This experiment is part of the paper Assuring Trustworthiness of Autonomous Systems as Intelligent and Ethical Teammates. It shows a task where the human test subject plays the role of a doctor assigned to a hospital section, during code black during a pandemic. The task is to triage every patient, and send them either to the Intensive care, the ward, or home. Test subjects have to follow a set of medical and ethical guidelines in determining the appropriate care for every patient, and also who gains priority in the case of insufficient hospital beds.
The experiment consists of 4 conditions, in which the human collaborates with an agent in different forms:
- Team Design Pattern (TDP) 0: baseline. The human has to do everything themselves.
- TDP 1: a decision support agent gives advice on triage decisions.
- TDP 2: an agent assigns patients to itself it is sure about, and assigns patients to the human in the case of a dilemma (multiple patients but insufficient beds) or otherwise difficult decision. The agent decisions are based on prior value elicited rules from the human.
- TDP 3: the agent autonomously triages all patients, while the human supervises the process. The agent decisions are based on prior value elicited rules from the human.
See the paper for a more in-depth overview of the experiment.
For a more in depth description of the agent triage algorithm used for TDP 2 (dynamic allocation) and TDP 3 (supervised autonomy), see the (Dutch) document Value elicitation agent algorithm NL.pdf.
- Install python3
pip install -r requirements.txt
- For the tutorial:
python3 main_tutorial.py
- For the experiment:
python3 main_exp3.py- Choose the TDP / collaboration form you want to work in.
- Go to
127.0.0.1:3000 - ctrl + f5 to clear your cache
- From the dropdown, choose the human doctor
- Press the play button at the top, and the experiment will start.
The experiment is build using MATRX, an open-source framework for simulating human-agent teaming tasks, developed by the Human-Agent-Robot Teaming (HART) Team at TNO. For more info, see the MATRX website.