What are GANs - Generative Adversarial Networks, the algorithms behind the most sophisticated deepfakes? How do they work? Are they only meant to make deepfakes or are there applications that are beneficial to society?
Their mechanism, their fields of application for the generation of text, images, sound and video will be explored and deepfake detection clues will be evoked.
In May 2019, Facebook came under fire from critics, the social network released a distorted video of US Democrat Nancy Pelosi in which she appears drunk: the video went viral, was seen by more than 2 million people in two days and allowed Donald Trump to discredit it again - Facebook does not consider itself a guarantor or judge of the veracity of the videos, advocates freedom of expression and announces that it will not remove deepfake. In June 2019, Mark Zuckerberg, founder and CEO of facebook is in turn staged in a distorted video in which his character announces that he is ready to use the billions of data stolen from citizens to control the future and that he owes it all to "Spectre". In January 2020, Facebook published a press release banning deepfakes. Deepfakes now appear to be a propaganda tool whose impact on society is still little controlled.
A websession by the Institut des Actuaires. (In French language)