AILSA CHANG, HOST:
AI is everywhere, right? Like, you're probably seeing it all over your social media feeds. Remember those cute bunnies jumping on a trampoline? Fake. And now because there are so many of these fake videos, there are now AI video critics of sorts.
(SOUNDBITE OF ARCHIVED RECORDING)
JEREMY CARRASCO: In this video, different AI model. I think this is Kling, which is a Chinese model, because there's Chinese text on the ceiling. And when he runs his hand through his hair, his pinky gets completely jacked up there. They're selling research chemicals, skin care products. There's no way this is real.
CHANG: That is Jeremy Carrasco. He's a video producer who's spent years screening real videos for errors. Now he's turned his careful eye to AI. And on TikTok and Instagram, he teaches people how to spot the fakes. Carrasco says spotting fakes has gotten a lot harder as AI models have improved. Missing limbs and fingers just aren't a dead giveaway anymore. But the models still aren't perfect.
CARRASCO: For example, there are AI scams of people who are sitting in hospitals asking for money. But if you send those images to a doctor, they're going to realize that the equipment is completely wrong or halfway done. If you're looking at a video of a rock climber, the ropes are going to be unsafe or not tied down to anything.
CHANG: (Laughter).
CARRASCO: And as a casual observer, you might not know that. But the good news is you have other superpowers or other domain expertise that can be useful in other situations.
CHANG: So even if the scene looks realistic, your own logic might tip you off to a fake. He says even something like the camera angle can help you think critically about what you're seeing.
CARRASCO: Absolutely. The main difference between these videos and our real videos is that there isn't a real person behind the camera. So a lot of the times, the camera movement doesn't make sense. And it just requires stepping back and asking, who would've filmed that and why? There were a lot of viral videos of dogs saving babies from falling shelves where you had a perfectly positioned camera.
CHANG: Yes.
CARRASCO: You know, that wouldn't happen in real life. And if it looks like someone's holding a cellphone and a baby is sticking something into an outlet, that person would've put the film down to help, you know?
CHANG: (Laughter) Right, right. Or like a shot of a cat jumping on a bed of somebody in the middle of the night and grabbing a snake off of it. Like, why would you have a security camera panned on yourself while you're sleeping at night?
CARRASCO: Exactly. I mean, not all of them are that obvious. But I'll tell you, if you can catch those obvious ones - and pet videos are great because pet videos show up on all of our feeds. If you can catch those obvious ones and learn from them from other details, then you're well on your way to becoming better at spotting it overall.
CHANG: Well, I saw that TikTok has a policy that requires creators to label all AI-generated content that contains realistic images, audio and video. That's the policy. How reliable is that policy? And how widespread are policies like that among social media companies these days?
CARRASCO: Very unreliable. And they have some detection. But the detection seems to be more based on descriptions or hashtags rather than actually analyzing the video. From what I can tell, no social media platform is taking detection or labeling very seriously.
CHANG: OK. Well, that leads me to my next question. Some of your videos talk about technical watermarks that AI generators embed in their content. But are those reliable indicators? I mean, how are those watermarks even visible to the casual consumer who encounters these videos?
CARRASCO: Yeah, there are visible watermarks and invisible watermarks. So AI companies like Google are trying to input invisible watermarks that are embedded into the pixels. But if you give a photo that was AI generated but wasn't made by Google to Google, they won't be able to tell you if it's AI.
CHANG: Right.
CARRASCO: So there isn't a good, reliable, centralized solution right now. But there's definitely some promising ones that hopefully we start to see implemented soon.
CHANG: I mean, God, talking to you right now and thinking about all these methods to use to spot fakery, you know what I feel like is getting lost here is just the spontaneous delight you would otherwise feel when you come across an amusing video.
CARRASCO: Yeah.
CHANG: And I'm wondering, how do you think that hurts all those video creators out there who are out there making real stuff?
CARRASCO: I agree with you. That's been the saddest thing because really the way to get good at spotting AI videos is to take that sense of wonder that you might've felt...
CHANG: Yeah.
CARRASCO: ...And you have to ask, wait, is that because it's AI now?
CHANG: Yeah.
CARRASCO: And the thing that I would say is that there are a lot of real creators who are still giving you that sense of delight. And you can just follow them and know that you can trust them to not trick you because a lot of real creators really have no interest in making AI videos. And the AI accounts really just are there to make AI. So I don't think that that sense of wonder needs to go away. You just need to know where to look for it and not rely on the social media platforms' algorithms to just present it to you.
CHANG: That is AI literacy educator Jeremy Carrasco. Thank you so much for your insights.
CARRASCO: Thank you, Ailsa. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
