Classification of the lunar surface pattern by AI architectures: Does AI see a rabbit in the Moon?

TL;DR

The paper examines whether AI architectures perceive a rabbit in the Moon’s surface pattern and how this aligns with the Moon rabbit motif across cultures. It uses two tests—CLIP-based rabbit-face classification and a 1000-class ImageNet evaluation—across seven architectures, with carefully prepared binary silhouettes of lunar maria and orientation reflecting early-evening viewing. Results suggest a latitude-dependent tendency for rabbit-like classification at low latitudes and a tendency toward face-like classifications at higher latitudes, while ImageNet results show limited rabbit identification. The work argues that both static shape similarity and dynamic, cyclical patterns influence cultural imagination and proposes future work on video-based dynamic pattern classification to better capture the Moon-rabbit symbolism. The findings provide a quantitative lens on how AI and culture intersect in shaping pareidolic interpretations of natural patterns.

Abstract

In Asian countries, there is a tradition that a rabbit, known as the Moon rabbit, lives on the Moon. Typically, two reasons are mentioned for the origin of this tradition. The first reason is that the color pattern of the lunar surface resembles the shape of a rabbit. The second reason is that both the Moon and rabbits are symbols of fertility, as the Moon appears and disappears (i.e., waxing and waning) cyclically and rabbits are known for their high fertility. Considering the latter reason, is the color pattern of the lunar surface not similar to a rabbit? Here, the similarity between rabbit and the lunar surface pattern was evaluated using seven AI architectures. In the test conducted with Contrastive Language-Image Pre-Training (CLIP), which can classify images based on given words, it was assumed that people frequently observe the Moon in the early evening. Under this condition, the lunar surface pattern was found to be more similar to a rabbit than a face in low-latitude regions, while it could also be classified as a face as the latitude increases. This result is consistent with that the oldest literatures about the Moon rabbit were written in India and that a tradition of seeing a human face in the Moon exists in Europe. In a 1000-class test using seven AI architectures, ConvNeXt and CLIP sometimes classified the lunar surface pattern as a rabbit with relatively high probabilities. Cultures are generated by our attitude to the environment. Both dynamic and static similarities may be essential to induce our imagination.

Figures (5)

• Figure 1: (a): Contrast of the surface of the Moon. (b): Illustration of the Moon rabbit in the Japanese book Zouho-houryaku-ohzassho (1781 CE, Author’s collection). While the Moon rabbit makes an elixir of immortality in Chinese culture, in Japanese tradition, the Moon rabbit is pounding rice cake on the Moon. On the left side of the rabbit, there is an osmanthus fragrans tree, which is also dipicted as an object on the Moon in Asian tradition.
• Figure 2: Two types of similarity between the rabbit and the Moon as the origin of the Moon rabbit. It has been suggested that the Moon rabbit originated from the pattern of the lunar maria, which is similar to the shape of a rabbit. In cultural anthropology, it has been indicated that the origin of the Moon rabbit is also linked to the cyclic appearance of rabbits and the Moon. While the similarity of shapes is considered static similarity, the relationship of their behaviors can be termed dynamic similarity. In this work, the main focus is on static similarity, as evaluated by the AI architectures. The effects of dynamic similarity are discussed in the discussion section.
• Figure 3: Postures of the full Moon simulated by Stellarium (https://stellarium.org) at different times and latitudes in January (a) and July (b) in 500 BCE. The times indicate the local time at each latitude. The images of the Moon are cited from Stellarium under the public domain licenses of NASA & JPL.
• Figure 4: Process of image preparation. Using OpenCV, four types of images were prepared.
• Figure 5: Tested lunar images and probabilities of "rabbit" and "face" judged by CLIP. Selected classes (those with higher probabilities) are shown in bold red. Attention maps for the selected class are overlaid on each image. The red areas indicate the regions that CLIP focused on when selecting the classes. At a threshold of 90 without blurring, an attention map for "face" is shown because the probabilities for both classes are the same. "A," "B," "C," and "D" in the images selected as "rabbit" correspond to the images tested in the ImageNet test, as shown in Table \ref{['imagenet']}.