Artificial Social Intelligence for Successful Teams (ASIST) Study 2

<p>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.
</p>
<p>
A readme file in each folder (e.g., readme_audio.txt) describes the folder's contents in detail.
</p>
<p>
(1) Data in the studywide folder will be of interest to researchers who conduct any analysis with any data from ASIST Study-2, because these files contain data that describe the study overall, the data used to evaluate AI, or the coding of data.
<br>
<br>
(2) Data in the surveys folder will be of interest to researchers who study individual differences and their effects on behavior.
<br>
<br>
(3) Data in the testbedmessages folder will be of interest to researchers who study individual and team behavior or who use any other components of this dataset, because these are machine- and human-readable text (json) records of the state and behaviors of study participants, and of the state of the task environment.
<br>
<br>
(4) Data in the transcriptions folder will be of interest to researchers who study language use. The audio source of these imperfect machine transcriptions can be found in study video files and audio files.
<br>
<br>
(5) Data in the audio folder will be of interest to researchers who study language use, or who wish to validate, contextualize, or specify transcriptions, testbed messages, and certain survey data.
<br>
<br>
(6) Data in the video folder will be of interest to researchers who study machine vision, or who wish to validate, contextualize, or specify transcriptions, testbed messages, and certain survey data.
</p>