Deepfake AI can sabotage national security, according to a cybersecurity expert.
Cybersecurity Expert Warns About Deepfake AI
Dr. Tim Stevens, director of the Cybersecurity Research Group at King's College London, said deepfake AI could create realistic images and videos that can undermine democratic institutions and national security. The tool's availability could be exploited by countries like Russia to achieve foreign policy objectives and threaten national security, Daily Mail reported.
He noted that it doesn't work in the high level of defense and interstate warfare but could manipulate democratic institutions and the press. Autocracies like Russia could reportedly exploit it to reduce the level of trust in those institutions and organizations.
For example, there is a deepfake video of Ukraine President Volodymyr Zelensky appearing to surrender to Russia. The clip circulated on Russian social media last year.
The video featured Zelensky from his lectern calling out his troops to lay down their weapons and acquiesce to Russian President Vladimir Putin's forces.
Fortunately, several internet savvy noticed the discrepancies between the color of Zelensky's neck and face. They also observed a strange accent and a pixelation around his head.
Mounir Ibrahim, who works for Truepic, a company that roots out online deepfakes, told the Daily Beast that the deepfake was poorly done.
Nathaniel Gleicher, head of security policy at Facebook's owner Meta, announced that the company removed the original deepfake clip for violating its policy against misleading manipulated media.
ALSO READ: Deepfake: 50-Year-Old Man Tricks Followers Into Thinking He is a Young Japanese Female Biker
What is a Deepfake?
Deepfake is artificial intelligence (AI) used to create convincing images, audio, and video hoaxes. The resulting bogus content is a portmanteau of deep learning and fake, according to WhatIs.com.
Deepfake can be used to have David Beckham deliver an anti-malaria message. However, the same technology can be used for heinous purposes and deceit.
For example, Barack Obama calling Donald Trump a "total and complete dips--t," and Facebook CEO Mark Zuckerberg bragging about having control of billions of people's stolen data are deepfake AIs, according to DailyMail.
The term "deepfake" was reportedly coined in 2017 when a Reddit user manipulated porn videos on the forum. The user swapped the faces of celebrities like Gal Gadot, Taylor Swift, and Scarlett Johansson onto porn stars.
It uses a subset of AI called deep learning to construct manipulated outputs. The most common method utilizes deep neural networks, encoder algorithms, a base video where to insert the face of someone else is, and a collection of target videos.
The deep learning AI will study the data in various conditions and find the common features between both subjects before mapping the target's face on the person in the base video, resulting in a convincing output.
Generative Adversarial Networks (GANs) are another way to make deepfakes. It employs two machine learning (ML) algorithms with dual roles. The first algorithm creates forgeries, and the second detects them. It will be completed if the second ML doesn't find any inconsistencies.
GANs' accuracy depends on the data volume. There are a lot of deepfakes featuring politicians, celebrities, and adult film stars because they have a lot of material that can be used to train the machine learning algorithm.
Check out more news and information on Deepfake in Science Times.
