The Vraagbaak AI in Education is collecting brilliant mistakes of AI in education.
If you have done an experiment or attempted an experiment with AI in education and got stuck, it did not set off, the outcome was not what you had expected, AI did not work as you had intended, you name it, then the AI Vraagbaak would love to hear from you!
For the month of AI, the Vraagbaak AI in Education is collecting examples of the great mistakes of AI in education.
Help us find the best practices of AI in education by sharing your mistakes.
Every example or experiment is welcome and relevant. This survey can be filled in completely anonymously. For inspiration, below are two different examples of experiments with AI in education that did not go as expected:
- A geography teacher used ChatGPT to grade final exams to experiment with how well ChatGPT can do this. He did this after he and the second proofreader had already checked the exams and students had already received their grades. A total of 25 students had agreed to let their exams be used for this experiment. In order for ChatGPT to check the exams, the answers first had to be put into a correct format to input a prompt for ChatGPT. The result was that ChatGPT often forgot to check all 25 students and give feedback to all of them. For instance, the first 12 went well, but the last 13 were forgotten. In addition, ChatGPT often repeated students' answers incorrectly when providing them with feedback.
- Perry et al., 2023 experimented in their research “Do Users Write More Insecure Code with AI Assistants?” They investigated if AI assistants help students program more secure code. It turned out that students who used AI to program code wrote less secure code than those who had not used an AI assistant. However, the students who used AI and were critical about the outcomes of AI and kept modifying their prompts had more secure code than students who did not use an AI assistant.
There are 2 questions in this survey.
