Article Review #2: Testing human ability to detect ‘deepfake’ images of human faces
Mason Phillips
School of Cybersecurity, Old Dominion University
CYSE 201S: Cybersecurity and the Social Sciences
Professor Yalpi
April 16th, 2026
BLUF
BLUF: AI deepfake technology is a serious AI threat and can be used to fake images and videos, which in turn can lead to the spread of misinformation and disinformation, potentially causing harm to people who are unable to tell the difference between what is real or fake. This study uses both qualitative and quantitative data to test 280 people on their ability to tell deepfake images from real images to test how real the threat of deepfake AI is in today’s times.
Relation to the Seven Principles of Social Science
This study directly relates to all seven of the social science principles. The first principle it relates to is the principle of relativism, which is the principle that says everything is connected in some way. The way you can make this connection in this study is that the subject’s ability to detect AI deepfakes is largely dependent on their technology usage, familiarity with technology, and prior interactions or experiences with AI. Parsimony is another principle that is also related with the study, and it is shown through the simpleness of the study, its conclusion, and how simple it was for the participants to understand. “The key aspect of the experimental task was a binary question asking participants whether they believed a given image was a deepfake or not… people are either right or wrong.” (Bray, Johnson, & Kleinberg, 2023). Ethical neutrality in this experiment is shown by the fact that the 280 people participating in the experiment were online and were not named during the experiment, with the ability for them to share their nationality being a completely optional field to protect the people who participated. Determinism, meaning that behavior is caused, is shown in the article and in the survey that was done by the people who participated in the survey experiment. They showed this when they looked at the images and chose whether they believed the image to be AI or real. “People may have their doubts about an image, which may in turn affect the decisions they make.” (Bray, Johnson, & Kleinberg, 2023). Two more principles, objectivity and empiricism, are both shown in the article through the use of concrete evidence that was collected from a large group of participants. The last principle, skepticism, is shown through the idea of the study by testing if people could tell the difference between AI and deepfake, keeping skepticism at the center of the experiment by questioning every image.
Experiment Details
The main question of the experiment and what was being tested was if people could accurately tell the difference between AI deepfake images of faces and real images of faces. The independent variable was the four different groups of participants, which include a control group, a group that was shown deepfake images beforehand so they had a loose idea of what to look for, a group that had a set of tips for spotting deepfake images displayed beforehand, and a group that had the tips displayed throughout the entire experiment. The dependent variables, or what was being affected by the independent variables, were the accuracy of deepfake detection, the participant’s reasoning behind their choices, and the confidence levels of their choices.
Research Methodology and Data Analysis
The experiment itself was an online survey and used both quantitative and qualitative data collection. Qualitative data collected appears when the participants gave their reason for why they thought an image was or wasn’t a deepfake image, and quantitative data comes from the accuracy percentage and the score of confidence that the participants gave. “For each image, participants were asked to label the image as ‘AI-generated’ or ‘real’ and to rate their confidence in their answer on a 10-point scale (from ‘No confidence at all’ to ‘Complete confidence’). They were additionally asked to answer a free-text question about the reasoning behind their decision.” (Bray, Johnson, & Kleinberg, 2023).
Other Concept Connections
Two concepts that connect well to the article are the psychologic theories. While in the module 5 PowerPoint the theories mostly deal with cyber offending, some of the psychological theories can help to explain the choices made by the people who participated in the experiment. The cognitive theory helps to explain why somebody makes the choices that they made along with how they then justify those choices as well, which could be used when trying to explain why the participant said a picture was AI or not, and this is shown when they type in the reasoning for their choice. The second theory that can be used is the behavioral theory. This can be used to explain the decisions made by the independent variable groups, since they were exposed to some form of indicators of how to identify AI deepfake images. I would argue that the participants being exposed to this information would align with the behavioral theory since the behavioral theory includes making decisions due to a learned behavior, which could come from both media and the environment.
Concerns of Marginalized Groups
A concern that could be brought up about marginalized groups is the concern for the older generation of people not being able to detect AI deepfake images as well due to the technology being recently exposed to the world. You could also argue that people from older generations are also more vulnerable to being misinformed or disinformed by an AI deepfake photo or video due to them not being as well educated about cyber risks or their inability to use technology efficiently. I believe that this experiment can be used as a contribution to society in the way that it could inform and raise awareness about the possibilities of deepfake scams and the disinformation that can be given by them, along with ways to potentially spot something as AI content in contrast to real media.
Conclusion
In conclusion, the article confirms that deepfake AI technology is a serious concern, with the participants in the study only guessing correctly an average of 60% of the time. With a recent explosion in AI innovation, the AI deepfake technology is only getting better at creating media that is indistinguishable from real media and making it more likely that somebody will fall for a scam, whether it be phishing, disinformation, or even a romance scam by a fake profile on a dating app.
Reference
Bray, S. D., Johnson, S. D., & Kleinberg, B. (2023, June 23). Testing human ability to detect ‘deepfake’ images of Human Faces | Journal of Cybersecurity | Oxford academic. Journal of Cybersecurity. https://academic.oup.com/cybersecurity/article/9/1/tyad011/7205694