An Empirical Study on Trust, Ethical Concerns, and Bias in AI-Powered Educational Chatbots

Abstract

In contemporary ages, Artificial Intelligence has swiftly changed all the paces of education. AI tools, such as ChatGPT, and other chatbots are being widely harnessed to support students in their learning. As AI-driven chatbots become more prevalent and engaged in the education landscape, key issues of trust, ethics, and anticipated bias in automated interactions are becoming a major focus. When the earlier research has greatly concentrated on theoretical discussions and isolated assessments, it has been observed that the comprehensive empirical studies amalgamating user perception with experimental validation was lacking. To fill this gap, this study used a mixed-method research approach, where both systematic literature analysis and primary data collection was performed. A total of 134 research papers were read, from which 29 papers were chosen for detailed analysis to identify educational chatbots' dimensions of trust, ethics, and bias. To build on this, an investigation study was conducted on a sample of n = 120 students. They were divided into two groups. One group was engaged with an AI Chatbot developed by the researchers and another group was trained through conventional teaching methods including giving notes to the group and giving structured assessments. Both groups were given a standardized survey instrument to assess their perception of trust, ethical reliability and bias in learning interactions. It was concluded that there is a positive influence of the use of chatbot in learning on perceived accessibility and perceived engagement, and the level of trust increased by about 18.4% from the traditional group. Likewise, the score for ethical perception rose by about 13 percent, which meant that it became more in line with the principles of fairness, transparency, and accountability. The analysis of bias-related parameters, however, showed that there was an almost 8% mean bias indicating in need for further refining to reduce bias and make it more neutral to ensure the equality of responses. The differences observed was validated statistically by two sample t-test and it was found to be significant (p<0.05). This study contributes by offering an incorporated experiential context for assessing AI-powered educational chatbots and suggests hands-on understandings for developing more trustworthy, ethical, and unbiased AI systems in education.