Description
AI can be a fantastic aid to learning physics, but it is alas altogether too easy for students to use it to replace their learning, rather than assisting it. We present a series of experiments in trying to train and encourage students to use AI to enhance their learning, spread across several 1st and 2nd year physics and astrophysics classes.
Students are trained in prompting techniques and in the pedagogy of effective and ineffective learning.
In our first experiment, students used AI to help them understand any aspect of the course they were struggling with, and they submitted their chat logs for us to mark and provide feedback on. The major issue here is that students often did realise how poor their understanding was of certain topics, and hence did not seek help.
We next tried to force students into a better realisation of the gaps in their understanding, by setting challenging problems each week as prompts. Once again, the students would use AI (or any other resources they preferred) to better learn the content, and would then submit "evidence of learning", marked on an effort basis. This seemed to work well during the course, but final exam results suggest that the students had not acquired a high level of proficiency from it. We suspect that the students get the AI to provide the answer too early, before they have really struggled with solving the problems themselves.
A third approach is to give students extremely challenging problems to solve - problems that are too hard for the students to solve themselves, and which are too hard for current AIs to solve without careful prompting. Finding suitable problems is extremely hard and constantly changing! This can work well, but only in very specific contexts.
| I am the presenting author | Yes |
|---|