Microsoft President Brad Smith’s LinkedIn post featuring an unusually dapper dog has stirred more than just curiosity. It heralds the launch of Microsoft’s latest initiative—a quiz designed to test the public’s ability to distinguish between genuine and AI-generated images.
The “Real or Not” quiz, a part of Microsoft’s broader AI literacy campaign, challenges users to identify which of 15 images in each quiz are real and which are deepfakes. Despite multiple attempts, I managed to score only 67%, highlighting the increasing sophistication of these deceptive technologies.
Deepfakes—AI-generated content that can include images, videos, or audio recordings—are notoriously difficult to detect. As these technologies advance, their capacity to spread misinformation grows, raising concerns about their potential impact on election integrity and public trust.
To aid in combating the spread of deepfakes, Microsoft has launched an educational resource page. The site offers general guidelines for improving media literacy and handling suspicious content, including:
- Verify the authenticity of sources.
- Confirm the accuracy of political or voting information before sharing.
- Report suspected deepfakes to social media platforms.
- Continuously enhance media literacy skills in light of evolving technologies.
- Verify voting plans with official authorities.
While these tips are useful, they may not directly help in pinpointing nuanced deepfake anomalies, such as those potentially seen in distorted images of public figures.
Recent efforts in similar vein include quizzes from Seattle-based AI nonprofit TrueMedia and The New York Times, which also test users’ ability to identify AI-generated human faces.
Additionally, Microsoft and Smith released a report titled “Protecting the Public from Abusive AI-Generated Content,” advocating for swift actions by policymakers, civil society, and tech industry leaders to address the rise of abusive AI content.
The challenge of identifying deepfakes underscores the importance of staying vigilant and informed as technology continues to evolve.
Related topics:
What Is Residual Learning for Image Recognition