Deepfakes, explained

Brut America

Is this video of Tom Cruise fake or real?  Deepfakes — or computer-generated fake videos — are on the rise, and they can be dangerous...

Transcript

00:00I'm going to show you some magic, it's the real thing.

00:07I genuinely love the process of manipulating people online for money.

00:11President Trump is a total and complete dipshit.

00:14Moving forward, we need to be more vigilant with what we trust from the internet.

00:30If humans are already tricked by, you know, simple spam emails,

00:33I think deepfakes are going to be really effective in the future,

00:36if there aren't tools in place against them.

00:45There was a project called the Video Rewrite Program,

00:48and that was the first major application of a deepfake.

00:50You take one person talking in a video track and sync it to a different audio track,

00:54and, you know, manipulate their face to make it seem like they're talking from the other audio track.

00:59I never met Forrest Gump. I'm a little teapot, short and stout.

01:08The term deepfake comes from deep learning mixed with fake event.

01:11So the fake event is the, you know, output of what a deepfake comes up with.

01:15And then deep learning is the process to come up with the algorithms

01:20and everything to, you know, combine the image with the video.

01:23Deepfake relies on a neural network,

01:25which basically takes a large sample of data and finds patterns throughout them.

01:30So you can take an image and apply it to a video of somebody moving their face.

01:42What's up, TikTok?

01:48Every now and then I like to treat myself.

01:50Deep Tom Cruise, if that was used for malicious purposes,

01:55you know, you could use a platform like TikTok to go viral and spread,

01:58you know, pretty dangerous news.

02:07As a human, if we pull up a video of someone,

02:09if the voices sound similar enough, I feel like the image of the,

02:13you know, the deepfake will kind of trick the ears into like, you know,

02:16being like, oh, hey, that's that person's voice.

02:18And, you know, there are pretty good voice actors who can almost replicate the real thing.

02:39And they take a train, a ton of data of other humans moving their faces,

02:51and then they can apply it to a still image.

02:53But what they have to do is they first have to use algorithms to take that image

02:57at a low quality resolution and turn it into a high quality resolution.

03:00It's bringing deepfakes to the masses in a way that hasn't really been seen before.

03:04I think by being able to take like, for example, like an old deceased family member,

03:08taking one of their pictures and kind of, you know, turning them back into life,

03:11that could be a really cool experience for a lot of people.

03:23Oh, my God.

03:38Taking a world leader and making a deepfake out of them.

03:46And basically, they could, you know, say whatever they want.

03:49They could say things to cause public unrest, you know, say like, I guess,

03:53give dangerous information or dangerous, I guess, like commands, you know, to people,

03:58you know, family members can be impersonated.

04:00So scams can happen that way as well.

04:01I'd say like 10 years ago, when text was the biggest thing, text and images,

04:05it wasn't nearly as big of an issue.

04:07But now, you know, I already feel like the shorter form video platforms

04:11already have a huge like misinformation issue and fake news issue,

04:15because video is so convincing.

04:17I'm sure like hackers are going to get much more creative about this stuff,

04:21especially going forward.

04:29I think big tech is going to be banding together and focusing on tools

04:33that can help prevent deepfakes or at least catch them right away.

04:36And probably like provide labels to say, hey, this like, you know, isn't good,

04:40especially, you know, on Facebook, Instagram, TikTok, any of these apps.

04:45Social media companies are the ones that need to be focusing on creating these tools.

04:48I think at the forefront of all of this,

04:50because they're the ones who'd be most heavily impacted and their users as well.

Category

Transcript

Recommended