Computational photography
The unintended impact of an unattributed statistic
There’s a niche genre of article I am always enticed by: investigations into unattributed statistics. I’m powerless to resist.
I think the first time I felt the rush that comes from reading that everything you’ve been told is a lie was Vanessa Friedman’s 2018 New York Times piece “The Biggest Fake News in Fashion”. In it she uncovered that the then-oft-cited “fact” that fashion is the second-most polluting industry, after oil, wasn’t actually the case.
I felt that same rush last year, reading a post on Matt Klein’s ZINE. He was tracing the source of the prediction that “as much as 90% of online content may be synthetically generated by 2026”.
I’d seen this stat quoted by the World Economic Forum, Axios and a lot of people on Reddit, but where did it come from? Anywhere at all? I was on tenterhooks.
The evidence trail started in a Futurism article, which cited an Innovation Lab of Europol report. “Not a shabby source,” Klein thought, until he accessed it to find that particular statistic had been removed. An archived version led him to a footnote, which led him to a book on deepfakes by well-respected expert Nina Schick, which finally led him to a conversation she’d had with Victor Riparbelli, co-founder of $2.1bn synthetic media startup Synthesia.
What he’d actually said, per Schick, is that synthetic video may account for up to 90% of all video content in as little as three to five years. So, what does Riparbelli think of what this potentially off-hand comment had snowballed into? I took the investigation straight to the source… I asked him.
“I mean, it’s kind of fun when I get to see it in the wild,” he laughs. Back when he said it, in 2019 or 2020, he says, “it was definitely a non-consensus view”, other people didn’t believe it, “but, I think, now it’s coming out to be true”.
“It doesn’t get attributed to me anymore,” he continues, joking that he should get his lawyers involved. “I guess I’m proud that I was somewhat right.”
Riparbelli doesn’t think 90% of the content we actually interact with online will be AI-generated though. He says there will be a huge “deluge of content”, way more than anyone could consume, and the best will float to the top (leaving a lot of AI slop down at the bottom).
But 2026 is next year, I point out to him, that’s pretty soon for 90%. “It depends on how you define synthetic right,” Riparbelli says, suggesting that Instagram filters on selfie videos technically make them AI-generated. Even iPhone images are adjusted by AI, he explains.
I was skeptical, until I watched the Apple Event earlier this week, and the images being produced by the new Air and 17 Pro models looked just too perfect.
“We apply deep learning models for demosaicing, which constructs the image from the raw output,” Patrick Carroll, manager of iPhone camera architecture, said on the livestream. This kind of “Photonic Engine” technology has been being used since the iPhone 14, but now that the Air has only one rear camera, Apple is relying more on AI to fill the gaps. As John Ternus, Apple’s SVP hardware engineering confirmed during the event: “With the power of computational photography, this incredible system enables the latest signature features of the iPhone camera.” One of those features is the new Bright style, which actually brightens the subject’s skin tone.
It made me wonder about the line between genuine and synthetic content, which – as an iPhone 11 and 35mm film camera user – had previously seemed quite solid. Not in terms of telling real from fake, that has been blurring for years. But generating something with AI still felt active to me, different to effortlessly snapping a photo of my dog or friends at a party. In a world where that’s not the case, though, Riparbelli’s prediction is more than feasible.
I did of course ask Victor about more than this one statistic, you’ll get to hear his thoughts on building in Europe vs the US and how he wooed Mark Cuban in a few weeks. Until then, here are your tabs:
Emotional alchemy
Creative technologist Tigris Li has developed an AI system that can turn the stories you tell it into bespoke fragrances. The project was born out of the question “what if you could smell people’s memories?” and Li’s belief that “play is essential for innovation, because it’s a universal language”. You can watch her talk through it in this video of the latest Tiat Place salon
The art of persuasion
Humans are pretty easy to influence: we are more likely to follow instructions if they come from an authority and more likely to buy something if it seems scarce. These are just two of Robert Cialdini’s principles of influence, which – professors at Wharton have discovered – also work on LLMs. You can see all the (honestly quite entertaining) details here
Zendaya, eat ur heart out
Danger Testing’s latest app iChallengers turns your phone into a tennis racquet, complete with an impressive thwunk in every swing. Download it (if you have an iPhone) and experience the same childhood glee you felt when you picked up a Wii remote for the first time
Machined imagery
More than 100 years ago, a Welsh singer named Margaret Watts Hughes invented an unusual recording device, an Eidophone. She would sing into it, using the resulting vibrations to move pigment around into textural images. You can see them in this piece from the Public Domain Review
Hunger strike
Alys Key went down to the Google DeepMind offices this week to speak to the two guys quietly protesting outside. Inspired by a similarly intense caloric restriction happening in California, Michaël Trazzi and Denys Sheremet are aiming to increase attention on AI safety. Why’d they pick DeepMind? Because they thought Demis was the most likely to listen
Black market GPUs
Though Jensen Huang swears there’s “no evidence” that Nvidia chips are being smuggled, it turns out there’s quite a lot of it. Jonathan Stein digs into this emerging shadow economy, and a potential solution, in this piece for Transformer
Paper rocks
I’ve been messing around with Paper’s public alpha this week and the instant gratification is real. I made an animated Open Tabs logo in minutes. Get in there and play
