Graphic codes across times and cultures consistently share certain visual characteristics. According to the ecological hypothesis, this is because glyphs reflect the input statistics to which our visual system has adapted. We computationally model this hypothesis by employing a drawing-based signaling game involving two AI models to explore factors that impact empirical regularities in the surface form of artificially evolved glyphs and their similarity to human visual signs. In our first experiment, we investigate the role of the models' perception system on glyph line orientation and symmetry. We find that these characteristics are impacted by the input statistics of data used to pre-train models and, to a lesser extent, canvas shape and architectural model properties. Our second experiment analyzes the grapho-phonemic mapping that emerges when we integrate representations learned by a deep learning model trained for speech conversion into our setup.

Although AI has made large strides in recent years, state-of-the-art models still largely lack core components of social cognition which emerge early on in infant development. The Baby Intuitions Benchmark was explicitly designed to compare these "commonsense psychology" abilities in humans and machines. Recurrent neural network-based models previously applied to this dataset have been shown to not capture the desired knowledge. We here apply a different class of deep learning-based model, namely a video transformer, and show that it quantitatively more closely matches infant intuitions. However, qualitative error analyses show that model is prone to exploiting particularities of the training data for its decisions.