LLM-Guided Reinforcement Learning Improves Personalized AI Tutoring
A study introduces a tutoring platform that combines a generative AI chatbot with a reinforcement learning algorithm for organizing practice problems. In contrast to typical GenAI chatbots that respond to student inquiries, this system takes a proactive approach by choosing problems based on interaction cues. The platform was implemented during a five-month Python course in ten high schools in Taipei, collaborating with the Taipei City Government and the American Institute in Taiwan. Students were assigned to either a fixed order of practice problems or an adaptive sequence. The research, published on arXiv, investigates whether proactive guidance enhances the effectiveness of GenAI tutoring.
Key facts
- Generative AI is reshaping education by enabling personalized tutoring.
- Existing platforms focus on reactive chatbot tutors that answer student questions.
- The study hypothesizes that proactive guidance can improve chatbot tutor efficacy.
- A new platform integrates a GenAI chatbot with a reinforcement learning algorithm.
- The algorithm sequences practice problems based on student-chatbot interaction signals.
- The platform was deployed with the Taipei City Government and American Institute in Taiwan.
- A five-month Python course was conducted across ten high schools.
- Students were randomized between a fixed and adaptive practice problem sequence.
Entities
Institutions
- Taipei City Government
- American Institute in Taiwan
Locations
- Taipei
- Taiwan