AI in Education, Vol. 1, Pages 3: AI, Ethics, and Cognitive Bias: An LLM-Based Synthetic Simulation for Education and Research

AI in Education doi: 10.3390/aieduc1010003

Authors:
Ana Luize Bertoncini
Raul Matsushita
Sergio Da Silva

This study examines how cognitive biases may shape ethical decision-making in AI-mediated environments, particularly within education and research. As AI tools increasingly influence human judgment, biases such as normalization, complacency, rationalization, and authority bias can lead to ethical lapses, including academic misconduct, uncritical reliance on AI-generated content, and acceptance of misinformation. To explore these dynamics, we developed an LLM-generated synthetic behavior estimation framework that modeled six decision-making scenarios with probabilistic representations of key cognitive biases. The scenarios addressed issues ranging from loss of human agency to biased evaluations and homogenization of thought. Statistical summaries of the synthetic dataset indicated that 71% of agents engaged in unethical behavior influenced by biases like normalization and complacency, 78% relied on AI outputs without scrutiny due to automation and authority biases, and misinformation was accepted in 65% of cases, largely driven by projection and authority biases. These statistics are descriptive of this synthetic dataset only and are not intended as inferential claims about real-world populations. The findings nevertheless suggest the potential value of targeted interventions—such as AI literacy programs, systematic bias audits, and equitable access to AI tools—to promote responsible AI use. As a proof-of-concept, the framework offers controlled exploratory insights, but all reported outcomes reflect text-based pattern generation by an LLM rather than observed human behavior. Future research should validate and extend these findings with longitudinal and field data.

Source link

Ana Luize Bertoncini www.mdpi.com