Detecting Deepfakes Using Signatures: a User Experiment
Keywords
Loading...
Authors
Issue Date
2022-06-27
Language
nl
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
A deepfake is either a video where the actor is swapped with someone else who was not
originally there or a lip-sync. Deepfakes are currently used for more damaging than beneficial
purposes and are mainly associated with disinformation. Problems with deepfakes
arise for different stakeholders, like individuals, companies and in fields such as politics
and forensics. There are many solutions being developed to reduce the problems that deepfakes
bring. Most of these are on the basis of Artificial Intelligence (AI). However, it is
sensible to look into solutions outside of AI. The solution proposed here is letting content
creators sign their videos. This is to provide viewers with reliable background information
about the creator/poster of the video. With this information, the viewer can make a
better informed decision about the credibility of the video. The goal of this research is to
examine and analyse if video signatures effectively helps humans to judge the authenticity
of videos, both real and fake. The research question therefore reads: “What effect does
providing relevant/non-relevant background information about the source of a video, in the
form of video signatures, have on people’s credibility assessments of the signed real and
fake videos?” This question was researched by presenting participants with 20 videos, which
were either fake or real. To these videos, no signature, a relevant signature (Obama signs
an Obama video) or a non-relevant signature (a random person signs an Obama video)
was added. The participants needed to indicate how certain they are the video is either a
deepfake or real. Based on the analysis of 47 participants’ results, it can be concluded that
video signatures do indeed have an effect on peoples’ credibility assessment of the videos.
The non-relevant signatures have a positive effect on peoples’ credibility assessment of
fake videos, but a negative effect on peoples’ credibility assessment of real videos. The
relevant signatures have a positive effect on peoples’ credibility assessment of real videos.
This makes the addition of signatures to videos online a solution that is worth researching
further and possibly even implementing in the future.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen