Seeing Through Deception: Assessing Human Ability to Detect Deepfakes: A Comparative Study of IT and Non-IT Professionals in Swedish Organizations

Sammanfattning

This thesis explores the human ability to assess manipulated media and examines whether IT expertise influences detection performance. While machine-aided detection methods like deep learning have shown promise, human judgment remains understudied in identifying fabricated images, videos, and audio. The study investigates whether Swedish office IT professionals and non-IT professionals differ in their ability to to detect deepfakes and if general awareness or knowledge of generative artificial intelligence affects detection accuracy. A quantitative approach was employed using an online survey, with statistical analyses including T-test and Spearman’s correlation to compare the groups and explore how familiarity affects accuracy. The findings indicate that human ability to detect deepfakes is limited, with respondents correctly identifying 57.6% of the presented content. Although IT professionals reported higher accuracy and greater certainty, the difference was minimal, suggesting that IT expertise alone does not significantly enhance detection. Instead, a higher level of familiarity with relevant concepts was associated with better performance, emphasizing the importance of increasing public awareness and education to counter the challenges posed by manipulated media. Future research could examine deepfake detection in realistic, contextual settings, such as social media, using a collaboration of human and machine-aided approaches.