AI’s Learning Process: A Digital Collage Machine

AI art models don’t “imagine” or “create” the way humans do. Instead, they rely on vast datasets scraped from the internet, learning through a process known as deep learning and neural network training. These models analyze patterns, compositions, and stylistic features from millions of existing artworks—often without the artists’ permission.

AI art generators like Stable Diffusion and Midjourney use training datasets such as LAION-5B, which contain massive collections of images, including artwork from DeviantArt, ArtStation. Many of these artists explicitly state “Do not use for AI training”, yet their work is absorbed into datasets without consent. This article mentions the unauthorized use of the original creator’s photography by LAION-5B.

But how does this differ from human artists learning from their predecessors? The answer lies in how AI reconstructs images.

AI-Generated Text vs. AI-Generated Images

Some argue that AI simply follows the same process as human artists—learning from existing works and creating something new. However, there is a fundamental difference in how originality is perceived in AI-generated text versus AI-generated images.

AI-generated text, such as content from ChatGPT, does not directly copy sentences from books or articles. Instead, it reprocesses language statistically, making it difficult to pinpoint exact plagiarism. The result is often an entirely new arrangement of words, even if influenced by existing materials.

AI-generated images, however, function differently. Instead of merely reprocessing, they often replicate key visual elements from specific artworks. Unlike text, where ideas can be paraphrased, an AI-generated image can mirror the brushstrokes, shading, and even the composition of real artists. Some AI-generated works have even been found with faint traces of original watermarks, proving that elements were directly lifted from existing pieces.

This discrepancy leads to a double standard: If an AI-generated novel is considered “original” because it reshuffles words, why is an AI-generated painting considered original when it visually resembles human-made artwork?

This has led many artists to describe AI-generated art as “Frankensteining” —stitching together bits and pieces of different works to form a new composition, but without actual creative intent.

AI Training and the Uncontrollable Ingestion of Unauthorized Works

As AI-generated content continues to evolve, a critical issue remains unsolved—the unauthorized use of artists’ works in AI training datasets. Even when AI companies claim they use only publicly available or licensed datasets, human intervention makes it impossible to prevent unapproved data ingestion.

The Inescapable Problem: AI Can Always Be Fed Unauthorized Data

Even if AI companies attempt to regulate their training datasets, there is no way to stop individuals or rogue developers from feeding AI unauthorized material.

Unauthorized AI training happens in multiple ways:

• Individuals can manually upload unlicensed images to train personal AI models.
• Companies and developers can secretly collect unauthorized artworks and integrate them into private datasets without public disclosure.
• AI tools allow users to upload reference images for style-based AI generation, effectively making unauthorized AI training accessible to anyone.

This means that even if an artist explicitly states “Do not use for AI training,” their work can still be taken, manipulated, and replicated in ways they never approved.

AI’s Irreversible Learning: Once Trained, It Can’ t Forget

Even if AI companies agree to remove certain works from their datasets, the fundamental nature of AI learning makes true deletion impossible.

AI models do not simply store images in a database; instead, they extract stylistic patterns and embed them into a broader style model. Even if an artwork is deleted, AI retains knowledge of its brushstrokes, composition, and color techniques. This is similar to human memory—if you lose a book, you may forget specific details, but you can still summarize its story in your own words.

This raises serious concerns for artists:

• Even if an artist successfully requests AI companies to remove their works from training datasets, AI can still generate images that bear an uncanny resemblance to their style.
• There is no effective legal framework to define “style theft,” meaning artists have no way to protect themselves from AI mimicry.
• Many users do not realize that AI-generated art may be built on stolen work. Raising public awareness is crucial, as consumer-driven pressure on AI companies could help push for more ethical practices.

The Question We Must Ask Ourselves

As AI-generated art becomes more sophisticated, we must ask whether we are witnessing an evolution of creativity or the commodification of human artistic labor without consent. If AI-generated images continue to replicate and commercialize the styles of living artists, should laws be adapted to protect creative expression?

And perhaps the most pressing question of all: If AI can replicate any artistic style with near-perfect accuracy, what does originality even mean in the age of machine learning?