I don’t watch TV much, and haven’t for quite awhile. It’s a bad use of my time, and never more than recently, when a friend sent me a link to a YouTube group called Isabella Advice. The link was to a video on prostate health, which is a topic of considerable (if fairly recent) interest of mine.
Well. What we have is a video of Isabella, a pleasant-looking forty-ish African American woman who says she’s an MD. It’s a 24-minute video, and I watched it to the end. By the time I reached the end, I knew the truth: The video was generated by an AI, and although the advice was fairly sound (and mostly aligned with the rest of my research) the presentation was a little, well, freaky.
What I noticed first were her hand gestures. There were five or six variations, but each variation was used identically many times during the video. She raises her hands to the same height, sometimes alone, sometimes clasped, sometimes just rising at once without touching one another. Her hands remained in her lap unless she was raising them, and the position of her body didn’t change except for a very slight up-an-down bounce. I’ve seen enough videos to know that presenters rarely sit this odd sort of stock still—not for 24 minutes, at least.
There are captions of what she’s saying, and her voice is good, if a little monotonous. But then…the caption contained a weight, something like 20 lbs. She didn’t say “twenty pounds.” She said, “twenty ell-bees.” Later the caption read ”with your daily 30-inute” with the initial “m” missing, and that’s precisely what she said, “thirty in-oot.” (It’s at 16:35 in the video.) She was literally pronouncing typos.
That nailed it. I’m guessing that someone who knew something about prostate health wrote a script, and the AI voice read it mercilessly verbatim.
When I’ve watched YouTube, it’s usually because somebody is doing something visual and explaining it: crafts, ham radio, building electronic projects, robots, telescopes, and so on. The visual focus isn’t 100% on the presenter. I could have read what the AI was saying in probably five or six minutes. For the most part, I research by reading, unless there are necessary graphics. Watching an AI sitting in a chair reading a script for half an hour is truly a waste of my time. I won’t do it again, but I’m going to watch for other examples of AI in use, not just on YouTube but anywhere else I run across it. If I spot any, you’ll read about them here.
Like you, I watch very little TV. But after my heart surgery, I was pretty much out of it for a couple months. During that time, I watched a lot of youtube documentaries and how-to videos.
An amazing number of them are now AI generated. The images are often very good, but flawed if you watch closely. And the speech is very repetitive and includes frequent mis-pronounciations.
The actual information is also suspect. They tend to exaggerate, and be expressing someone’s opinion rather than factual information. It’s believable if you don’t know the subject; but if you do, you can spot the errors.
My wife caught one of these on YouTube the other day and we watched one of the suspect channel’s videos together. Quite fun. The cheerful and excited woman in the thumbnail becomes a cold, grim-faced apparition who could plausibly be human (perhaps being held hostage?), even with the monotone and apparently wearing the same clothes in every video (even though, very weirdly, she’s only ever shown for literally a few seconds at a time). But the illustrative shots all throughout the video are depressingly “good AI”: almost-Walmart aisles full of mysterious products with bizarre blurred-together substitutes for brand names, packaging that moves by itself while the camera pans over it, bags of play-kitchen chicken quarters, pots of slightly-too-perfect chili being prepared by eerie almost-hands presumably belonging to the blandly uniform-faced animate mannequins (sorry, women) who are shown opening the Facebook-slop-post-filter refrigerator. I even think some of the ideas expressed in the videos may be decent, we just don’t want to bother digging enough to decide what we can trust if the channel author couldn’t be bothered to have a face or a name or use words himself / herself. What a time we live in.
Oh cool: I just went to check the Isabella AIdvice video you linked to, and it’s apparently now “private.” The entire channel has disappeared from YouTube search results.
Yup. Isabella Advice is just, well, gone. I’m assuming it’s a coincidence; I don’t have enough followers to trigger a critical mob. Nor is it worth my time to try and see why the channel is gone. I’m trying to get a novel written. But it was a good example of AI being used to replace human presenters in videos.
Apparently YouTube is now force labeling videos as AI generated if its AI notes tell-tale signs. I’ve only seen it on a few shorts, but not on others that were obvious AI.
The YouTube rule that creators need to self police their content (stating their content is AI generated) does not seem to work.
Also, I’ve read that the required or forced AI label was formerly hidden inside the expanded video description for long-form videos, but should now be more prominently displayed.
We’ll see how this plays out.