Google today released a data set of 3,000 deepfake videos the company created. Google says these videos will hopefully allow researchers to develop methods to combat malicious deepfakes. Deep Fakes are videos, that use machine learning to change the faces and even the voices of someone on video.
The below YouTube video is a fairly harmless deepfake that puts Nick Offerman’s face on every person in the Full House intro. But we’ve also seen malicious deepfakes used to put the faces of celebrities into porn videos and even political content.
Other companies are also concerned about deepfakes. Earlier this month Facebook, Microsoft and 10 universities partnered to offer a $10 million contest spur the development of tools to identify deepfaked videos.
For Google’s part, they recorded actors in various scenes and using publicly available deepfake generation tools created a database with thousands of deepfakes researchers can now use as a reference model.
From Google’s Blog post.“Since the field is moving quickly, we’ll add to this dataset as deepfake technology evolves over time, and we’ll continue to work with partners in this space. We firmly believe in supporting a thriving research community around mitigating potential harms from misuses of synthetic media, and today’s release of our deepfake dataset in the FaceForensics benchmark is an important step in that direction.”
I am reminded of the quote from Batman Begins when Commissioner Gordon says “We start carrying semi-automatics, they buy automatics. We start wearing Kevlar…they buy armor-piercing rounds.” The reference data may well improve the deep fakes more than we have already seen.
I’ll leave you with the deepfake of Donald Trump inserted into the TV show Breaking Bad explaining money laundering.
Header Image: Google
A small town girl living in a robots world. But these robots only exist online and don’t look like Gabriel Luna. I cover things relating to AI and cybersecurity, topics that are increasingly converging.
Good write-up. I absolutely love this website. Stick with it!