OpenEXPO Virtual Experience 2021 had an exceptional guest of honor : Chema Alonso. The popular security expert also gave a lecture covering topics as interesting as cybersecurity and how deepfakes and AI can influence it.
With the advancement of artificial intelligence, cybersecurity faces new challenges. Currently, AI allows for relatively easy identity theft, leading to the rise of deepfakes that flood social media and the internet.
A deepfake allows, for example, the use of an existing video featuring a person, replacing their face with another , and inserting a cloned voice uttering words they would never have spoken. This could lead to terrible hoaxes, especially if used against political leaders or those with significant influence over the population.
They have now become one of the most sophisticated techniques for spreading fake news and disinformation campaigns. They can even significantly influence the increase in cyberattacks , as Chema Alonso pointed out at OpenEXPO Virtual Experience.
And it's a more worrying problem than it seems. Until 2019, there were fewer than 15.000 deepfakes circulating online. By 2020, that number had reached almost 50.000 fake videos , 96% of them pornographic. And the figure continues to grow, creating new cybersecurity challenges.
To detect these deepfakes, Chema Alonso points to two forms of analysis :
- Forensic analysis of the images.
- Removal of the biological data from images.
The renowned expert has delved into this topic in his talk for OpenEXPO Virtual Experience 2021, and together with his team, has been able to develop a plug-in for the Chrome web browser with which any user can select a video and run tests to detect these DeepFakes.
This plugin implements 4 scientific studies to combat these scams:
- FaceForensics ++: checks based on a model trained on its own database.
- Exposing DeepFake Videos by Detecting Face Warping ArtifactsCurrent AI algorithms often generate images of limited resolutions, and this tool detects those limitations with a CNN model.
- Exposing Deep Fakes Using Inconsistent Head Poses- A swap is performed between the original and synthesized face, so that causes errors in the 3D head pose. With a HopeNet model, these inconsistencies can be detected.
- CNN-generated Images Are Surprisingly Easy To Spot…for now: It can be confirmed that the current images generated by CNN share systematic flaws.
The topic discussed at OpenEXPO is extremely interesting and an equally necessary tool , given that these deepfakes are so prevalent these days…
More information – Official Event Website