VR Surveillance as a Catalyst for effective Bias Mitigation in Policing
PhD-project by Bas Boing (Chief Inspector with the Dutch police and PhD Researcher at Psychology of Conflict, Risk and Safety)
Police departments across the world increasingly face allegations of ethnic profiling.1 This practice has considerable implications for individuals, communities and law enforcement agencies, as it has been associated with deteriorating community trust, which in turn make policing more challenging. For these and other reasons, police chiefs want their agencies to invest in programs to increase public trust by professionalizing police stops.
However, attempts to improve officers’ knowledge and skills so as to remedy potential biases toward certain demographic groups have proven to be challenging and largely ineffective. Part of the problem is that ethnic profiling is a topic that seems to be of little interest to many police officers. Research shows that law enforcement officers cannot see themselves having prejudices. Many perceive claims of ethnic profiling as false and heavily exaggerated. In fact, scholars observed some police officers to perceive these claims as an attack on their craftsmanship and integrity. For these reasons, police officers do not want, and/or do not see any urgency, to engage in training.
At the same time experts have warned that the attention for ethnic profiling, and in particular the ‘racist cop stereotype’, can make police officers feel anxious about appearing racist, which allegedly makes them avoiding interactions with community members with a minority background. In the literature, this phenomenon is known as racist stereotype reactance. When police officers cease doing stop and searches, it can lead to depolicing and deteriorating crime levels in neighborhoods, some experts fear. Police officers have also mentioned this as yet another reason why they are against the attention for ethnic profiling, as it scares good police officers to do stop and searches.
It is thus not only important to make police officers aware of their unconscious bias, but also to make them more resilient and confident, so they to continue to do police stops.
1 The definition used by the Council of Europe and adopted by most police departments in Europe is: ‘The use by the police, with no objective and reasonable justification, of grounds such as race, color, language, religion, nationality or national or ethnic origin in control, surveillance or investigation activities.
What is needed
The best way for people to become aware of their bias is through direct learning experiences. However, direct learning is difficult to facilitate. Role playing is time consuming, actors are usually expensive, and simulation outcomes are hard to compare, as circumstances leading up to these outcomes may differ from each other. Interactive Branching Scenario (IBS) in virtual reality (VR) offers a possibility. In VR, individuals can enjoy highly realistic scenarios in a controlled environment, while IBS offers the possibility to explore how different choices lead to different (more or less effective) outcomes. This makes IBS in VR an interesting choice for training to improve self- and group-efficacy.
Another advantage is that VR can be used for research purposes as well, for instance to monitor behavior. VR has been used, for example, to study consumer choices of grocery shoppers, but also how flight personnel know how to carry out safety instructions in case of an emergency. Potentially it is possible to observe who police officers select for police stops, and why. As such, it should be possible to discover to what extent they engage in ethnic profiling, or if they experience racist stereotype reactance and do exactly the opposite. Hence, having a system in place that can be used both for training and research purposes, can help police agencies to get the right intelligence, while also improving officers’ awareness and skills.
Using the VR for training
Given these advantages, the Dutch police has developed an interactive VR simulation, that can be used for both purposes. For the first Europol Innovation Award ’21, we have submitted only the project related to training. We reiterate the aim of the training here. Training consists of two parts.
- The first part of the training consists of a simulation in VR. The simulation puts the participants in a 360-degree environment in the shoes of a police officer. In the simulation the participants see a variety of people, but can initiate an interaction with only some of them (i.e., only the actors, not regular pedestrians). Once they have selected someone, participants are presented with several questions on how to proceed. Each question (story branch) contains two or three options to choose from; options are always both escalating (e.g., identification check, search, fine, arrest), and de-escalating (e.g., cease the interaction).
- The second part of the training consists of group dialogue. Directly after completing the VR simulation, participants are encouraged to discuss their choices in group conversations. Discussion leaders (all in-service trainers) first identify officers who have all made identical choices and then allow one party to present their case. After these participants have made their case, participants who have made different choices are invited to speak. When hearing these different experiences, participants belonging to the first group may feel challenged and may become uncertain about their own choices. This may lead them to reevaluate an initial point of view, causing them to engage in higher-level cognitive reasoning, and, ultimately, stimulate awareness of their own potential biases in day-to-day police activities.
Picture 1: classroom with police officers doing VR-training
Using the VR for research
Studying ethnic profiling
To determine to what extent police officers use race or ethnicity in stop and search decisions, we did an experiment with deepfake technology. For this study, we exposed 96 participants to a scenario, where the participants found themselves patrolling a square in the financial district in Amsterdam. In total there were four different characters in the simulation that could be selected for a conversation / ID-check. One of these individuals was a man sitting on a motorcycle, who seems a bit restless as if he is waiting for someone. In the original version, the man has a Dutch appearance (see picture 3), while in the parallel one we have used deepfake technology to give this man a non-western appearance (see picture 2). Half of the participants were exposed to the scenario with the deepfake, while the other half stayed with the original one. Prior to the experiment, participants did not know to what condition they are assigned to, nor did they know what the study was about. In our research we wanted to discover who of these two characters is more likely to be selected for a conversation with the police, and subsequently an ID-check. We used retrospective-think-aloud-protocol (in the form of open and closed survey questions) to capture the participants’ motivation for selecting (or not) any of the individuals, and to make sure that their decisions was not somehow mediated by the technology.
Picture 2: Deepfake Picture 3: Non-deepfake
Studying risk-avoidance and depolicing
Same technology and methodological framework were used for a study to assess racist stereotype reactance and depolicing. To be able to study this behavior in the VR-simulation, participants were asked to watch a short film prior to the experiment. Half of the participants saw a film in which the starring police officers were accused of racism, while the other half saw a random introduction video. It was expected that participants who saw the manipulation would recognize these accusations from their daily practice, which would then activate the racist stereotype. We expected that these officers, as a result, would be more likely to avoid the deepfake (man with non-western appearance). This would point towards risk-avoidance and depolicing. Again, we used open and closed survey questions in addition to capture the participants’ previous experiences with racism allegations, which may influence their decisions.
Relevance
The studies help us to understand the role of race or ethnicity in stop and search decisions, and also how the perception of negative stereotypes against the police can mediate that behavior. When the avoidance of ethnic minorities in traffic stops is mediated through higher awareness about racial bias and its impact, this can be seen as a positive thing. However, when this avoidance is mediated through fear and uncertainty, this could lead to depolicing and ultimately result in unsafe neighborhoods. It is important that police departments are keen on both issues and keep the balance in their policies to mitigate ethnic profiling.
Links and showcase videos
- To get an impression about how the Dutch police uses the VR simulation for training purposes, take a look at this showcase video: https://vimeo.com/586728895/1a433e6949
- To get an impression about how the Dutch police uses the VR simulation with deepfake technology for research purposes, take a look at this showcase video: https://vimeo.com/801931504/a323610694