A new typography experiment called Ghost Font is drawing attention for doing something unusual. It produces texts that are readable by humans yet cannot be recognized by many vision-based AI systems. Instead of displaying visible letterforms, the project uses moving dots to reveal hidden words. The test demonstrates an important distinction between human and AI perceptions rather than creating a new form of digital security.
Ghost Font is an invention of Eric Lu, that attempts to bridge the gap between human and machine perception. No visible characters appear in Ghost Font. Rather, there are thousands of tiny dots that move across the screen. The dots representing hidden letters move in one direction, while the remaining dots move in a different direction.
Human brains tend to group these moving dots and reconstruct the invisible word. But once the dots stop moving, the message becomes a part of visual noise.
Many current multimodal AI systems analyze videos as a sequence of individual frames. Since Ghost Font removes the clear outlines and high-contrast edges that optical character recognition depends on, AI models often fail to identify the intended message. Some even read an intentionally placed decoy text instead.
Through the experiment, it is clear that human vision perceives motion differently compared to the vast majority of AI. Moreover, the advancements in image recognition are not always matched by advances in understanding motion.
Despite its increasing popularity, Ghost Font is neither an encryption method nor any other means of protecting data from AI. According to researchers, Ghost Font cannot fully protect information, as special algorithms for processing computer vision, such as optical flow algorithms, can decipher the hidden text by analyzing the movement of frames.
As AI video understanding improves, future models are expected to decode such animations more reliably. Experts therefore view Ghost Font as a perception experiment rather than a long-term defence against AI systems.
Also Read: Apple-OpenAI Legal Dispute Takes New Turn Over Email Mix-Up