AI and intellectual property: who owns the work

Duration 4:01

Who owns what an AI produces? In 2024, the US Copyright Office ruled: a work generated entirely by an AI, without a human hand, is not protectable. In parallel, the European AI Act is coming into force with its intellectual property provisions. This film traces copyright, patents and the real underlying question: human contribution.

Chapters

Every chapter below is clickable.

Transcript

Who owns what an artificial intelligence produces? The question is no longer abstract. It is already before the courts and the regulators. In 2024, the US Copyright Office ruled. A work produced entirely by an AI, without a human hand, is not protectable. The reasoning is precise. Copyright protects a human creation. Without an identifiable human author, there is quite simply no rights holder. For a company, the consequence is direct. A visual generated without any human intervention may belong to no one. And become freely reusable. The European Union is moving in parallel. Its regulation on AI is coming into force. It contains provisions that bear on intellectual property. The debate did not begin with the recent chatbots. As early as the 1950s, whether a machine could be creative was already discussed. The real turning point comes in the 2010s. Neural networks and machine learning then make the question central. Language models made it tangible. OpenAI's GPT showed that a machine could write a credible text. The other front is patents. The first battles posed a simple question. Can an invention found by an AI be patented? In the United States as in Europe, the law answered no. An invention must be attributable to a person to give rise to a patent. So the real subject is not the machine. It concerns inventiveness and the share of human contribution. On the creative side, three names come up again and again. OpenAI with GPT, Stability AI with Stable Diffusion, and Midjourney for images. The stakes go well beyond marketing and text. With its AlphaFold system, Google DeepMind cracked protein folding. A major advance for biomedical research. And the same question comes back. Who owns a result produced by a machine? Beneath these systems, the same mechanics. Machine learning and deep neural networks learn from vast volumes of data. Their exact scale is kept deliberately opaque. The number of parameters in these models is not always published by their developers. The result is unsettling. A reader cannot always tell this text apart from one written by a human. Hence the legal puzzle. The answer lies in traceability. These systems must be able to be audited, to understand how a decision was reached. Everything hinges on a single definition. What is a genuine human contribution? Without it, neither copyright nor patent applies. Finally, there remains the heavy question of liability. If an AI infringes existing rights, who must answer for it before the law? For a company deploying AI in contact with the public, these debates become very concrete. Data governance is no longer a theoretical detail. At Koaee, we design conversational assistants for businesses. Our servers are in Europe, and no conversation is kept. They understand speech and speak sixteen languages. A company's customer addresses the assistant in their own language, and receives a reply in that same language. This video only skims the subject. The full article is waiting for you in the comments. Curious about AI in general? Subscribe and leave a like. See you soon on koaee.ai.

Read the full article

Conversational assistants for business