In an increasingly technology-driven world, the dissemination of Artificial Intelligence (AI) is already a reality. Therefore, its regulation has been the focus of international attention, and several countries, including Brazil, are trying to get ahead in this regard. Here, several points of disagreement have contributed to the postponement of the voting on Bill (PL) 2,338/2023, and recently the Federal Court of Accounts (TCU) identified several risks related to these regulations that are under consideration in the National Congress.
The work of the TCU put the Brazilian Artificial Intelligence Strategy (Ebia) into perspective, addressing and analyzing how this could hinder its implementation, as well as the possible impacts resulting from the eventual approval of these proposals on the public and private sectors. Established by an ordinance from the Ministry of Science, Technology, and Innovation (MCTI), Ebia assumes the role of guiding the Brazilian State in favor of initiatives, in their various aspects, that promote the development of Artificial Intelligence solutions, as well as their conscious and ethical use.
For the professor at the University of Brasília (UnB) and researcher in the field of technological innovation applied to health, Dr. Paulo Henrique de Souza Bermejo, it is especially important to consider whether Brazilian regulation may restrict innovation, going against Ebia. It is worth mentioning that Ebia has six main objectives: to contribute to the development of ethical principles for responsible AI development and use; to promote sustained investments in AI research and development; to remove barriers to AI innovation; to train and empower professionals for the AI ecosystem; to stimulate Brazilian innovation and AI in the international environment; and to promote cooperation among public and private entities, industry, and research centers for the progress of Artificial Intelligence.
Paulo highlighted that excessive regulation can overburden and increase the cost of developing AI systems in the country, leading to excessive spending of time and money for companies to comply with complex regulatory rules, instead of investing resources in improving the technology. This would promote the monopoly of a few financially capable companies to regularly create systems in order to comply with the possible published standards. Thus, startups and smaller companies would have no way to compete with this, risking the objectives of Ebia itself. "All of this scenario could result in the loss of international competitiveness of various productive sectors, so it is very important to find a balance between regulation and innovation to ensure that AI is developed and used ethically and responsibly, without hindering its evolution," added the professor.
He explained that one of the most essential aspects of the field in Brazil is to position the country as an international competitor, both in the development of AI itself and in its application in the productive sector. "The amount of investment and the regulation proposed in each country will directly influence their positioning relative to other countries. This means that the domestic industry can have more autonomy, developing its own technology and aiming for export, or becoming more dependent on importing technology from other countries. In a globalized market, Brazilian companies need to be capable of competing with foreign ones, keeping up with technological advancements to make the production process more efficient," he explained.
The professor also pointed out that the regulation distinguishes AI systems from other systems. In my view, in the medium and long term, this will apply to virtually all types of software, considering that less intelligent software systems tend to become obsolete and be replaced more quickly by others. In other words, this could possibly lead to the implementation of a series of restrictions on all types of software, including those made abroad. That is, it is expected to impact the national software industry as well as Brazilian society as a whole, as it could delay the introduction of new technologies in the country, which would make the national productive system, related to various sectors of the economy, more efficient and competitive," he emphasized.
Another aspect to consider concerns restrictions on experimentation, since academic research in AI usually involves experiments with new methods and algorithms. "Restrictive regulations can limit researchers' freedom to test other approaches, which can hinder the discovery of new solutions. Additionally, they can impede collaboration among researchers, whether from different institutions or countries. This can happen, for example, if regulations deny the sharing of essential datasets for AI advancement, limiting the ability to create more efficient algorithms," he endorsed.
The need for flexible regulation
The researcher indicated that an ideal AI regulation in Brazil would be flexible, aligning with society's understanding of the technology. "Since there are so many factors involved, the best approach is not to rush regulation in Brazil, but rather to deepen the debate and include all stakeholders, such as the government, private sector, civil society, and academia, in the decision-making process. Regulation is very necessary, but it must be done cautiously to address all aspects of the current circumstances. Especially because, given the rapid evolution of these technologies, new situations constantly arise that are not yet foreseen in current laws or in some of the bills under consideration in Brazil. Therefore, the urgent thing is to discuss and reflect on the topic, but not necessarily to reach a conclusion."
In another aspect of this context, according to Paulo, it is important to keep in mind that AI applications cover different sectors of the economy, so that each of them would require specific regulations, making it more reasonable to adopt different governance measures, as is the case in the USA.
Copyright and data mining
For the teacher, regulation that exceeds reasonable limits can hinder not only potential but also the ability of an entire community to advance and prosper. Therefore, one of the points discussed is how to balance the protection of individual and collective rights without hindering technological progress, especially regarding data mining.
According to Paulo, data availability and mining are crucial for the development of AI, especially for machine learning models, which require the exploration of large volumes of data. One example of this lies in the realm of large language models, such as GPT-3 and 4, the basis for ChatGPT, which are trained on vast amounts of internet texts, learning to generate human-like content, understand instructions, among other things.
The issue that is outlined, then, as he stated, is about the copyright of the data used for training. "A disproportionate regulation decision can hinder the development and use of AI systems that require the processing of large volumes of data, including text, audio, and images. This would create even more obstacles for the national market," he argued.
In another perspective of this context, the specialist noted that the issue of copyright has been widely debated worldwide, with these technologies receiving numerous criticisms. For Professor Bermejo, it is necessary to reach a consensus, in the sense that the major manufacturers of these technological instruments come to an agreement with content creators in case of intellectual property violations.
Finally, according to the specialist, training language models in Portuguese is a key point because although commercial models like Gemini and ChatGPT "speak" our language, it often happens through translations, which can result in biased and incorrect responses. On the other hand, Brazilians also produce a large volume of data, which is targeted by foreign "big techs" capable of training large artificial intelligence models, making them potentially profitable alternatives within this market, with an AI that is genuinely Brazilian.
