Encouraging bystander helping behaviour in a violent incident: a virtual reality study using reinforcement learning.

Area of Science:

• Psychology
• Computer Science
• Human-Computer Interaction

Background:

• Virtual reality (VR) offers controlled environments for studying complex social behaviors.
• Traditional VR studies have fixed scenarios and delayed data analysis.
• Adaptive VR environments that respond to participant behavior are needed.

Purpose of the Study:

• To investigate the use of reinforcement learning (RL) to encourage helping behavior in a virtual reality (VR) aggression scenario.
• To develop a method for real-time data processing and environmental adaptation within VR.

Main Methods:

• A VR scenario was created where a virtual character was attacked, with participants acting as bystanders.
• Reinforcement learning (RL) was implemented to varying degrees to control virtual character actions.
• Forty-five participants were divided into three groups: no RL, medium RL, and full RL.

Main Results:

• A greater extent of RL operation led to a higher number of helping interventions by participants.
• The adaptive VR methodology allowed for real-time data collection and processing.

Conclusions:

• Reinforcement learning (RL) can effectively encourage specific participant behaviors, such as helping, in adaptive VR environments.
• This adaptive VR approach offers an alternative to traditional experimental designs and can guide participant actions.