AI Voice and Deep Fake

Day 137 / 365

There was a time where a photograph of something happening was a certain proof that it happened. Fake or doctored photographs could easily be distinguished. Then came photoshop, and over a few years it changed everything. Professional editors became really great at this stuff, and we could no longer rely on photographic evidence.

But we still had video and audio. If we have a video of two people talking, we would be able to say with great confidence whether it’s real or not. That’s how sting operations work right. But soon, even this is going to change.

Deepfake

Deepfake is a technique that uses deep learning to create fake videos by superimposing someone else’s face on an existing video. And this is far more advanced than the face swap filter on your snapchat. If done well, it can be hard to identify a deepfake with the naked eye. You can see this video that comedian Jordan Peele made of Obama

It could be that easy to put words into someone else’s mouth!

Well you would think that it would be tough to fake the voice, at least. But with AI nothing seems to be out of reach. We have programs now that can be fed with the audio of someone saying something, and then they can generate an audio of them saying something entirely new!

Joe Rogan is a comedian who has also been doing daily podcasts for the past decade. he has over a 1000 podcasts. Using all that audio, someone created a AI generated fake audio of Joe saying random stuff, stuff he never said before.

We are surely living in interesting times. Real life is beginning to feel like a black mirror episode.

This post is part of my 365 Day Project for 2019. Read about it here

Yesterday’s blog — What’s the Box Office?