Article Review 1 “Human accuracy identifying AI generated human faces.” 

Michelle Cabral 

CYSE 201S 

11 February 2024 

https://academic.oup.com/cybersecurity/article/9/1/tyad011/7205694

Introduction 

When looking at cyber security and the social sciences, we have been taught the importance of learning and understanding the human factors that come into play. These factors include how humans use technology, for good or malicious purposes, how they can implement techniques for better use the technology, and most importantly protect themselves from cyber threats. With increasing AI (artificial intelligence) capabilities, it is important to look at what human users can do to protect ourselves from cyber incidents caused by AI. This article specifically looks at the ability of individuals to detect if an image of a human face is authentic or AI generated. The hope of this study, using the scientific method, is to determine the accuracy of properly identifying if an image is authentic or AI generated. 

The Experiment 

While previous studies have been identified to test human accuracy versus AI generated content, this case specifically differed in three major ways. First, the study used an a-priori statistical power analysis to ensure sample sizes would be sufficient to detect differences between conditions that were constructed to avoid cognitive overload. Secondly, the images were to be shown individually as they would in a real-world scenario, not compared to a second image, and limited to twenty so as not to lead to cognitive overload for the participant. Lastly, the study delves into the participants confidence levels on their decision making thus producing an open-ended question for each image, allowing the participants to annotate their reasoning for their decision to include selecting specific areas of the image that guided to their conclusions. It is also important to note that participants were compensated for their time and were informed immediately before the experiment began that the top fifty percent of correct answers would receive a bonus payment. The experiment was given using a web application written in Django and hosted on a secure server in The Netherlands. The hopes were to answer three specific questions. “RQ1-Are participants able to differentiate between deepfake and images of real people above chance level? RQ2-Do simple interventions improve participants’ deepfake detection accuracy? RQ3-Does a participant’s self-reported level of confidence in their answer align with their accuracy at detecting deepfakes?” (Bray et al., 2023) 

The Results 

On average, the participants were correct about sixty percent of the time with a standard deviation of about fourteen percent. While it can be noted that participants in the group with advice and reminders throughout the experiment were marginally better than those with less familiarization on how to spot the AI generated images. Overall, none of the interventions led to a significant improvement. The data collected also breaks down each group’s level of confidence when answering about each image. This data shows that participants who were in the group with constant reminders became less accurate at identifying the authentic images.  

Conclusion 

This study shows the complexity of AI capabilities in generating false images and how even with constant education humans can be fooled into believing such material is authentic leading to the importance of social studies as it pertains to technology. As technology develops, specifically AI, further human factors need to be studied to learn how to combat these advancements. AI generated images specifically can be used for malicious intent such as political propaganda, creating false accounts for programs that require facial recognition, and a company/governments warfare technique.  

References 

Bray, S. D., Johnson, S. D., & Kleinberg, B. (2023). Testing human ability to detect ‘deepfake’ images of human faces. Journal of Cybersecurity, 9(1). https://doi.org/10.1093/cybsec/tyad011 

Article Review 2 “Cyber Bullying before/during the COVID-19 Pandemic”

https://vc.bridgew.edu/ijcic/vol7/iss1/3/

Principals of social sciences include social research. This topic uses survey methods to collect personal data about cyberbullying experiences prior to and during the COVID-19 lockdown. It hypothesized the following,

“H1: Cyberbullying victimization has increased during the COVID-19 pandemic. H2: Social media usage is positively associated with cyberbullying victimization (both before and during the pandemic). H3: Perceived social isolation is positively associated with cyberbullying victimization (both before and during the pandemic)”.

Data was gathered from a survey from 331 participants of currently enrolled college and university students residing in the United States who were at least 18 years old. Analytic strategies were used to explore the difference in victimization experiences before and during the pandemic. The data was also broken down by marginalized groups. Covering experiences from majority 18–24-year-olds (69.4%) as well as gender with 60.4% identifying as female. The topic and findings show that while social media usage did increase, cyberbullying was not influenced by the increase of usage. However, there was an increase of perceived social isolation and social media became more of an entertainment outlet versus a social connection outlet. Other marginalized groups were referenced in from other studies of grades 4-12 slightly younger than this study. This study also found no increase in cyberbullying frequency. This was perhaps due to increased supervision, smaller class sizes, or family members’ presence. Concerns were raised if perceived social isolation was/is a predictor of cyberbullying victimization.

Module 9 discusses social media and shows the increasing usage over 2017 to the predicted 2027. This increase only further emphasizes the need for more research on how social media sites affect the average user. Also, more research is needed to identify more patterns of cyberbullying victims and their aggressors.

Overall, this study, while the majority of those surveyed identified as female, shows the pattern or women falling victim more often than males. This calls for more research into if this proves that women are more active on social media, if women are more susceptible of letting online actions/words of other affect them deeper, or if women are victimized more due to their mental health prior to accessing social media sites.

Refrences
Neuhaeusler, N. S. (2024). Cyberbullying during COVID-19 pandemic: Relation to perceived social isolation
among college and university students . International Journal of Cybersecurity Intelligence & Cybercrime,
7(1), – . DOI: https://doi.org/10.52306/2578-3289.1140
Available at: https://vc.bridgew.edu/ijcic/vol7/iss1/3
Copyright © 2024 Nadya Stefani Neuhaeusler