Fake news is about to be a bigger problem than it already is thanks to a research team from the University of Washington which has used Artificial Intelligence (AI) to convert audio files into realistic mouth movements and further adding these realistic looking mouth movements onto an existing video.
To demonstrate the technology and how it is trained and works, the researchers used a mash-up of audio and videos of former USA president, Barack Obama.
Demonstration of the AI technology used to produce the fake videos.
So far, the spread of fake news has somewhat been limited to written content which attempts to masquerade itself as legitimate news content by using variots techniques such as writing on trending topics and using similar names/web addresses to established news outlets. Although fake news somehow manages to spread, to the discerning eye and analytical mind a few checks can reveal a fake news article from a real one. The fake news problem has been credited with getting Donald Trump elected, it is not however limited to the USA and it is also quite prevalent in Afrika too.
In the lead up to the 2017 Kenya presidential elections set to take place on 8 August 2017, there has been a scourge of fake news trying to sway the elections to an extent that Facebook released an educatonal tool to help Kenyans spot fake news. Even then, the fake news in Kenya have mostly been limited to written content. However, we did experience a glimpse of the future of fake news, and the closest we could get to what the University of Washington is working on, when fake videos made to look like BBC and CNN news reports made the rounds on WhatsApp and social media. The videos were so believable, in that they looked like BBC and CNN reports if you didn't look closely, that the broadcasting companies issued statements denouncing them as fake.
"We show that by training on a large amount of video of the same person, and designing algorithms with the goal of photorealism in mind, we can create believable video from audio with convincing lip sync. This work opens up a number of interesting future directions," said the research team from the University of Washington in the research document titled "Synthesizing Obama: Learning Lip Sync from Audio".
Download "Synthesizing Obama: Learning Lip Sync from Audio"document.
Although the researchers from the University of Washington do not acknowledge that this AI technology can be used for fake news, they do explain that it is more difficult to "train" it using a non-celebrity. This is because of the lack of hours of training data such as audio and video, which in turn likely means, should the technology be publicly available, it would be likely easier to use it to produce fake videos of popular people such as politicians.
Not only does this technology present us with a potential fake news problem in the future, it likely also will brings into question some video evidence that is used in legal proceedings.