Generative AI is about to collide with Hollywood.
This is an extract from New World Same Humans, a weekly newsletter on trends, technology, and society by David Mattin.
📥 You can sign up here for the full experience in text and podcast form. 📥
*
This week, further glimpses into the halls of mirrors taking shape around us via generative AI.
UK-based voice technology startup ElevenLabs launched a new text-to-speech model that generates eerily pitch-perfect, human-sounding voices. Here’s a snippet:
I listen to a lot of audio books. To my ear, the voice reading Gatsby above sounds indistinguishable from those of the handful of actors — male, American, blessed with a soothing voice — who narrative most of them.
What’s more, the tool allows anyone to create a highly convincing voice clone in seconds, simply by uploading a few short clips of the voice they want to recreate.
And that’s what caused all the trouble this week. Within days, people had used the tool for all kinds of mischief, including using a voice clone of actress Emma Watson to read passages from Mein Kampf, and sending a cloned Ben Shapiro on a racist rant about Alexandria Ocasio-Cortez. Much of this content was shared on the infamous troll’s paradise that is 4Chan.
Three days after launch ElevenLabs withdrew free access. They’re now restricting access to the ‘build your own clone’ feature to paid users, and say they’re working on a tool that will allow for the near-instant detection of AI-generated voices.
The announcement echoed one made this week by The Big Player in generative AI:
OpenAI’s new tool will allow users to identify text written by a generative model, including by GPT-3.
This week, OpenAI announced that ChatGPT has hit 100 million users just two months after launch. The vast popularity of the tool has led to speculation that the internet is about to be hit by a tsunami of AI-generated junk content and disinformation.
⚡ NWSH Take:
The ElevenLabs story is a signal of the potent difference between really good and perfect when it comes to generated/deepfake content.
Just a few months ago publicly text-to-voice tools were generating voices that sounded good, but a little robotic. ElevenLabs elevated fidelity to perfect; cue the spectre of a million convincing celebrity says hateful things fakes.
No wonder, then, that AI detection tools are about to become big business.
Right now, these tools are in their infancy. Pretty soon, internet browsers will come with AI detection as standard.
The broader message here? New forms of generated content — including voice clones — are about to transform media and entertainment.
Back in New Week #100 I wrote on how an AI will voice Darth Vader in Disney’s Obi-Wan Kenobi series; this week brought news that AI startup Metaphysic — best-known for their viral Tom Cruise deepfakes — will deploy its technology to make Tom Hanks appear younger in his next film.
How long before a Hollywood film uses AI to reincarnate a much-loved star who is no longer with us?
But it won’t only be Hollywood and media giants that leverage generated media; new tools will mean new creative possibilities for all of us. One glimpse? Check out this person who automated the creation of a personalised podcast; he uses ChatGPT to collect and summarise stories on topics of interest, and ElevenLabs to read out the summaries using a clone of his own voice.
*
You've just read an extract from this week's New World Same Humans, a weekly newsletter on trends, technology, and society by David Mattin.
Also in this week's instalment:
🤖 DHL say new warehouse robots will help their human workers have more fun
🌎 An AI model suggests we're going to break through 1.5C of global warming sooner than we thought
...and much more!