Ambigrams are images than can be viewed both right side up and upside down. Illustrators in Japan and Western Europe explored the genre in the 18th and 19th centuries to express virtuosic drawing technique and comment on mankind’s dualities. Golan Levin and Lingdong Huang made their _Ambigrammatic Figures _using StyleGAN2, an AI protocol that simultaneously generates outputs and scores them for accuracy, then re-generates to achieve higher-scoring results. The wonky shapes, lurid textures, and inhuman uncanniness resulting from GAN processes define the aesthetics of early AI art.
Levin and Huang’s initial experimentation led them to a novel generation process. They trained their model on right-side-up faces, then prompted it to create upside-down faces using only the images in its dataset. _Ambigrammatic Faces emerge from the tension between those two commands. The final outputs—like #39, #43, and #55 _seen here—have sickly gray skin, splashes of pallid blue, and oddly placed wrinkles. They don’t look fully human or inhuman from any angle. We can clearly see the software’s limitations. But _Ambigrammatic Faces _aren’t meant to be perfect.
Levin and Huang note that early ambigrams “were often used to depict uncomplicated dualities,” whereas their project reflects “the moral ambiguities of a darker and more uncertain time.” Indeed, the bizarre expressions and rubber-mask compositions of their Ambigrammatic Faces never settle in a stable form. They are forever becoming or unbecoming.