Deepfakes have become a serious problem for public people such as politicians and celebrities, and it is a matter that does not stop worrying because its improvement advances at a pace that suggests that in a few years we will not be able to distinguish reality from "fiction".
In this regard, Google has also expressed concern, and going into practice, has announced the launch of a database of 3,000 videos with deepfakes that specifically created for subject researchers to improve their detection tools, and thus combat malicious uses of this new genre of videos.
To create the 3,000 videos, Google has paid actors to be able to record their face and later experiment on their faces with those of other people. Both versions are uploaded so that researchers fully understand what the system is producing. The database can be downloaded from the FaceForensics GitHub page. To download it, yes, It is necessary to fill out a form and give you permission.
What deep learning gives you, deep learning takes it away from you
Google begins its statement highlighting everything that deep learning has enabled in recent times, something unthinkable a few years ago, and highlights the role of generative models that have allowed hyperrealist images to be created, musical compositions made by machines or even to write complete news with meaning that suggest that the role of journalists may be outdated .
But, the negative part of this comes when lto technology that allows to achieve new solutions to problems also creates them. This is a fear regarding artificial intelligence in general, and with deepfakes we are already seeing its potential to produce content on a huge scale that can be used for malicious purposes.
To produce the 3,000 videos, Google has collaborated with Jigsaw (an Alphabet incubator whose challenge is to improve the security of any type on the Internet), and The database has been incorporated into the FaceForensics benchmark with the Technical University of Munich and the Federico II University of Naples, which is a project in which the Mountain View company itself collaborates. It is a repository created to measure the parameters of the algorithms and thus learn to better identify when a video is false. Google promises to continue updating the database as technology evolves.
More information | Google AI
Share Google has released 3,000 deepfakes videos so researchers can use and fight them