Fooled twice: People cannot detect deepfakes but think they can

Area of Science:

• Digital Media Forensics
• Cognitive Psychology
• Human-Computer Interaction

Background:

• Deepfakes, hyper-realistic manipulated audio-visual content, pose significant challenges to online information veracity.
• Research on the psychological impact and detection of deepfakes by humans is limited.

Purpose of the Study:

• To investigate the human ability to detect deepfakes.
• To assess the effectiveness of awareness and financial incentives on deepfake detection.
• To explore the cognitive biases influencing deepfake perception.

Main Methods:

• A pre-registered behavioral experiment involving 210 participants.
• Assessing participants' accuracy in distinguishing deepfakes from authentic videos.
• Analyzing cognitive factors, including biases and self-perceived detection abilities.

Main Results:

• Participants demonstrated unreliable deepfake detection capabilities.
• Neither increased awareness nor financial incentives significantly improved detection accuracy.
• A bias towards perceiving deepfakes as authentic was observed.
• Participants consistently overestimated their own deepfake detection proficiency.

Conclusions:

• Individuals exhibit a 'seeing-is-believing' heuristic for deepfake evaluation.
• Overconfidence in low detection abilities exacerbates susceptibility to deepfake influence.
• Current detection strategies are insufficient, highlighting the need for improved media literacy and technological solutions.