we’ve been watching a new Silicon Valley legal drama airing on CBS; Cupertino1 https://en.wikipedia.org/wiki/Cupertino_\(TV_series\) centres around two attorneys who join forces to sue their former employer, an AI behemoth, starting in the process their own boutique law firm dedicated to fighting the Goliaths of the tech world.
while the plot has a larger AGI twist (revealed by episode 2, so you don’t need to wait long), a subplot is that one of the main characters, before being dismissed over fears of a sexual harassment suit against one of the co-founders, had her voice and identity cloned by the company without her consent for their new AI model.
life mimicking art, the same plotline is now playing out IRL.
when ElevenLabs2 https://en.wikipedia.org/wiki/ElevenLabs launched its AI voice tools in January 2023, they were all the rage in the business world. like all the other magic bullet AI products out there, these tools would relieve us of the daily slog and “free up” our time to do… well, I’m not sure that part was ever made clear. regardless, we’d pay them in exchange for being rich and happy, because that’s the AI promise 🙄
fast-forward three years, and podcasters like Chris Williamson3 Chris Williamson hosts the Modern Wisdom podcast, which has been downloaded north of a billion times are facing the inadequacies of clear legal or protective frameworks against AI voice cloning. Williamson alleges that a popular synthetic British voice on the platform sounds like him, right down to the idiosyncrasies of his Stockton-on-Tees accent.
Williamson, in reaching out to ElevenLabs, faced what reads to me like the same response Cupertino character Olivia Siffre comes up against in court; essentially: we’re not the problem, your self-inflated sense of your own importance is. according to Williamson’s account, ElevenLabs said the voice came from another performer and that its “similarity test” did not support a match with his.
problem solved 🤷♀️ his account of friends sending him videos of what they presume to be his voice endorsing various products he has no affiliation with notwithstanding.
a voice itself isn’t protected by US copyright law in the way a particular recording can be. other protections can apply, including state-level publicity rights and federal law against false endorsement. the legal test depends on the claim: whether people are likely to believe someone endorsed a product can matter, but there’s no single “reasonable person” test governing every voice-cloning dispute.
as the seedy part of the crypto world demonstrated in all its money-grubbing glory years without needing AI deepfakes, using a celebrity’s name or likeness to sell (scam) people into buying your product or investing in your exit scheme is big business. if digital impersonation erodes trust merely because your identity is attractive enough to be leveraged as a resource in a stranger’s endeavours, then AI-as-enabler supercharges that to a most dangerous level: that of idea semination… spreading ideas under someone else’s name.
if seeing is believing, then hearing a media personality’s voice attached to something that reinforces your existing beliefs—or draws you towards more radical views by virtue of having a voice you respect attached to a line of thinking you’ve yet to go down—is one of the many factors to navigate in the growing polycrisis threatening our sense of reality. (this over and above the fact that AI’s sycophantic tendency to tell us exactly what we’d like to hear can make escaping the echo chamber that much harder.)
even where legal protections exist, enforcing them can still depend on a particular victim having the resources to take a gazillion-dollar company to court. legaltech is in a time crunch to catch up to the rapidly developing state of AI that—like all pioneer tech—massively outpaces the rate at which bureaucracy moves.
here’s Williamson on his legal predicament:
19
footnotes
-
Chris Williamson hosts the Modern Wisdom podcast, which has been downloaded north of a billion times ↩