An AI was trained to create fake videos of former U.S. President Barack Obama and match them to fake audio with authentic effect.
The phenomenon of “fake news” is rampant on social networks–which are the primary sources of news and information for many people.
The implication of widespread fake articles and misleading news is that we no longer know who gives us the info, where it comes from, and what the agenda may be behind disseminating the fake story.AI can create fake videos from scratch.Click To Tweet
As if fake stories on social media didn’t already have as much impact as traditional media, now we have to add “fake videos” to the equation.
And we don’t mean traditional cut-and-paste editing techniques. Instead, we mean creating video from scratch, using AI.
Could Fake News make you a President?
Since the results of the U.S. Presidential election last November, the spread of disinformation has been at the heart of a huge debate.
In a “true” article, researchers from Stanford University and New York University have shown that fake news was “both widely shared and heavily tilted in favor of Donald Trump” but they couldn’t assess whether this was pivotal in the 2016 U.S. Presidential elections.
One of the most surprising fake news stories in the run-up to the election was one wherein the Pope himself endorsed Donald Trump. This ran contradictory to the Pope’s stance at the time, so many people quickly saw through it–even though a few were fooled.
Yet, what if this fake news story had an audio clip of the Pope endorsing the 2016 U.S. Republican Presidential candidate?
Hacking the voice of the Pope, or any other celebrity, is easy enough nowadays thanks to new tech.
Lyrebird, the Canadian startup, has proved its machine learning system capable of synthesizing any voice from a one-minute audio sample.
Now, let’s turn it up a notch. What if we had a video of the Pope endorsing now-President Donald Trump?
It could be that, if a video like that was published at the last-minute when it’d be too late to debunk, many potential voters would be confused.
Think of all the possibilities and consider that this has already been done with former U.S. President Barack Obama.
Deep Learning System Creates Fake Videos From Audio
Researchers at the University of Washington have succeeded in creating fake speech and synched video of Barack Obama.
To synthesize Obama, scientists trained a neural network with many hours of the former president’s speech footage to learn how to match audio features to his mouth shapes.
The AI then synthesize high-quality mouth textures that match accurate lip sync, and the results are straight-up scary.
Researchers admit that their system still presents some weaknesses, like the perfect alignment between the mouth and the face, or the concordance of facial emotions to the tone of the speech.
Although this innovation raises many questions about its possible uses, scientists think that it could have several “legit” applications.
I’m just thinking of one application here, but the realistic conversion of audio to video could be used to improve video conferencing. Skype and Messenger, for example, could use this method to produce quality videos despite irregular connection and poor resolution.