This study tested the capability of artificial social intelligence to make inferences and predictions concerning the state and behaviors of 3-person teams executing an urban search and rescue task in Minecraft.

The ASIST Study-2 dataset was developed in a human subjects research study designed to assess the capability of artificial intelligence to infer the state and predict the actions of members of a three-person team executing an urban search and rescue task in Minecraft. The data were developed under Contract No. HR001119C0130 to the Defense Advanced Research Projects Agency (DARPA). The dataset comprises approximately 2,100 files and 300GB of data. We have partitioned the full dataset into folders that support research in specific areas. Thus, researchers can more easily download only the files of value to them.