Using ChatGPT as a virtual mentor for K-12 students learning science improved student grades by 30% and provided both functional and emotional support when teachers were unavailable.
Objective: To evaluate the real impact of using ChatGPT as a virtual mentor for K-12 students learning science, within a blended learning framework that combines in-person teaching with AI-assisted independent learning.
Methods:
- Systematic assessment of ChatGPT's performance through a 52-question test covering K-12 chemistry and physics
- Quasi-experimental study with control and experimental groups (23 students total)
- Student grades analyzed before and after ChatGPT intervention
- Student perceptions measured using 5-point Likert scale surveys
- Evaluation conducted over multiple academic terms using both GPT-3.5 and GPT-4 models
Key Findings:
- ChatGPT demonstrated high accuracy in K-12 science (9.3/10 with GPT-3.5, improved to 9.7/10 with GPT-4)
- Experimental group showed 30% grade improvement compared to control group
- Student perception of ChatGPT as a useful educational tool increased from 37% to 70% after extended use
- AI provided both functional (academic) and emotional support for students lacking access to help at home
- 90% of students in experimental group showed improved performance regardless of initial proficiency level
Implications:
- ChatGPT can effectively supplement traditional teaching when teachers are unavailable
- The tool helps democratize access to educational support
- Can provide both academic and emotional support for students
- Particularly beneficial for students without access to tutoring or homework help at home
- Promotes student autonomy and self-efficacy
Limitations:
- Small sample size (23 students from one school)
- Non-randomized experimental design
- Potential Hawthorne effect (subjects aware of being studied)
- Some mathematical hallucinations in ChatGPT's responses
- Limited ability to handle image inputs/outputs
Future Directions:
- Conduct larger-scale randomized studies across multiple schools/regions
- Evaluate impact across different subjects and educational levels
- Assess long-term effects on student learning and development
- Study impact on teachers and teaching methodologies
- Investigate applications for students with learning difficulties
Title and Authors: "Use of ChatGPT as a Virtual Mentor on K-12 Students Learning Science in the Fourth Industrial Revolution" by Rafael Castañeda et al.
Published On: December 5, 2024
Published By: Knowledge (MDPI journal)