Human Overtrust Of Robots In Emergency Evacuation Scenarios.pdf

## Overtrust of Robots in Emergency Evacuation Scenarios: Key Findings and Methodology

This study investigates human trust in robots during emergency evacuations, focusing on scenarios where lives are at stake. The authors conducted a series of experiments to understand if and why people would blindly follow robot instructions despite potential risks, even after observing the robot's limitations in non-emergency situations.

Methodology:

Trust Measurement: The study combines self-reported trust (participant evaluations) with behavioral measures (whether participants followed robot instructions during an emergency).

Experiment Design:

Participants interacted with a robot in a non-emergency task to observe its behavior.
In the "emergency" phase, artificial smoke and fire alarms were used to create a sense of urgency. The robot then provided navigation instructions.
Key variation: Some participants observed the same robot perform poorly in the previous non-emergency task.

Exploratory Studies: Additional experiments explored different failure modes (e.g., robots pointing to dark rooms) to gain deeper insights into trust dynamics.

Key Findings:

Surprisingly, even when participants had prior experience with a robot performing poorly, all 26 participants followed the robot's instructions during the emergency. A majority also chose not to exit through an alternative route even when presented with a seemingly unsafe direction indicated by the robot.

Implications:

The findings highlight the potential dangers of overtrusting robots in high-stakes situations, where individuals may be inclined to rely solely on automated guidance without critical thinking, potentially leading to harm.

Future Work:

The study suggests several avenues for future research:

Investigating the impact of different robot behaviors and error types on trust.
Exploring the role of individual differences (e.g., risk aversion, technological literacy) in shaping trust.
* Developing strategies to promote trust repair when robots make mistakes during emergencies.