Biases, illusions, and decisions: to what extent can we trust artificial intelligence?

Abstract

The Artificial Intelligence (AI), developed since the 1950s, is a general-purpose technology that requires critical analysis of its ethical, epistemological, and sociotechnical implications. While it promises innovation and efficiency, it raises concerns such as algorithmic biases, social bubbles, hallucinations, and model opacity. In this theoretical essay, AI reliability is discussed from three main perspectives: biases, “illusions,” and the decisions made based on its outputs. A qualitative methodological approach is adopted, centered on a literature review and engaging with authors such as Rivoltella (2024) and Fantin (2024). It is shown that algorithmic biases reflect prejudices embedded in training data, as seen with ChatGPT and facial recognition systems that fail to identify black faces, reinforcing hegemonic cultures and affecting the educational field. The “illusions” contribute to the spread of misinformation and strengthen the post-truth era. In this context, AI literacy becomes essential so that users can recognize limitations and flaws. Automated decisions often rely on opaque models that compromise explainability and require active human mediation. Furthermore, the concentration of AI development within large private corporations exacerbates the lack of transparency and deepens technological inequalities, especially in peripheral countries. It is concluded that trust in AI should be understood as an ongoing and situated construction, intertwined with human decisions, economic interests, and power structures. Thus, trusting AI requires maintaining responsibility over its use, through critical mediation and grounding in principles of justice, equity, and cognitive freedom, reinforcing the urgent need for critical education and AI literacy.