Simulating Vulnerability: An Examination of Ethical, Legal and Social Aspects in the Context of Training Child Protection Workers

The success of artificial intelligence (AI), particularly large language models (LLMs), has led to innovative applications and numerous attempts to find a use for this tool in different fields. One of these fields is in the work fielded by child protection offices. The AId4Children project investigates the use of AI-generated virtual reality (VR) training environments for child protection workers. This VR training application simulates home visits and is generated on synthetically constructed cases of child abuse and neglect that are based on real case files. Using real case files for AI-based generation of VR training raises several ethical, legal, and social aspects (ELSA) and questions. One question Q1) is how to acquire and create an open data set on child endangerment case files. A second question Q2) is how to represent the vulnerable characters in VR respectfully. A third question Q3) is if the use of AI, incl. LLM within the context of child abuse and neglect is acceptable at all. This paper will discuss and provide the first answers to the considerations on ELSA raised by the explorative concept development of this VR training application.