In what might be the most bureaucratically sensible response to the increasingly uncanny world of artificial intelligence, Meta has announced that developers using its “Voicebox” generative AI model will now need signed consent forms from real people if those people’s voices have been used in training or output. In other news, the 21st century continues to feel like a strangely dystopian HR seminar.
This new requirement, outlined in Meta’s updated “Responsible Use Guide” for generative audio tools, insists that developers seeking access must not only specify their intended use, but also solemnly swear that no human voices were harmed in the making of their dataset without prior approval. Consent, it seems, is not only sexy again, but also a compliance checkbox in Meta’s AI playbook.
Voice cloning, for the uninitiated or the blessedly skeptical, involves training a model on recordings of a person’s voice so it can later generate eerily accurate audio clips without requiring that person to say the words themselves. This technology has already ruffled quite a few ethical feathers, particularly when it comes to impersonating celebrities or, even more disturbingly, loved ones in scam calls. Meta, sensing the potential for headlines even worse than “Tech Giant Accidentally Starts AI Cold War,” has decided to preempt some of the chaos.
“We are committed to developing and deploying AI responsibly,” Meta said, presumably while side-glancing at every other tech firm still treating ethical concerns like optional toppings on a corporate sundae.
While Meta’s Voicebox model isn’t publicly available, it is being shared with “select partners” under controlled conditions. These partners now have to pledge in writing that every ounce of virtual vocal power they unleash is consensually sourced. Essentially, Meta wants to make deepfakes polite before they become completely unmanageable.
The move has drawn some cautious praise from digital rights advocates, although many still express concern that the industry as a whole remains several steps behind the people who are already using AI to make prank calls sound like Morgan Freeman. Regulation may lag behind innovation, but at least Meta’s trying to staple a permission slip to progress.
If you want to clone someone’s voice now, apparently all you need is cutting-edge AI and some light paperwork.