Brazilian cultural diversity still does not appear in data that train AI

Amid the discussion of Law No. 11,645/2008, which determines the mandatory inclusion of Afro-Brazilian and indigenous history and culture in basic education curricula, a provocation arises: does Artificial Intelligence know the culture and plurality of Brazil?

Bamboo Data, a Brazilian datatech specialized in cultural datasets, brought to light the debate about representation in AI. Widely used by students, the technology continues to be trained, for the most part, with data produced outside the country, which expands the discussion about cultural representation in the artificial intelligence environment.

The scenario shows that the main AI models were trained with data produced in Europe and North America, leaving aside territories, languages, cultural references and forms of social organization that are part of the reality of a large portion of Brazilian students.

According to studies by the University of Southern California (USC), 38.6% of the "facts" used to train AI systems present distortions that affect non-Western groups, for example.

In this context, the debate is part of a global agenda that questions the governance of artificial intelligence, algorithmic diversity and digital sovereignty, in addition to discussing how these issues can be addressed based on Brazilian contexts and particularities.

In the educational environment, the discussion is even more relevant, since it is essential to value the ethnic-cultural diversities of students, avoiding historical erasures that education seeks to overcome.

USA says it will accelerate development and use of AI for security | NOW CNN

Bamboo Data is a Brazilian datatech dedicated to structuring and licensing cultural datasets for Artificial Intelligence training. The company works to collect, organize and annotate multimodal data aimed at training AI systems, focusing on communities historically underrepresented in digital environments.

Source: CNN