Get a free exercise
AI-Enhanced Curricula: Stanford Scholars Explore How AI Can Transform Education
Developing new curricula for K-12 students is often a time-consuming and complex task, but large language models could streamline this process, making it faster and more effective.
Date of writing
October 25, 2024
Time of reading
2 minutes
Developing new curricula for K-12 students is often a time-consuming and complex task, requiring teachers to carefully design, test, and adjust educational content, according to Stanford University. Researchers from Stanford University are now exploring how large language models (LLMs) could streamline this process, making it faster and more effective.
In a study led by Joy He-Yueya, a PhD student at Stanford's AI Lab (SAIL), the team proposed using AI to generate and evaluate educational materials. "It’s a slow process with many logistical challenges. We thought, there might be a better way," He-Yueya explained. With support from the Hoffman-Yee Research Grant, He-Yueya collaborated with professors Emma Brunskill and Noah D. Goodman to investigate the potential of AI as a curriculum evaluator.
The researchers tested GPT-3.5's ability to predict student performance by evaluating math lessons tailored to different student personas. Their goal was to determine whether the model could replicate key educational phenomena, such as the Expertise Reversal Effect, where advanced learners benefit from minimal guidance, and the Variability Effect, which cautions against overwhelming students with too many practice problems.
The results were promising—AI assessments mirrored patterns well-known to education psychology, suggesting that LLMs could effectively evaluate instructional content.
Building on these insights, the researchers introduced an Instruction Optimization Approach, using one AI model to create new materials and another to evaluate them. This approach was tested with math worksheets, and feedback from 95 experienced teachers showed a high level of agreement with the AI’s evaluations.
While AI won’t replace teachers, it could become a valuable tool for instructional designers. "Our hope is that this approach could help support teachers and instructional designers," said Emma Brunskill, emphasizing the complementary role of AI in education.
For more details, read the full study published at the Educational Data Mining Conference: Evaluating and Optimizing Educational Content with Large Language Model Judgments.
In a study led by Joy He-Yueya, a PhD student at Stanford's AI Lab (SAIL), the team proposed using AI to generate and evaluate educational materials. "It’s a slow process with many logistical challenges. We thought, there might be a better way," He-Yueya explained. With support from the Hoffman-Yee Research Grant, He-Yueya collaborated with professors Emma Brunskill and Noah D. Goodman to investigate the potential of AI as a curriculum evaluator.
The researchers tested GPT-3.5's ability to predict student performance by evaluating math lessons tailored to different student personas. Their goal was to determine whether the model could replicate key educational phenomena, such as the Expertise Reversal Effect, where advanced learners benefit from minimal guidance, and the Variability Effect, which cautions against overwhelming students with too many practice problems.
The results were promising—AI assessments mirrored patterns well-known to education psychology, suggesting that LLMs could effectively evaluate instructional content.
Building on these insights, the researchers introduced an Instruction Optimization Approach, using one AI model to create new materials and another to evaluate them. This approach was tested with math worksheets, and feedback from 95 experienced teachers showed a high level of agreement with the AI’s evaluations.
While AI won’t replace teachers, it could become a valuable tool for instructional designers. "Our hope is that this approach could help support teachers and instructional designers," said Emma Brunskill, emphasizing the complementary role of AI in education.
For more details, read the full study published at the Educational Data Mining Conference: Evaluating and Optimizing Educational Content with Large Language Model Judgments.