A paper based on LAW-GAME, presented by Helvia at the Edge Intelligence 2022 Conference
Helvia presented the paper “AI-assisted Serious Games: Interrogating an Avatar in Virtual Reality” at the Edge Intelligence 2022 – Emerging Tech Conference, jointly organized by HETiA and the University of Patras.
The paper is based on the LAW-GAME project, which develops a social AI-powered Virtual Reality game platform to cover the training needs of Law Enforcement Agencies, providing a powerful alternative to traditional classroom-based learning.
Focusing on Police Interrogation, one of the platform’s gaming modes, the paper explores the implications of implementing a realistic interactive dialogue between a human player (the police officer) and an AI-assisted non-player character (the suspect).
About the LAW-GAME Project
LAW-GAME aims to train LEAs, focusing on the transition between theory and real-life practice through serious games in a safe and controlled virtual environment. To achieve this, the project develops an intuitive, configurable, user-centered, social VR game engine, enhanced with AI-assisted procedures and emotion recognition features.
The project has 4 gaming modes:
- The CSI Game: Examining the crime scene and all kinds of evidence in order to solve the case
- The Police Interview Game: Practicing interrogation and negotiation skills versus an AI-assisted suspect / perpetrator
- The Terrorist Attack Game: Attempting to predict terrorists’ actions and terminate a critical situation
- The Car Accident Game: Collecting scene’s data and preparing a realistic accident report
The Police Interview – Interrogation Game
The aim of this game is to question the suspect in order to validate the gathered evidence and get a confession of the crime. The basic elements of the game are the following:
- The place: a police interrogation room
- The characters: (a) the player – a police officer who interrogates the suspect, (b) another police officer who observes as per formal police procedure, i.e., a non-player character (NPC); and (c) the suspect, i.e., an NPC controlled by the game’s AI engine.
- The storyline: a description of the incident and the reasons for the interrogation
- The evidence: each game has a set of evidence which were gathered by the police investigation team prior to the interrogation
The requirements of the project for a realistic, interactive dialog come with a number of challenges, i.e. the game engine should be able to:
- Understand what the player / police officer asks and respond accordingly
- Differentiate the responses based on the context of the dialogue, i.e. depending on the player’s previous questions
- Recognise and respond accordingly to different police officer’s emotions
The proposed solution for interrogating an AI-assisted 3D avatar includes 4 mechanisms:
- Stage progression: We divide the game into 3 distinct stages – Question → Confrontation → Confession.
- Questions’ classification: We classify relevant questions into evidence categories as questions within the same category can contextually influence responses of the same category and gradually progress the storyline forward
- Dialogue management: We combine 2 AI approaches – transformers-based language models to classify each message into the relevant questions and an action-driven framework of preconditions and postconditions to manage the unfolding of the storyline.
- Emotion loading: We regard emotion as an additional dimension that affects the response formulation (e.g. expressing anger) but not the response context.
Part of this work has been carried out in the scope of the LAW-GAME Project, which has received funding from the European Commission under the H2020 programme (Grant Agreement No. 101021714).
The authors acknowledge support and contributions from all partners of the project.