AI, Storytelling, and Truth
The Beatles Documentary and the Slippery Slope of Authenticity with Technology
Over the weekend, I binged all 8 hours of Get Back, the Peter Jackson Documentary on the Beatles. I really enjoyed it. It was hypnotic to watch the sessions of rehearsal and dialogues from a by-gone era. In particular, it gave someone like me (Elder Millennial), for whom the Beatles were always a little more mythical than tangible, a real sense of the personalities, dynamics, and talents.
One thing that struck me (other than the fact that I think McCartney and I have the same hair and beard shape) was the quality of the video, the pictures, and the audio. It wasn’t like watching a documentary from the early 70s — it seemed in many ways like a modern production and made the watching of it all the more accessible.
Restoration Helps Storytelling
I shouldn’t have been surprised — Jackson has become the go-to person these days for film restoration techniques ever since They Shall Not Grow Old, the 2018 World War I documentary that used 100-year old film. To use that film, they had to use every modern digital trick in the book. Techniques included rotoscoping, adding missing frames, updating the timing, filtering out grains, scratches, and colorization.
Jackson used many of the same techniques here, plus a couple of new ones, including using machine learning on the audio tracks to isolate and amplify vocals, guitars, etc.
Of course, no “upgrade” of this sort is going to be perfect, as some point out, and there are times when you know you’re seeing something that has been smoothed or softened or even slightly altered.
It’s a tradeoff we make to access the signal from the noise. As an audience member, it certainly has the intended effect.
The documentary got me thinking about that slippery slope of restoration in general and black-box technologies like machine learning and AI in particular. As these tools are used more and more (and get better and better), the lines are blurring between historical representation, interpolation, and interpretation.
One of my favorite examples of AI-in-film-restoration comes from the YouTube channel of Denis Shiryaev, a product director of Neural.love, where they restore (very) old films, such as New York in 1911 or Apollo 16 on the moon in 1972.
What’s interesting as well is not only the technological advances themselves, but the perception of technology that may or (may not) be behind the scenes.
In fact, in Shiryaev’s compilation video, he is very explicit that these videos are not historically accurate, and are only meant to serve as interpretations.
We are moving beyond restorations, as we aren’t just cleaning up dust specs here or fixing a broken splice there. The techniques fill in missing data from extrapolated datasets of millions of faces, millions of hours of related film and coloration data.
You can’t talk about these blurring lines without talking about deepfake, the technology behind digitally replacing faces with the likeness of someone else. If you haven’t been paying attention, the technology has become very good (Tom Cruise, Barack Obama, The Shining with Jim Carrey).
Earlier this year, the documentary Roadrunner, about the late Anthony Bourdain, sparked some controversy by using a “deepfake voice” of Bourdain to read his private journals.
What’s interesting about the rise of these examples is that for now, they still generate headlines and are stories in and of themselves. But a lot of people are (rightfully so) worried about the implications of a deepfake world, where we can’t readily identify authenticity, generated or altered content, or underlying intent. While it might be a challenge to create a deepfake of Barack Obama and pass it off as legitimate, that won’t necessarily be true for your average person and the cascading implications.
Taken to its logical conclusion, it seems like all media will correspondingly become a digital wasteland of trust, where, like with the rest of the Internet, we’ll be able likewise to believe and disbelieve anything that we see.
And my hopes aren’t too high on blockchain figuring this out.
Perceptions of Technology
In a completely unrelated venue in recent weeks, we saw the impact of perceptions around video AI technology in the trial of Kyle Rittenhouse, when the lawyers for the defense argued that it was using pinch-to-zoom on a video displayed on an iPad manipulated the footage. The full quote is telling:
“iPads, which are made by Apple, have artificial intelligence in them that allow things to be viewed through three-dimensions and logarithms[...] It uses artificial intelligence, or their logarithms, to create what they believe is happening. So this isn’t actually enhanced video, this is Apple’s iPad programming creating what it thinks is there, not what necessarily is there”
Yes, it’s particularly inane, and it’s unclear if this was a legitimate concern. But it’s not as if this belief that underlying algorithms are manipulating images is somehow unbelievable.
This concern (given weight by the judge, by the way) is a bit of a waterline for seeing how far we’ve come when it comes to not knowing and not trusting the underlying technologies (which we shouldn’t by default).
I don’t have a pithy conclusion to this — it’s both a fascinating technological landscape (check out Two Minute Papers Youtube Channel for incredible examples), as well as an inevitable minefield of ethically dubious opportunities.
All of this started with me wanting to say that I enjoyed the Beatles documentary. And that’s true — it was excellent.
But it also makes me feel like if we zoom in on what’s left after we open Pandora’s Box, that by using AI image enhancement, we’ll see that what we thought all along was going to be “hope” that was left, was in fact just “nope.”