Machine Learning Models and Their Development Process as Learning Affordances for Humans

This paper explores the relationship between unsupervised machine learning models, and the mental models of those who develop or use them. In particular, we consider unsupervised models, as well as the ‘organisational co-learning process’ that creates them, as learning affordances. The co-learning process involves inputs originating both from the human participants’ shared semantics, as well as from the data. By combining these, the process as well as the resulting computational models afford a newly shaped mental model, which is potentially more resistant to the biases of human mental models. We illustrate this organisational co-learning process with a case study involving unsupervised modelling via commonly used methods such as dimension reduction and clustering. Our case study describes how a trading and training company engaged in the co-learning process, and how its mental models of trading behavior were shaped (and afforded) by the resulting unsupervised machine learning model. The paper argues that this kind of co-learning process can play a significant role in human learning, by shaping and safeguarding participants’ mental models, precisely because the models are unsupervised, and thus potentially lead to learning from unexpected or inexplicit patterns.