Developer Eric Lu has created Ghost Font, a typeface that humans can read easily but advanced AI models struggle to process. The experimental project uses moving dots to form letters, exposing gaps in how current artificial intelligence interprets dynamic visual patterns.
Developer Eric Lu has launched an experimental typography project called Ghost Font that highlights a notable difference between human vision and modern artificial intelligence capabilities. The font is designed to be readable by humans while remaining difficult for many leading AI vision models to correctly interpret.
How Ghost Font Operates
Unlike standard fonts that use static letter shapes, Ghost Font utilizes thousands of tiny, moving dots. These dots move in specific patterns to form characters that are recognizable to the human brain. The design relies on human motion perception, where the brain naturally combines continuous movement into coherent text. When the animation is paused, the letters disappear, and the display appears as a collection of random dots. In recent demonstrations, while human observers could easily identify the hidden words, several advanced multimodal AI systems either failed to read the text or misidentified the intended message entirely.
Challenges for AI Vision Systems
This project demonstrates how current AI models process visual information. Many top-tier vision systems often analyze video content frame-by-frame as discrete images. Because the information in Ghost Font is dependent on continuous motion over time, these AI models struggle to piece together the patterns. In some test cases, the models were also misled by decoy text, leading to inaccurate results. This highlights a gap in how machines handle dynamic visual data compared to human observers who excel at recognizing patterns within movement.
Not a Security Solution
While the project has generated significant interest, it is not currently a replacement for standard cybersecurity measures like encryption. Tech observers and experts note that while the font poses a challenge to current models, it is not an impenetrable barrier. Techniques such as optical flow analysis—a computer vision method used to track motion—could potentially recover the hidden text. Furthermore, as AI developers continue to train models on more complex temporal data, the ability of machines to interpret such motion-based patterns is likely to improve.
Potential Future Uses
Despite these limitations, the concept has prompted discussions regarding new ways to verify human users online. Similar to the early days of CAPTCHA, which used tasks easy for humans but difficult for early computers to solve, Ghost Font could find applications in human-verification tools or privacy-focused communication. The long-term practical utility of such experiments remains to be seen, as the technology sector continues to advance in visual processing capabilities.