Simulating Social Situations in Immersive Virtual Reality - A Study of Bystander Responses to Violent Emergencies

The goal of this research is to show how immersive virtual reality (IVR) can be used to study human responses to extreme emergencies in social situations. Participants interact realistically with animated virtual humans. We show this through experimental studies of bystander responses to a violent confrontation, and find that there are conditions under which people intervene to help virtual characters that are threatened. We go on to show that a reinforcement learning (RL) method can capture the types of actions of virtual humans that lead to greater intervention on the part of the bystander. It has been shown that people tend to respond realistically in social situations depicted in IVR when they have the illusion of ’being there’ (Place Illusion, PI) and that what they perceive appears to be really happening (Plausibility Illusion, Psi). This has enabled IVR technology for the study of several fields including human behavioural studies, social phobia treatment and both physical and psychiatric rehabilitation. Additionally, IVR helps to overcome ethical issues such as deception that can arise from the nature of the study. The highly controlled environment reduces the variations induced by the repetition of the study, and thus, increases the internal validity. Furthermore, IVR allows setting up life-size computer-generated simulations and enables the possibility of interaction with natural body movements, making their responses close to being authentic thus increasing the ecological validity. We carried out a series of experiments to understand the circumstances likely to make people intervene when they witness a confrontation between two people. Faced with a potentially violent situation, any individual has to decide whether to intervene to try to prevent the violence, or do nothing. Evidence demonstrates that factors such as a shared social identity between the bystander and either the victim or the aggressor, the presence of other bystanders or authority figures and their behaviour influence people’s responses to an emergency. But for ethical and validity reasons, it is very difficult to set up studies with real actors which observe how people react to violence. The results of the experiments in this thesis show that the likelihood of a bystander intervention can be increased or diminished if the bystander perceives the other people present in the scene as sharing some type of affiliation with him. The main experiment variable that we manipulated was whether the people in the scene supported the same football team as the bystander, or on the contrary they did not have an explicit association with any team in particular. Additionally, this thesis provides evidence that the bystander effect also occurs in an immersive virtual environment (IVE) and that the strength of this phenomenon varies depending on the social identity of the characters present in the scene. The last part of the thesis shows that RL can provide learning capabilities to a computer to study human behaviour and use the knowledge towards eliciting pre-determined responses from real people in IVEs, such as increasing the likelihood of intervention in a violent emergency