Abstract
Values or principles are key elements of human society that influence peopleto behave and function according to an accepted standard set of social rules tomaintain social order. As AI systems are becoming ubiquitous in human society,it is a major concern that they could violate these norms or values andpotentially cause harm. Thus, to prevent intentional or unintentional harm, AIsystems are expected to take actions that align with these principles. Trainingsystems to exhibit this type of behavior is difficult and often requires aspecialized dataset. This work presents a multi-modal dataset illustratingnormative and non-normative behavior in real-life situations described throughnatural language and artistic images. This training set contains curated setsof images that are designed to teach young children about social principles. Weargue that this is an ideal dataset to use for training socially normativeagents given this fact.