Leveraging Generative AI Models to Explore Human Identity

TL;DR

The paper investigates human identity indirectly by aligning diffusion-based face generation with identity formation, treating initial noise as innate factors and color-specific external noise as environmental influences. Using a pre-trained Latent Diffusion Model trained on CelebA-HQ, it demonstrates that color-conditioned noise injected during early denoising steps yields continuous, color-coherent changes in generated faces, enabling a visual metaphor for identity fluidity. It introduces Fluidity of Human Identity, a video artwork produced from AI-generated frames that embodies this fluidity without manual editing. The work bridges psychoanalytic theory, AI-based art, and diffusion-based generative methods, suggesting that AI can serve as a medium to explore and express aspects of the human mind and identity, with potential extensions to cognitive exploration via LLMs.

Abstract

This paper attempts to explore human identity by utilizing neural networks in an indirect manner. For this exploration, we adopt diffusion models, state-of-the-art AI generative models trained to create human face images. By relating the generated human face to human identity, we establish a correspondence between the face image generation process of the diffusion model and the process of human identity formation. Through experiments with the diffusion model, we observe that changes in its external input result in significant changes in the generated face image. Based on the correspondence, we indirectly confirm the dependence of human identity on external factors in the process of human identity formation. Furthermore, we introduce \textit{Fluidity of Human Identity}, a video artwork that expresses the fluid nature of human identity affected by varying external factors. The video is available at https://www.behance.net/gallery/219958453/Fluidity-of-Human-Identity?.

Figures (8)

• Figure 1: Correspondence between the human identity formation process and the face image generation process.
• Figure 2: The mechanism of diffusion models. The training stage consists of a forward process, where noise is incrementally added to a training image, and a denoising (reverse) process, where the model learns to progressively denoise and recover the image. After training, the diffusion model can generate a new image by applying the denoising process to pure random noise.
• Figure 3: Diffusion model outputs based on color-specific additional noise.
• Figure 4: Path 1, shown on the left, represents a trajectory resembling a ball bouncing on the ground. The generated human face images on the right illustrate variations as the color of the additional noise changes along this path. Blue and green points indicate the start and end points, respectively, while red points represent sampled color coordinates along the path.
• Figure 5: Human face images generated when the color of the additional noise is adjusted almost continuously with very small sampling intervals in Path 1.