Abstract:

The rapid adoption of generative AI (GenAI) in education has led to widespread use of AI chatbots by students. However, most existing systems are designed as general-purpose conversational tools, with limited consideration for pedagogical alignment. As a result, their use often introduces risks such as superficial learning, overreliance on generated answers, and reduced transparency in the learning process.

This paper presents a shift from tool-centric AI chatbots to pedagogically aligned AI tutors, grounded in insights from focus groups with students and education professionals. These stakeholders highlighted key limitations of current chatbot use, including the lack of alignment with course objectives, insufficient support for learning processes, and limited teacher control. The proposed design principles are further informed by real-world deployments of an AI tutor (ART: Augmented Reality Tutor), piloted across four countries and nine universities, with over one thousand student users. Usage data from these implementations provides additional insight into how students interact with AI tutors in real-life educational contexts.

Based on these combined insights, we propose a set of design principles for AI tutors in education. These include:
(1) teacher-controlled and curated knowledge sources,
(2) alignment with pedagogical models and learning objectives,
(3) transparency through source attribution and reasoning traces,
(4) support for active learning through questioning and feedback, and
(5) student modeling to enable personalized guidance.

The paper illustrates how these principles are implemented in ART, an AI-powered tutoring system designed to support learning processes rather than simply provide answers. Through a comparison between conventional chatbots and pedagogically aligned AI tutors, the contribution offers practical guidance for educators and developers seeking to integrate GenAI in a responsible and educationally meaningful way.

Keywords:

AI Tutors, pedagogical alignment, GenAI.