An artificial neural network can be trained to provide effective feedback on trainee teachers’ diagnostic reasoning, according to Michail Sailer and colleagues in Germany and the UK.
One key skill teachers learn during training is to diagnose what kind of difficulty a student is having and then clearly explain and justify their diagnosis.
One accessible way of practicing these skills is to use simulation based cases. In this study, the simulations were presented using the computer platform CASUS.
The simulations provided materials such as transcripts of teacher-parent conversations, assignments, and descriptions of student behaviours. Appropriate diagnoses ranged from specific learning difficulties, such as dyslexia, to disorders such as attention-deficit (ADD/ADHD).
In order for the trainees to learn from the practice cases, they need appropriate feedback on both their diagnoses and justifications. Since this is potentially very time-consuming for teacher trainers, it is common to have ‘static’ feedback in the form of preprepared expert diagnoses for comparison.
To provide ‘adaptive’ feedback to individual trainee responses, Sailer et al made use of artificial intelligence; in particular, natural language processing (NLP) which can analyse and respond to human language, thus providing real time feedback. The NLP-based algorithm was trained on relevant diagnostic contexts making use of data from previous trainees.
For the study, 178 pre-service teachers were recruited. As well as comparing ‘static’ vs ‘adaptive’ feedback, Sailer et al wanted to assess the accuracy of diagnoses made collaboratively vs individually. Trainees were put into one of four groups: working collaboratively with 1) static or 2) adaptive feedback; working individually with 3) static or 4) adaptive feedback.
Analysis of trainee responses showed that the use of adaptive feedback improved the quality of the trainees’ justifications – whether working individually or collaboratively – but not the accuracy of their diagnoses.
Students working collaboratively actually tended to have poorer diagnostic accuracy, which was somewhat ameliorated by adaptive feedback, especially for more complex cases; a curious and unexpected outcome worth further study.
Overall, the use of adaptive feedback of this kind shows potential for providing immediate, personalised training. The initial setting up of the algorithm is still quite time-consuming, however, in this case, 40 expert feedback paragraphs were supplied for each case. For large teacher training programmes, however, this initial expert-time investment could pay off in expanded training opportunities with limited further investment. At least we do still need human experts – for now.
REFERENCE
- Sailer, M., Bauer, E., Hofman, R., Kiesewetter, Glas, J., Gurevych, I. and Fischer, F. (2023) Adaptive feedback from artificial neural networks facilitates pre-service teachers’ diagnostic reasoning in simulationbased learning, Learning and Instruction. https://doi.org/10.1016/j.learninstruc.2022.101620
