Real or Fake
Overview
The RealorFake project investigates the human ability to recognize artificially generated faces using Generative Adversarial Networks (GANs). GAN technology, which can create hyperrealistic images, poses both incredible opportunities and significant ethical challenges. This project explores how people perceive, trust, and remember artificial versus real images, aiming to shed light on human interaction with synthetic media.
Research Goals and Motivation
In recent years, GANs have become instrumental in creating realistic human faces for applications in marketing, media, and beyond. While these advances enhance visual content, they also raise concerns, particularly around the misuse of synthetic media, commonly known as "deepfakes." The Real or Fake project aims to understand how accurately people can differentiate between real and synthetic faces and to examine the emotional and cognitive factors that influence their judgments.
Methodology and Platform
Our custom online platform, RealorFake, allows participants to engage in a structured survey. In this three-part survey, participants:
- Identify if a face is real, fake, or edited.
- Choose the most memorable images.
- Indicate which images they trust most.
Key Findings
The survey revealed fascinating trends:
- Recognition: Participants often struggled to distinguish GAN-generated faces from real ones.
- Memorability: Real photos tended to be more memorable, although some fake images also left a lasting impression.
- Trust: Surprisingly, participants trusted GAN-generated faces more than some real ones, suggesting an intriguing response to synthetic media.
Implications
This study highlights a crucial challenge in today’s digital age: as synthetic media becomes indistinguishable from reality, understanding human perception is essential to navigating both its ethical and practical applications.