Ethical considerations when conducting machine learning research

The development of machine learning tools has opened new pathways for the field of physics education research. These pathways, however, come with pitfalls that are not always present when using traditional research methods. In our group's experience using machine learning to predict student persistence, we have come across several ethical considerations that have changed the way we approach the work. Machine learning models can, for example, exacerbate any initial demographic bias in their training data. Furthermore, models that rely on data scraping often run into issues around data privacy regarding what information is collected from students. Finally, we found that while machine learning models could predict persistence, the autonomous nature of the model means that implementing those results ethically is not trivial and requires more work than just creating the model. While the field of computer science has explored the ethics of machine learning, there has not yet been an attempt to study how these apply to physics education research. In this talk, we discuss the potential issues with using machine learning and big data to study students, as well as how we feel this research can be done ethically.