Child avatar project: Investigative interview training through VR outperforms traditional methods
Published:
In a new study, researchers from SimulaMet and OsloMet found that using virtual reality in training police and child protective services for conducting investigative interviews with children surpasses traditional methods.
Given the potentially profound consequences, the necessity of effective training in conducting investigative interviews with abused children is vital. With traditional methods, the quality of these interviews often fall short, and the existing training programs are not enough to ensure that professionals are following best practice guidelines.
This is the background for the project that began in 2021 - to develop an artificial child avatar to improve interviews by police and child protective services (CPS) with maltreated children. The result is a training program with a system for simulating interactive environments with alleged abuse victims using a child avatar.
How the system works
The system is trained on a dataset consisting of transcripts from interviews conducted by professionals according to specific guidelines, and consists of simulated dialogues between professional actors portraying five- to seven-year-old children and an interviewer who could be from CPS or law enforcement. The conversations aim to replicate an investigative interview with a child who may have been a victim of abuse. The dataset of mock interviews was provided by the project’s partners at Griffith University in Australia.
The system consists of several components. First, a language component is designed to analyse a child's conversation and produce coherent, realistic, and suitable responses to an interviewer's questions. Next, there is a speech component for speech-to-text and text-to-speech synthesis. Finally, the front-end system offers four different interactive environments: a VR environment, a 2D environment, an audio environment, and a text-based environment.
A user study was conducted to assess the efficacy of four interactive environments. Professionals from Child Protective Services (CPS) and students of child welfare participated in the study. Results show that VR outperforms in four out of five quality aspects, with 66% of participants preferring it for immersive and realistic training.
“Results from the user study showed that the VR-based system demonstrated significantly greater engagement and the highest average values across all learning experience metrics. Indicating the project's potential to improve training quality by providing a more immersive, lifelike environment that enhances knowledge transfer and the overall learning experience," says PhD student Syed Zohaib Hassan, first author of the article providing these results.
Optimizing question quality
The quality of the questions posed during interviews with children who have been victims of abuse is crucial. The best-practice recommendations emphasize open-ended questions, active listening, and providing support.
“We have also developed a feedback system that accurately classifies questions based on these best-practices - distinguishing between recommended and not recommended questions, and enhancing learning effects by offering effective feedback during training," says Hassan.
Through this system, the researchers achieved a balanced accuracy of 87 percent in classifying these two categories. Surpassing the previous state-of-the-art results of 72 percent.
In tandem with the outcomes from the user study, they believe that these findings will play an important role in the continuous development of better training programs.
“These insights are crucial for the development of effective training programs in the field of child welfare and law enforcement. The study also reveals the potential of VR and advanced technologies in other sorts of professional training for sensitive, stressful and critical tasks.”
Syed Zohaib Hassan, PhD at SimulaMet
What lies ahead?
The project team now intends to continue their studies with more participants to get a clearer picture.
The future work will concentrate on enhancing the child avatar system’s quality aspects, to provide a more flexible and efficient interview training platform that will contribute to effective learning.
“Our future focus revolves around refining the dialogue model to generate more realistic responses by using the advancements in large language models, and incorporate more realistic-looking avatars that will be generated by neural radiance fields or generative adversarial networks," says Hassan.
In addition, emphasis will be placed on enhancing the feedback quality and exploring the optimal ways that feedback can be effectively utilized to enhance the overall learning experience.
The child avatar project
The project titled “Professionals interviewing maltreated children supported via artificial avatars” is drawing on expertise in developmental psychology and artificial intelligence. The project manager is Gunn Astrid Baugerud from OsloMet Faculty of Social Sciences - the facilitators of the project.
SimulaMet is providing expertise on artificial intelligence, with Syed Zohaib Hassan, Pegah Salehi, Steven Hicks, Saeed Shafiee Sabet, Michael Riegler and Pål Halvorsen as the coordinator from SimulaMet. Members from OsloMet provide expertise on developmental psychology, with Miriam Sinkerud Johson and Ragnhild Klingenberg Røed in addition to Baugerud. Other partners in the project are Griffith University in Australia and the University of Cambridge in UK. It is funded by The Research Council of Norway, and will conclude in March 2024.
Reference article:
Associated contacts
Syed Zohaib Hassan
Email Syed Zohaib