The thing I find most interesting about people saying AI generated content is easily noticeable, is that they're always presented a question of "Do you think this is AI generated content?". Nobody consumes media with a list of questions they ask themselves for each piece of content. I expect people in the comments to explain how obviously fake it is, but barring the ridiculous scene, I'd 100% believe that was legitimate if I saw it with no AI warning.
> I'd 100% believe that was legitimate if I saw it with no AI warning.
And if it weren't such an obviously fake situation. Imagine it were more subtle, like Cruise out on a date in a high-end bar, coming onto a woman and overstepping boundaries, taken with a concealed cell phone.
> Before we asked “was that recorded in analog or digital?” Then we stopped asking.
Very few people did, and the consequences of that are incomparable. But I'm afraid your idea of commodity/ease of use will make much of the media AI mediated (at least). And like analog/digital, nobody will notice the difference.
What do you mean by saying that the consequences of not always asking whether something was recorded in analog or digital are incomparable?
Asking that question seems relatively irrelevant to me, especially now that pretty much everything is at some point converted to digital anyway, if only for ease of distribution.
In that sense filming on film, for example, is only an aesthetic choice and a choice about the tools you want to use on set, anyway. Which is not to say that it doesn’t matter or doesn’t make a difference. But in cinemas they will still project the digitized version of that film. And people will still distribute digitized scans of their “analogue” photos online.
I mean, even “digital” cameras record decidedly analogue voltages from photosites, the difference is mostly about at which point in the chain you put the analogue digital converter.
In the sense that not knowing whether it's analog or digital doesn't alter the information. It's the source plus some noise; our eyes and ears perceive more or less the same, the interpretation is identical. But an AI-mediated video can invent or change the information in ways we we're not equipped to deal with. In that sense, you cannot compare analog-digital vs original-artificial.
For a year or so I've had the idea of setting up one of those "can you spot the AI generated content" tests except pitching good AI generated content against bad AI generated content, but keeping it absolutely secret that the good ones are AI generated. I suspect very close to 0 % of the test-takers would come out and say that "I'm confident all examples I saw were AI generated".
I think people would unconsciously shift their mental bar to the point where the bad stuff makes them believe the good stuff is real.
I see this "fake tom cruise" serie of video a lot on youtube (probably the same guy, because it's the same voice and same height). It's so perfect it made me wonder if the guy's isn't already a tom cruise lookalike and the AI part only adjusts minor details.
You only feel like that because you know it's a deepfake beforehand. It's a real person, an actor with a swapped face, so whatever he's doing with the movement of his shoulders neck or head is completely natural.
Nah, the head is too small for the body and the movement is wrong, maybe a person might take it as phone things or whatever but we have months! until this is actually perfected :)
Is there a technology we might use someday to prove the provenance of digital media? Like every video and photo has a blockchain entry or something. And any time there’s a modification, that gets recorded too. So in the future we might right click any random video and see its file history.
Wow, that looked incredible. It does worry me that you could generate a video about a celebrity / VIP in a compromising situation, and that it goes viral before anyone can disprove it. Alternatively, could there be a video scanner that tells you if the video has AI "fingerprints", and this would be built into social media platforms with a "likely AI generated content" warning on posts.
Or, we could just remove video/audio as evidence. Even with fingerprinting, it can still be faked enough to get past judged and elected folks who are already 200 years older than they should be while in office.
People are easily faked and, imho, it’s safest to assume everything is faked.
It’ll be fine. Keep in mind that the world has run on handwritten signatures for a few hundred years. And anyone could say anything in a letter with no “pixels” to prove it right.
Digital media as strong evidence has been a novelty.
If anything the real thing to watch out for will be the echo of national digital ID. This would certainly solve the mild/moderate issues of deepfakes while (if done poorly) risking a serious upgrade to surveillance state/capitalism.
I wonder if having all this "strong evidence" has caused us modern people to lose the social norms and relationships that people had to depend on back when human trust was more important. Like, good fences make good neighbors but what happens when all of a sudden the fences don't work anymore and we have to be good neighbors without them?
I'm guessing an uncomfortable transition period at least.