YouTube creators must now 'fess up to using AI in vids YouTube has warned content creators that, beginning today, certain uses of AI in their videos must be clearly labeled. The team at the Google-owned video site said a new tool has been added to the Creator Studio, effective immediately, where uploaders can tag videos containing altered or synthetic media that could be mistaken …
I rarely go to Youtube; but I was just downloading some Light classical, and on one of them one of those obnoxious adverts --- this was for a medical thing so doubly obnoxious --- a male testifier was reading out a spiel, and the rendition was imperfect; as if each sentence was a thing of beauty in itself, and with unwarranted gaps.
So I'm only guessing, but am assuming they saved money by not paying an actual actor, and rather fed the dopey script into an artificial voice generator. No doubt this is the future we were promised.
I don't see how Google is going to effectively police this. Human review doesn't scale up well, and YouTube has millions? billions? of videos. So they'll have to be using software to check the videos for violations.
Our local police force has an acronym: AVANHI, which stands for, "Asshole-Versus-Asshole, No Humans Involved." Given the YouTube/AI* situation, perhaps that acronym might need to be re-defined as, "Artificial intelligence-Versus-Artificial intelligence, No Humans Involved".
*Yes, I'm more-than-thoroughly-aware that the current crop of much-touted software has nothing to do with true artificial intelligence.
(Icon for, "Seeking true intelligence.")
I don't see how Google is going to effectively police this. Human review doesn't scale up well, and YouTube has millions? billions? of videos. So they'll have to be using software to check the videos for violations.
They'll police it as well as they do the "paid promotion" tag, which is to say not at all.
I started playing around with this GAN stuff last weekend, via a free website. It was interesting, yet frustrating -- it's like sitting in the back seat of a car, reaching forward, putting your hands over the driver's eyes, and shouting commands at the driver: "Forward. Left. More left. No, RIGHT!" Keywords were sometimes ignored, or misapplied. Ask for "green eyes", and rarely, you'd get green eyes, but usually, you got a person with a green shirt, or a green skirt, or green pants. The same set of keywords, in the same order (I cut-and-pasted from a text editor into the web page), re-used, got you different results than the first time you used them. Keywords didn't have their normal English meanings. "Dark skin", and "black skin" were interpreted as, "African", and you'd get someone with African features. Similarly, "Asian" was intrepreted as, "Korean". "Breasts" was interpreted as "ginormous hooters", and attempting to modify that via the descriptors, "medium", or "small" might-or-might-not work. There was an "anti-" field, for things you don't want in your generated image. "Nude" or, "nudity" in that field usually, but not always, worked. I got images of women with four breasts. I got an image of a girl with three arms. I got an image of a little girl with a cat leaping out of her chest, Alien "Chestburster"-style, though without the gore. It's not that the GAN is "jumping the guardrails", it's more that you're driving on long stretches of mountainous road where the guardrails were made of wood and had rotted away years ago.
I ended up with a synthetic cutie, which, cat-ears aside, is, to me, obviously fake. She has an unnatural growth on the lower side of her left pinkie finger. Her skin is too smooth. Her left arm is in a position which I cannnot replicate, as it will not rotate sufficiently-far to do so. And her left middle finger has its fingernail on the bottom side of the finger.
But, I am looking at this image on a large flat-panel display, and it fills the screen from top-to-bottom as part of my wallpaper. Most people watching YouTube/TikTok videos are doing so on the relatively-tiny screens on their cellphones. Hence, even if those watchers were detail-oriented (I suspect most are not), they (and I) would have a hard time noticing these clues. Thus, not-great-quality is good-enough if you're producing fake media ("Tom Cruise 'says' Joe Biden wore lederhosen in kindergarten.") for viewing on YouTube or social media.
You can program a bot to watch YouTube vids by the 100,000,000x and provide likes subscriptions notifications, and even a few pithy comments in the discussion forum! I might be a bot, myself, “whose” to say I’m not!
(Why, just yesterday I harnessed artificial intelligence to selectively blur the background out of a couple snapshots of my own farinaceous visage, what’s this? Yes the FBI has come knocking at the portcullis, toodle-oo!)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
