ÃÛÁÄÖ±²¥

The global political reaction to this burgeoning industry is mixed. While the United States and China embrace AI development, the European Union has stopped local companies in their tracks with its heavy-handed AI Act.

Proponents of this regulation argue that AI is bound to violate intellectual property. The AI Act even allows copyright holders to ÃÛÁÄÖ±²¥œopt outÃÛÁÄÖ±²¥� of having their data used for machine learning without giving a precise mechanism on how to do so, potentially resulting in a slew of infringement lawsuits in the future. Moreover, the idea of plagiarizing AI is expressed and even taken to the courts not only by regulators but also by writers and artists, even in the generally AI-friendly United States.

The idea is this: To be able to perform, AI had to be trained on creative work done by people. Thus, AI is said to not be ÃÛÁÄÖ±²¥œreally creative,ÃÛÁÄÖ±²¥� but imitative, borrowing from the creativity of the original authors and designers ÃÛÁÄÖ±²¥” effectively plagiarizing them.

If we try to apply the same standards to human creative work, we quickly see that this does not make sense. Without being trained on work done by others, people would also be utterly incapable of creative work.

With the standards AI seems to have to fulfill to be deemed ÃÛÁÄÖ±²¥œnon-plagiarizingÃÛÁÄÖ±²¥� or original, no scientist should read other papers from his field before writing his own. No artist should study at a school of art before contributing to it, and so on. No journalist should do an ounce of research before writing an article because surely, information and maybe even phrasing would enter his brain from these other texts and end up in his.

In other words, when it comes to judging AIÃÛÁÄÖ±²¥™s creative capabilities, people are suddenly elevated to magical creatures that generate ex nihilo (out of nothing) information in their brains. The harsh truth is that the amount of information required to perform creative work can never arise from random particle movement in the brain but must, in some form, be derived from information that entered the brain from the outside. In this sense, people are not different from AI.

We know that. This is why itÃÛÁÄÖ±²¥™s not considered plagiarism to have done some research or studying before embarking on creative work ourselves. On the contrary, it is considered essential. Taking information from various precursors and mixing it together is the usual creative process.

Plagiarism is a real issue between human creators, but it is never judged at the input level. Instead, the output is considered: In fiction and art, the final work should not be too similar to something a single other source created. In non-fiction, parts of oneÃÛÁÄÖ±²¥™s work can be taken directly from different sources, but it is required to cite them.

The EU AI Act and the lawsuits turn this principle on its head by judging input: Denying AI to be trained on oneÃÛÁÄÖ±²¥™s data is like denying humans to read oneÃÛÁÄÖ±²¥™s book or article or study oneÃÛÁÄÖ±²¥™s work of art for fear that the creative work they produce afterward might contain traces of the proprietary information they took in. This is an incredibly restrictive idea of intellectual property that has never been applied to human creators, and for good reason.

The same standards should be applied to AI. Any judgment of plagiarism should take place at the output level. Doing this, the justification for regulation and legal actions quickly falls apart.