Human learning and machine learning: Building bridges or integration?

Abstract

At the core of Empirical Modelling is an activity we call ‘making construals’. A construal is a software artefact that embodies how we think about something, or make sense of something. For example, it might be a visualisation of a car engine with gears and controls that behaves — through interaction — like the physical car. We shall show a construal of MENACE : an early example of a simple machine (made with matchboxes) that learns to improve its own performance at playing noughts and crosses. Some experts in machine learning contrast the ‘big data’ methods of training networks with the use of explanatory models. It is proving difficult, but desirable, to integrate these approaches. We'll suggest why Empirical Modelling might offer some useful insights into this problem.