Lingcam

This project, titled Lingcam: A Camera That “Develops Photos with Language”, aims to highlight the biases inherent in language. The artwork uses ChatGPT to verbalize every element and compositional detail captured by the sensor, and then directly reuses these words as prompts to generate new images.

For example, when photographing a coastline, the phrase “beautiful beach with a sunset” will generate a Ferris wheel due to the AI’s bias. In English-speaking countries, “beach + sunset” is easily associated with Santa Monica Beach, where abundant training data indicates the presence of a Ferris wheel. Consequently, the AI will add a Ferris wheel that is not mentioned in the original description, which is difficult to imagine from the perspective of non-English-speaking cultural contexts.

Many AI image generators today are multilingual, so we can enjoy their high performance without giving much thought to how they work. Yet even if a user provides input in a language other than English, the prompts are typically processed internally in English, with the resulting images influenced by the inherent biases of an English-language framework.

The camera can also use a language other than English for a photo description and use that language as a prompt without translating the language unintentionally, generating images that reflect the context of that language.

Though the images produced by this camera are algorithmically generated, the results are instantaneous and the act of pressing the shutter makes it feel like a photograph, expanding the concept of photography. Just as it is said that people perceive the world through language, the camera’s ability to reinterpret images through words reveals the bias of language.