Unveiling Algorithmic Biases Through Personas: A Comparative Analysis of ENEM’s Essays on Gemini

• Luana B. Mendes UFMT
• Keslley W. C. Guimarães UFMT
• Igor C. Valenciano UFMT
• Bianca Cristina O. do E. S. Silva UFMT
• Patricia de Souza UFMT
• Thiago M. Ventura UFMT
• Allan G. de Oliveira UFMT

DOI: https://doi.org/10.5753/eniac.2025.13714

Resumo

The advancement of generative Artificial Intelligence (AI) models raises questions about their objectivity in evaluation tasks. This study investigates whether AI can assess essays in Portuguese impartially and how assigning author personas affects their evaluations. Using a comparative methodology, 801 essays were evaluated by a human and by the Gemini AI using anonymously and with five different personas. The AI demonstrated bias, assigning higher scores to personas perceived as more experienced and lower scores to less experienced ones. The research concludes that AI systems can amplify demographic biases, highlighting the risks of their application in educational contexts and the need to develop audit mechanisms to ensure equity.

Referências

Belenguer, L. (2022). “AI bias: exploring discriminatory algorithmic decision-making models and the application of possible machine-centric solutions adapted from the pharmaceutical industry.” In AI and Ethics (2).

Bozza, S., Roten, C.-A., Jover, A., Cammarota, V., Pousaz, L. and Taroni, F. (2023). “A model-independent redundancy measure for human versus ChatGPT authorship discrimination using a Bayesian probabilistic approach.” In Scientific Reports, 13(1), p.19217.

Bui, Ngoc My; Barrot, Jessie. "Using generative artificial intelligence as an automated essay scoring tool: a comparative study". In Innovation in Language Learning and Teaching, p. 1-16, 2025.

Dwivedi, Y.K. (2023). “‘So What If ChatGPT Wrote it?’ Multidisciplinary Perspectives on opportunities, Challenges and Implications of Generative Conversational AI for Research, Practice and Policy.” In International Journal of Information Management, 71(0268-4012), p.102642.

Ferrara, E. (2024). “The Butterfly Effect in artificial intelligence systems: Implications for AI bias and fairness.” In Machine Learning with Applications, 15(15), p.100525.

Heggler, J., Szmoski, R. and Miquelin, A. (2025). “As Dualidades entre o uso da inteligência artificial na educação e os riscos de vieses algorítmicos.” In Educação & Sociedade,46.

Jain, L.R. and Menon, V. (2023). “AI Algorithmic Bias: Understanding its Causes, Ethical and Social Implications. International” In Conference on Tools with Artificial Intelligence.

Krasanakis, E. and Papadopoulos, S. (2024). “Towards Standardizing AI Bias Exploration.” In Workshop on AI bias: Measurements, Mitigation, Explanation Strategies.

Liu, Y., Kong, W. and Merve, K. (2025). “ChatGPT applications in academic writing: a review of potential, limitations, and ethical challenges” In Arquivos brasileiros de oftalmologia, 88(3) .

Salminen, J., Wenyun Guan, K., Jung, S.-G. and Jansen, B. (2022). “Use Cases for Design Personas: A Systematic Review and New Frontiers.” In CHI Conference on Human Factors in Computing Systems, pp.1–21.

Stein, K., Harvey, A., Lopez, A., Taj, U., Watkins, S. and Watkins, L. (2024). “Eliciting and Measuring Toxic Bias in Human-to-Machine Interactions in Large Language Models.” In 2024 IEEE 15th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON), pp.13–19.