I “A.I.’d” My Own Campaign Rally to Test Trump’s Claims

🌈 Abstract

The article explores the potential for using AI-generated images in political campaigns, specifically in creating a fake campaign rally photo. The author experiments with various AI tools like Midjourney, DALL-E, and RunwayML to create a convincing-looking rally photo and video, and discusses the implications of such technology being widely available.

🙋 Q&A

[01] Experimenting with AI Tools

1. What AI tools did the author use to create the fake campaign rally photo? The author used several AI tools to create the fake campaign rally photo:

• Midjourney, an advanced AI image generator, to generate the base image
• DALL-E to create the campaign logo and poster
• Adobe Photoshop's Generative Fill and Spot Healing Brush tools, which use AI, to edit the image

2. How did the author evaluate the quality of the AI-generated image? The author noted that while the initial Midjourney output had some obvious flaws, the Version 6.1 system produced a reasonably realistic image that captured the aesthetic of the real rally photo. However, the author also pointed out that the text on the signs still looked fake.

3. What did the author do to further improve the fake rally photo? To improve the fake rally photo, the author used Photoshop's AI-powered editing tools to remove the text on the signs and replace it with a campaign poster generated by DALL-E. The author also added a custom 747 plane livery to the image, again generated by DALL-E.

[02] Implications and Limitations of AI-Generated Images

1. How does the author view the threat of AI-generated fake images in political campaigns? The author believes that creating a reasonably convincing AI-generated rally photo is now relatively easy, and that anyone with some Photoshop skills and access to these AI tools could do it. This raises concerns about the potential for such images to be used to sow doubt and misinformation in political campaigns.

2. What does the author suggest as a way to verify the authenticity of political images? The author argues that relying on technical means to detect AI-generated images is a losing game, as the technology will only continue to improve. Instead, the author suggests that the best way to ensure the veracity of photos is to trust the creator of the photo, not the photo itself. Provenance, not visuals, is the crucial factor.

3. How does the author view the potential for AI-generated fake videos compared to fake images? The author notes that while today's AI image generators can produce reasonably convincing fake images, the current state of AI video generators is still quite limited. The video the author created using RunwayML's Gen-3 Alpha was clearly unrealistic and could not pass as a genuine video of a campaign rally. The author suggests that for now, video evidence can be a powerful tool for verifying the authenticity of political events.