Farid's research found that people are only slightly better than chance at spotting AI-generated fakes, even though most believe they are good at it. Here he explains why real-world conditions make people worse still.
…stealing somebody's voice and their likeness. And it's really, these are nasty and people aren't ready for it. Like we've gotten, maybe we've gotten ready for the spams and the malwares and the go out and buy me, you know, music gift card. But this one, I don't think people are fully prepared for.
And here's the wildest part. If you ask someone if they think they're good at spotting fakes, most people will tell you they are. Hani has actually tested that. And it turns out, humans are actually pretty bad at it. His research shows our ability to spot a fake is just slightly better than chance.
And here's the thing. You are in a lab setting. We've set you up for success. It is not politically charged. It is not emotionally charged. Now, imagine you are in social media, doom scrolling, politically charged, emotionally charged, things that you ideologically feel strongly about. You're way worse at it in that situation, but they don't know it. And that's really interesting. I don't mind you being bad at something as long as you know you're bad at it, but they're bad at it and they're confident in the wrong assessment.
And apparently, we have evolution to thank for some of this.
Our visual system didn't evolve to think, is this thing real or not? We are out in the world. And if we saw a tiger, run. There was no question of, well, is that really a tiger? You just ran and that's how you survived. But in this modern world, in this modern AI world, we have to really figure out…
So what does a guy who knows just how easy it is to impersonate someone do to protect himself? Well, for starters, he and his wife have a secret code word, something they can ask each other if they ever get a suspicious phone call. But Hani has a less secret spy recommendation for the rest of us.…
