Methods of deepfake detection based on machine learning

ML models frequently provide interpretability, enabling practitioners to comprehend the specific features contributing to the model's decision-making, and this transparency can facilitate identification of the cues the model depends on to differentiate authentic and counterfeit content.

“ML models frequently provide interpretability, enabling practitioners to comprehend the specific features contributing to the model's decision-making process. This level of transparency can facilitate the identification of the cues that the model depends on to differentiate between authentic and counterfeit content (Maksutov etal. 2020).”

Synthetic media are much more appealing and lifelike, with broad applications, making protection from them significantly more difficult.

“These media are also much more appealing and lifelike, with broad applications in a variety of contexts, all of which make protection from them significantly more difficult (Maksutov et al., 2020).”

The presence of visible or nonverbal clues for evaluating information has become diminished or nonexistent due to recent technological breakthroughs, heightening marketplace deception to unprecedented levels.

“The presence of visible or nonverbal clues (e.g., facial expressions, eye contact) for evaluating a piece of information has become diminished or even nonexistent as a result of recent technological breakthroughs (Maksutov et al., 2020; Ramadhani & Munir, 2020; Tong et al., 2020), thus heightening the degree of marketplace deception to unprecedented levels (Ho et al., 2016; Taylor, 2021).”

In a machine-versus-machine scenario, the whole detection method becomes obsolete.

“In this machine-versus-machine scenario, the whole detection method then becomes obsolete (Maksutov et al., 2020; Ramadhani & Munir, 2020).”

Addressing the surge of algorithm-generated misinformation opens up a new field of business for developing AI-based solutions and services that detect synthetic content from human-generated content and provide consumers with warnings when confronted with marketplace deception or suspicious content.

“Addressing the surge of algorithm-generated misinformation opens up a new field of business for developing AI-based solutions and services that detect synthetic content from human-generated content and provide consumers with warnings when confronted with marketplace deception or suspicious content (Maksutov et al., 2020; Torres et al., 2018; Zotov et al., 2020).”

Deepfakes are highly realistic synthetic media generated by algorithms and typically distributed as social media content.

“Deepfakes are highly realistic synthetic media generated by algorithms (Chesney & Citron, 2019; Maksutov et al., 2020) and typically distributed as social media content.”

Most studies have perceived deepfakes as a grave danger.

“Moreover, most studies have perceived deepfakes as a grave danger (e.g., Giansiracusa, 2021; Graham et al., 2021; Maksutov et al., 2020).”

The current knowledge on deepfakes is scant and diffuse.

“The current knowledge on deepfakes is scant and diffuse (Maksutov et al., 2020; Zotov et al., 2020).”