How Should AI-generated Content be Labeled?

🌈 Abstract

The article discusses the challenges of labeling AI-generated content to combat the spread of misinformation. It presents the findings of a study that examined which terms are most closely associated with AI-generated and misleading content across different countries. The article also discusses the considerations for governments, tech companies, and industry associations in developing effective content labeling strategies.

🙋 Q&A

[01] Labeling AI-Generated Content

1. What are the key findings of the study on which terms are most closely associated with AI-generated and misleading content?

• The phrases "AI generated," "generated with an AI tool," and "AI manipulated" were most closely associated with content generated using AI.
• "Deepfake" and "manipulated" were most closely associated with potentially misleading content, whether created by AI or not.
• Terms like "edited," "synthetic," or "not real" were not closely associated with either AI-generated or misleading content.
• The differences between "AI manipulated" and "manipulated" were quite striking, with the "AI" qualifier dramatically changing which content participants understood the term as applying to.

2. What are the two main purposes that content labels could serve?

1. Indicating that content was generated using AI.
2. Showing that the content could mislead viewers, whether created by AI or not.

3. What are some of the challenges in developing effective content labeling strategies?

• Self-labeling by creators has obvious disadvantages, as few will admit their content is intentionally misleading.
• Relying on machine learning, crowdsourcing, and digital forensics to label content will become more challenging as the lines between human-made and AI-generated content continue to blur.
• Labels that work well in one country may not resonate in others, so testing of labels needs to be done across different cultures and languages.

[02] Regulatory and Industry Efforts

1. What are some of the regulatory and industry initiatives mentioned in the article related to labeling AI-generated content?

• President Biden's executive order on AI security and safety, which includes new standards and best practices for labeling AI-generated content.
• The AI Disclosure Act of 2023 proposed by U.S. Rep. Ritchie Torres, which would require a disclaimer on any content generated by AI.
• The Coalition for Content Provenance and Authenticity's development of an open technical standard for tracing the origins of content and determining whether it has been manipulated.

2. What are the potential benefits of using disclaimers, watermarks, or other labels to indicate how content was created or whether it is misleading? Studies have shown that social media users are less likely to believe or share content that is labeled as misleading.