AI helps math teachers to build better “scaffolding”

For those who are outside of education, it may be surprising that the most difficult aspect of education is often not what is happening in class, but the preparation which must occur outside, beyond normal hours of work. The most difficult work is in the planning and structuring of lessons for classes with variable knowledge and skill levels. And, with the loss of learning the pandemic, American classrooms – in particular college classes – are more than ever filled with students at various levels of competence.

In this context, IT and IT researchers from Stanford University have evaluated the ability of large -language models to help college math teachers to create lessons on several levels that allow them to feed those who could have lagged while simultaneously holding the interest of more advanced students. Everyone wins, say the researchers, especially the teachers for whom the model is a great partner of thought resurfaces the ideas that they may not have considered.

“Teachers spend so much time adapting programs to the needs of their students, but no one really asks – how can we support them in this process?” said Rizwaan Malik, a Knight-Hennessy scholar studying the data sciences of education at the Stanford Graduate School of Education. Malik is the first author On a new studypublished in the British Journal of Educational Technology, Presentation of the task and evaluation framework.

The article presents the first assessment framework for the lesson scaffolding based on the processes of expert teachers and the first experiences that test and adapt the LLM for this task.

“The idea of ​​scaffolding is trying to support the program that helps all students, whatever their place, to access the content of the program,” explains Dora DemszkyEducation teacher of education data and the main author of the newspaper. The work of Demszky and Malik was supported by the AI ​​seed subsidy program (HAI) on the Stanford Institute for Human.

Study teachers to train the model

Before starting to experiment with LLM, Demszky and Malik analyzed teacher planning to understand the fundamental principles of scaffolding. This is perhaps the most difficult part of the planning of lessons, explains Malik, a former professor of familiar mathematics with whims and the commitment of time of planning of lessons.

“The premise of the project was to see what technology can do to help teachers take a study program and prepare class in class,” explains Malik. “We are not only creating a tool, but a framework that helps teachers effectively scam the curriculum, ensuring that the content generated by AI align with real needs in class.”

In their analysis, they identified three stages that teachers follow in the creation of course plans: observation (evaluation of the skill levels of their students), the formulation of an educational strategy and the implementation of a scaffolding course which meets the needs of all students.

Better warm -up

The AI ​​model was designed to generate “warm -up” exercises that help students activate previous knowledge. In user assessments, these exercises generated by AI have been better assessed than those created by humans in terms of accessibility, alignment on the learning and preferably teachers' objectives.

The highest rated approach has fed the model an additional data set of original program equipment and used complex and nuanced prompts lit by an expert educator (Figure 1).