Active physics learning data 2016 - 2016, UCL, NYU and MIT

In these experiments we bring together research on active learning and intuitive physics to explore how people actively about physical properties in "microworlds" with continuous spatiotemporal dynamics. In two experiments, participants interacted with objects in simulated two-dimensional microworlds governed by a real-time physics engine, with the goal of identifying latent physical properties of the objects in the scenes, such as their masses, and forces of attraction or repulsion. We find an advantage for active learners over passive and yoked controls, and show that active learners generate evidence specific to whatever physical property it is their goal to identify. Consequently, yoked learners do better when asked to identify the same property. Our active participants spontaneously performed various "natural experiments" which revealed the objects' properties with varying success. In our research papers we highlight, and begin to and formalize these experiments, and finally outline further steps to categorize and explore active learning in the wild.