robots ended up creating their own language that defied human comprehension.\\
Tales of the singularity.
“It’s all over by 2045 — we are no longer running the show,” predicts Regelous. “They will become so smart, and this won’t be about motion pictures. They will be curing cancer and fixing global warming.” But Regelous doesn’t envision a bleak future. “Anything significantly smarter than us is going to value us. It will think, feel love and value life,” he says, adding, “I hope I’m right. Otherwise, we are screwed.”
So in 28 years, the singularity will arrive? In 1984, I was at Simon Fraser University, studying computer science. There were many AI scientists in the faculty. I remember learning that in the 1960’s, their teachers had been sure that AI would be achieved in 10 years. I’m not saying Regelous is wrong, but I certainly wouldn’t go all in on his prediction.