Examining Self-directed Learning of Natural Science Categories Through Adaptive Learning Systems

Empirical studies of category learning traditionally utilize passive paradigms, where experimenters strictly control the information presented to learners. Recently, there has been growing interest in active learning paradigms, where learners are allowed to select category examples themselves. However, existing studies highlighting the benefits of active selection often employ artificial categories explainable by simple rules, potentially limiting generalizability of the findings to more complex natural categories.The project aims to examine the benefits and limitations of active selection in category learning in the context of teaching geological categories. It was found that active selection of training examples did not lead to better category transfer in learning rock categories compared to traditional "passive" learning. By implementing an adaptive category-sampling system that prioritizes the categories that are less well-learnt by individual learners, I found that category learning performance is enhanced by studying more examples from the worse performing categories, which is a learning strategy endorsed by successful active learners. On the other hand, learners tend not to select examples from the difficult categories to the same extent as the adaptive system suggested, with some learners showing reluctance to follow explicit recommendations based on the system. Additionally, presenting all examples from contrastive pairs of categories simultaneously, regardless of whether the categories were self-selected by the learner or determined by the system, also substantially improved category learning. In conclusion, incorporating adaptive example/category sampling and contrastive category pairing can effectively alleviate biases in self-directed learning and inform better instructional design for natural science education.