Human Cognitive Processing Strategies in the Detection of AI-Generated Synthetic Media

Abstract

Deepfake technology generated by artificial intelligence (AI) is becoming increasingly realistic. This technology not only challenges people’s ability to judge what they see but may also influence individual thought patterns. Current deception research generally holds that applying cognitive load to deceivers can prompt them to reveal more deception cues, without considering the cognitive impact of deception detection methods on observers. Addressing this gap, this study moves beyond the limitations of focusing solely on media content or detection mechanisms by examining how individuals perceive and interpret signs of manipulation when evaluating deepfake material, particularly their attention and cognitive processing strategies in multimedia deepfake detection tasks. Participants were asked to assess whether textual, image, and video materials are authentic or fake. The results showed that participants used different strategies in allocating perceptual and cognitive resources across the three media. Text and image materials required longer reaction times and led to more extreme judgments, indicating that they involved more effortful processing and cautious deliberation. Video materials, on the other hand, induced fewer extreme judgments, implying that dynamic cues help individuals detect anomalies intuitively. Although video information is commonly perceived as the most challenging media format to authenticate, participants were able to determine its authenticity more quickly than in other tasks. These findings highlight the importance of considering observer cognitive load in deception detection and offer theoretical implications for integrating cognitive load theory with dual process models of judgment in HCI contexts.

Keywords: Deepfake Detection, Cognitive Load Theory, Dual Process Theory, Information Evaluation

DOI: 10.54941/ahfe1007372