When you post a comment, write a blog entry, or share a piece of art online, you have a reasonable expectation. You assume another person will read, view, or engage with it. This unwritten social agreement, a sort of human-creative contract, has governed the web for decades. We share for connection, expression, and communication with other people. But Large Language Models (LLMs) operate outside this contract. They don't read for pleasure or understanding; they ingest data to extract patterns, style, and knowledge, ultimately building a commercial product.

This practice raises a fundamental question: who owns the data that fuels AI? While something may be publicly accessible, that doesn't automatically make it free for any and all use. The debate cuts to the core of digital ownership, moving beyond legal technicalities into the realm of moral rights.

Two long-standing philosophical ideas help frame why using our data to train AI feels like a violation. The first comes from the 17th-century philosopher John Locke. His labor theory of property states that when a person mixes their labor with a common resource, that resource becomes their property. If you pick an apple from a wild tree, your effort of picking it makes the apple yours. In the digital world, your labor is the act of creation: writing a poem, coding a program, or taking a photograph. By this logic, the content you create is an extension of your effort and belongs to you.

A second perspective comes from the 19th-century philosopher Georg Wilhelm Friedrich Hegel. For him, property wasn't just about labor; it was an essential expression of personality. Your belongings, creations, and even your ideas are external manifestations of your identity. According to this personality theory, when an LLM scrapes your personal blog to learn your unique writing style, it is consuming a part of your identity. The model isn't just taking text; it's abstracting a piece of you.

AI developers often defend data scraping under a legal doctrine known as fair use, which permits the limited use of copyrighted material without permission for purposes like criticism, research, and education. They propose a new interpretation called "fair learning," arguing that training an AI is a transformative act. The model, they claim, isn't just a database of stolen text; it

learns from the data to generate something entirely new. In this view, the AI is like a student studying a vast library to develop its own understanding of the world. Critics, however, argue this analogy is flawed. A student uses knowledge for their own growth; an LLM uses data to create a commercial product that can directly compete with the human creators it learned from. This isn't transformation, they argue, but a 'free ride'—a direct extraction of value from the labor and personality embedded in the training data.

Current LLM training practices often prioritize data quantity above all else, bypassing these nuanced moral and philosophical considerations. By scraping the web indiscriminately, developers absorb a rich tapestry of human expression created under a very different set of assumptions. The result is a powerful technology built on a foundation of uncredited and uncompensated human creativity.

This tension between innovation and creators' rights is actively being fought in courtrooms and parliaments around the world. The outcome will shape not only the future of artificial intelligence but also the very meaning of ownership in a digital society.