Artificial intelligence or reality?

Dr. Faculty Member Kristin Surpuhi Benli, a Software Engineer, stated that video and audio content produced with artificial intelligence is widely used today in many fields such as cinema, education, advertising, gaming, and media. "Advertisements, distance learning videos, characters in digital games, virtual news anchors, and content translated into different languages are prepared quickly and at low cost thanks to this technology. On the other hand, the same technologies can also be misused for purposes such as fake news production, fraud, and creating information pollution," she said.
Stating that with artificial intelligence, deepfake, and similar technologies, the concept of "reality" has undergone a radical transformation in the digital age, Dr. Faculty Member Kristin Surpuhi Benli said, "We now need to question whether the content we see, hear, or even read is truly real."
 

Dr. Faculty Member Kristin Surpuhi Benli from Üsküdar University, Faculty of Engineering and Natural Sciences, Software Engineering (English) Department, evaluated the topic of AI-supported audio and video production technologies.

What is "Deepfake" technology?

Dr. Faculty Member Kristin Surpuhi Benli stated that AI-supported audio and video production is generally referred to as "deepfake" technology. "In terms of word formation, Deepfake is derived from the combination of 'deep learning' and 'fake' words. This technology uses deep learning algorithms developed in areas such as facial recognition, voice analysis, and image processing to produce extremely realistic fake videos, audio, or images," she said.

Can be used by malicious individuals

Dr. Faculty Member Kristin Surpuhi Benli noted that a person's face, gestures, or voice can be imitated using deepfake technology. "Thus, a conversation or event that never actually happened can be presented as if it did. This technology can be used to create creative content, but it can also be used by malicious individuals to spread disinformation or commit fraud," she added.  

Fake news can be produced

Dr. Faculty Member Kristin Surpuhi Benli stated that video and audio content produced with artificial intelligence is widely used today in many fields such as cinema, education, advertising, gaming, and media. "Advertisements, distance learning videos, characters in digital games, virtual news anchors, and content translated into different languages are prepared quickly and at low cost thanks to this technology. On the other hand, the same technologies can also be misused for purposes such as fake news production, fraud, and creating information pollution," she stated.

What lies at the heart of deepfake videos being so convincing?

Dr. Faculty Member Kristin Surpuhi Benli explained that the reason deepfake videos are so convincing lies in the increasing realism and level of detail offered by artificial intelligence technologies, especially deep learning models, in image and audio production. "Details such as facial expressions, blinking, lighting, and skin texture can now be imitated much more naturally. Voice cloning technologies successfully copy speech rhythm and intonation. This advancement in image and sound quality makes fake content more convincing every day, making it increasingly difficult for the human eye or ear to detect them," she said.

The concept of "reality" is undergoing a radical transformation in the digital age

Dr. Faculty Member Kristin Surpuhi Benli, stating that with artificial intelligence, deepfake, and similar technologies, the concept of "reality" has undergone a radical transformation in the digital age, said, "We now need to question whether the content we see, hear, or even read is truly real. This situation brings with it more skepticism in accessing information, while also increasing the need for new methods and ethical standards that can confirm accuracy and reliability."

What are the distinguishing features in deepfake videos?

Noting that techniques and methods for detecting whether an image or sound is an AI product are increasingly developing, Dr. Faculty Member Kristin Surpuhi Benli continued:

"These detection processes are based on image processing, audio analysis, and machine learning techniques. In visuals, fake content is tried to be identified by analyzing facial expressions and gestures. Some distinguishing features to look out for in deepfake videos are; unnatural blinking, synchronization problems between mouth movements and audio, artificiality in skin tone transitions, unnatural light reflections, and distortions seen in hair and edge details. In audio analysis, details such as frequency spectrum, emphasis, intonation, and breathing are examined. Additionally, specially trained deep learning models are also effectively used in detecting fake content."

Advanced deepfake technologies; successful from facial expressions to eye movements

Dr. Faculty Member Kristin Surpuhi Benli also explained that images and sounds produced with artificial intelligence have often become so realistic that the human eye and ear cannot distinguish them. "Although some artificialities can still be sensed with careful examination, such as unnatural facial expressions, inconsistencies in eye movements, or the mechanical tone of the voice, these differences are becoming less noticeable. Especially advanced deepfake technologies can imitate many details, from facial expressions to eye movements, voice tone to breathing patterns, with high success. Therefore, short-duration or low-resolution fake content can often be perceived as real," she said.

The reliability of the account where the video is shared should be evaluated

Dr. Faculty Member Kristin Surpuhi Benli, stating that individuals should first investigate the source of the content to question the accuracy of a video shared on social media, concluded her remarks as follows:

"The reliability of the account where the video is shared should be evaluated, and it should be checked whether the content is published from official or recognized platforms. Comparing videos or news about the same event from different sources is an effective method for confirming the accuracy of the content. Furthermore, inconsistencies between audio and video in the footage should be carefully analyzed. In addition to all these, AI-supported detection tools also offer significant support in identifying fake content."