*Originally published in Jota.

**This is an AI-powered machine translation of the original text in Portuguese

Bill No. 2,338/2023, which regulates Artificial Intelligence (AI) in Brazil, is about to be voted on in the Chamber of Deputies. Its most controversial chapter deals with copyright, establishing, among its provisions, the obligation for developers of Generative AI (GAI) models to: (i) specifically indicate the copyrighted works used in the dataset for training the model; (ii) manage authors’ consent for the use of such works; and (iii) remunerate authors for the use of their works.

These three requirements would entail significant costs for GAI developers in the country, leading to three serious problems. First, they would reduce the quality and development potential of GAI systems. Second, they would create significant barriers to entry and to the development of AI models, potentially resulting in market concentration led by foreign companies. Third, they would cause national works not to be used by foreign leaders, alienating national culture from these new technologies that are widely used by Brazilian citizens.

The first negative consequence is a corollary of the increased costs of using data to train GAI systems, which, according to a report on the European Union, reduces the availability of high-quality data.

Under the remuneration model proposed by the Bill, such costs would be incurred (with the exception of scientific research) even if the model is not commercially exploited. Only a fraction of trained GAI models find applications and are actually launched. A recent study published by MIT shows that, despite high levels of investment, only 40% of GAI projects aimed at the general public are realized as commercial products, and only 5% of projects aimed at the corporate sector result in signed contracts. Experimentation is extensive and necessary. However, imposing costs already at the development stage undermines this driving force behind the development of new methodologies and technological advancement.

Regarding barriers to entry, for startups the three requirements represent prohibitive overhead: hiring or training compliance officers, lawyers, and copyright specialists; implementing record-keeping systems, audit reports, takedown and monitoring processes to track works added or whose use has been unauthorized.

Foreign companies that lead in digital services and have incorporated GAI functionalities into their products already possess scale, financial resources, legal teams, data curation infrastructure, and experience to absorb such costs. The same is not true for domestic entrants, which have been flourishing vigorously in Brazil. The competitive advantage for startups lies in agility, experimentation, rapid development cycles, and the launch of minimum viable products—advantages that would be virtually eliminated, thereby reducing innovation in Brazil.

Finally, it should be noted that leading foreign companies can develop models without relying on national works and still deliver products that are useful to Brazilian consumers. The proposed cost increases would certainly lead them to avoid using national copyrighted content or to curate datasets by selecting only public-domain works or works by already established artists, thereby elitizing culture. It would be fanciful to believe that Brazilian companies would be able to develop competitive models based on national content, given the disproportionately higher costs required to achieve comparable quality, which could exclude Brazil from this market—except, perhaps, in certain niche applications for corporate customers.

As a result, we would face the undesirable consequence of seeing national content not incorporated into this new technology and new way of consuming and producing culture. Considering the growing popularity of AI assistants and content-generation systems in Brazil, this could amount to a process of digital colonialism, in which access to technology implies access to content trained on data from another culture, with different values and historical roots.

The remuneration model proposed in Bill No. 2,338/23, however, threatens cultural and technological autonomy and even digital sovereignty in the country. It is important to remunerate authors for the use of works in the training of GAI systems and to foster Brazilian artistic and literary production, but not through the payment of copyright royalties.