🧠 AI Ed Wiki

📄 Research Article

The Path to Conversational AI Tutors: Integrating Tutoring Best Practices and Targeted Technologies to Produce Scalable AI Agents

● high · 2026-05-11intelligent-tutoring, generative-ai, llm, scaffolding, personalized-learning, feedback-loop, adaptive-learning

Authors: Kirk Vanacore, Ryan S. Baker, Avery H. Closser, Jeremy Roschelle Year: 2026 Venue: arXiv (cs.HC)

Synthesizes intelligent tutoring systems research and generative AI into a keep/change/center/study framework for conversational tutoring systems, arguing proven ITS technologies should anchor generative tutors while centering student meaning-making and agency.

Key Findings

  • The paper synthesizes intelligent tutoring systems (ITS) research and generative AI, proposing a keep, change, center, study framework for designing conversational tutoring systems.
  • Keep: proven ITS technologies — knowledge tracing, affect detection, and related student modeling — remain valuable for diagnosing students' cognitive and emotional states.
  • Change: tutoring delivery is transformed by generative AI's capacity for dynamic content generation and dialogic scaffolding, replacing rule-based, limited-scope feedback with flexible open-ended dialogue.
  • Center: the student's meaning-making and agency, engaging students' thoughts, questions, and misconceptions the way high-quality human tutors do.
  • Study: the field must test efficacy, student experience, and integration, since conversational tutors can now be built quickly and easily but their effectiveness is not yet established.
  • Legacy ITS technologies such as Knowledge Tracing, Knowledge Spaces, and Epistemic Emotion Detection can diagnose states like slips, lack of mastery, or misconceptions; large language models integrated with structured knowledge representations may identify not just whether an answer is correct but why a student responded that way.

    • From Problem Sets to Dialogue

    Whereas most current ITS function, in essence, as interactive and adaptive problem sets with feedback and hints, conversational tutors hold the potential to simulate high-quality human tutoring by engaging with students' thoughts, questions, and misconceptions through natural language dialogue. Earlier dialogue systems such as AutoTutor and Watson Tutor demonstrated the possibility but were limited by rule-based response generation. Generative AI changes the capacity profile: interactive feedback can encourage constructive behaviors such as self-repair and knowledge construction, and tutorial dialogue can follow up on errors with diagnosis questions, supporting seamless formative assessment.

    Implications for AI in Education

    The keep/change/center/study framework gives researchers and developers a discipline for building conversational tutors that are both scalable and pedagogically grounded. It warns against discarding decades of ITS research in the rush to generative models: Knowledge Tracing and affect detection remain the diagnostic backbone, while Generative AI supplies flexible delivery. For educators, the framework's emphasis on centering meaning-making and student agency aligns with Intelligent Tutoring goals and with Scaffolding that keeps the learner active; its insistence on studying efficacy and integration reflects the reality that easy deployment of conversational agents has outpaced evidence about what works.

    Connected Concepts

  • Knowledge Tracing
  • Intelligent Tutoring
  • Lifelong Learning
  • Personalized Learning
  • Affective Tutoring
  • Scaffolding
  • Pedagogical Agent
  • Pedagogical LLM Training

    • Connected Articles

  • Learnmate2 LLM Adaptive Learning — LearnMate^2: Design and Evaluation of an LLM-powered Personalized and Adaptive Support System for Online Learning
  • Codify Socratic Tutoring Programming — Codify: An Intelligent Socratic Tutoring System for Programming Education
  • Eduagentbench Agent Teaching Benchmark — Are Agents Ready to Teach? A Multi-Stage Benchmark for Real-World Teaching Workflows
  • AI Coaching RL Skill Development — AI Coaching for Accelerating Human Skill Development with Reinforcement Learning
  • AI STEM Bibliometric Trends — Why does AI unlock new possibilities in STEM education? A Bibliometric Analysis of Trends and Future Agenda
  • Retrieval Augmented Tutoring Algorithm Kite — Retrieval-Augmented Tutoring for Algorithm Tracing and Problem-Solving in AI Education

    • Citation

    Kirk Vanacore et al. (2026). The Path to Conversational AI Tutors: Integrating Tutoring Best Practices and Targeted Technologies to Produce Scalable AI Agents. arXiv:2602.19303. cs.HC.