This project was submitted by Christopher Tardy. It was one of the top submissions in our AI Alignment course (Mar 2024). Participants worked on these projects for 4 weeks.
As AI capabilities grow, distinguishing AI-generated content from human-generated content becomes increasingly difficult. This has implications for tracing misinformation, preventing intellectual property theft, and defining abstract notions of what is “real”. One solution is AI watermarking, which involves embedding metadata into digital content to validate its origin as AI-generated or human-generated. This aspect of AI safety is crucial because it’s easier to establish content’s authorship at creation than after it’s online. As AI-generated photos and text pose challenges for classification, companies like GPT Zero, Undetectable, and Turnitin have developed tools to identify AI-generated content, but they caution against using them to prosecute or punish individuals due to the technology’s potential flaws.
Read the full piece here.