Context: Communication and mutual trust are keys driver for effective teamwork in human teams. In human-AI teams, teams composed of both humans and artificial agents, communication and trust are also important. In this research project, we investigated how different artificial agent’s communication affect human’s trust and satisfaction, in such teams. Participants teamed up with artificial agents in an online setting (using 2D grid world) and their decisions were be logged. This dataset includes different metrics calculated based on the logs, self-reported questionnaire answers on trust and satisfaction, and free answers to open questions.
This dataset was created during the Research Project course of the Computer Science Bachelor's in Delft University of Technology supervised by Carolina Jorge and Dr. Myrthe Tielman. Five students ran a user study with six different conditions (the baseline, and five new developed by each of them). The full description of the user study and their individual results (i.e., pairwise comparison between their own condition and baseline) can be found in each of their thesis, linked in this page below.
Then, a full joint dataset was created and it can be found in "Full dataset.csv" (total 140 rows). To balance the number of participants per condition, we generated a "capped_dataset.csv" with 20 rows per condition (total N=120). We analysed differences among conditions, and rerun the pairwise comparisons, of "capped_dataset.csv". The code can be found in "Quantitative Analysis.ipynb". These results are to be published in a paper - the author contributions can be found in "author_contribution.txt".
The full code used for the generation of this dataset can be found in this Github repository: https://github.com/centeio/AT-Communication