Microsoft is working with Facebook to organize and run a contest that will help them better understand and detect deepfake, Facebook stated this Thursday.
The social media powerhouse has injected close to $10 000000 into this project with the aim of spurring research in the area of deepfake detection. In addition, Facebook has commissioned researchers to come up with realistic deepfake that will be used to calibrate detection systems.
The social media company also stated that videos, to be released this December, would work with paid actors. As the presidential election approaches in a year’s time, social media platforms are being pressured to deal with deepfake that uses artificial intelligence to come up with hyper-realistic clips of people saying things or perform acts they did not really take part in.
Although no one has come up with a deepfake with a political implication in America, such videos have the potential to create chaos as was demonstrated most recently with a cheapfake video of Nancy Pelosi, House speaker. The video was edited to fewer frames per minutes to make it seem like her speech was slow.
This past August, the Democratic National Committee showed how dangerous deepfakes can be by creating a video of Chairman Tom Perez, to make it seem like he had attended Def con hacker convention via skype.
Researchers are currently working on systems that will help authenticate every image or video at the capture point through digital watermarks. However, due to the rapid evolution of deepfake technology, there is an arms race in the making between video detectors and deepfake creators.
Deepfake technology is also becoming increasingly accessible to less knowledgeable creators. In fact, as recently as last week, some users of Zao, a Chinese App, were able to morph their likenesses into videos, which made viewers think they ad stared in the film. The app listed at the top in China’s app store. Critics have raised privacy concerns over the use of this app.
Deepfake creators have also capitalized on this app and the market as a whole to access easy to create deepfakes. Machine learning experts in Japan and Poland have made it incredibly easy for users to come across customized deepfakes.
They are also uploading systematic tutorials on YouTube at a fee of $30 as the cost of 50 words for an AI-based voice of a famous person like Trump. They are also creating websites that generate deepfakes.
This is not the first time that social media giant face, which does not have a policy on deepfakes, is funding such a challenge. This new contest, which is expected to build ties between the company and the researchers, will also indulge academics from institutions like MIT, Cornell Tech, UC Berkeley, Oxford University, University of Maryland, University of Albany-Suny and College Park.