--- title: "Branching scenarios vs AI characters in VR training" description: "Branching VR scenarios and AI roleplay characters compared for training teams: what each gives learners, where each needs care, and how to use both." canonical: https://peopletechrevolution.com/learn/xr/branching-scenarios-and-ai-characters --- [Learn](https://peopletechrevolution.com/learn) / [XR](https://peopletechrevolution.com/learn/xr/what-is-xr) # Branching scenarios and AI characters in VR training Learners want more ways to respond than a script allows. What each approach gives a training team. The request ## Learners want more room to respond In Google’s [CHI 2024 case study](https://doi.org/10.1145/3613905.3637128) of VR inclusion training, participants asked for more agency, more conversational turns and more options for what to do next. Two visitors wait, options between them. A librarian gestures beside an option panel. A practitioner moves a conversation block. [Download the Google research paper PDF](https://peopletechrevolution.com/uploads/docs/people-tech-revolution-google-vr-inclusion-training-chi-2024.pdf) Branching scenarios ## A script your experts sign off Every line a character says is written and reviewed before anyone puts on the headset. An instructor points along a route. ### What a branch gives you Each choice leads to a scene your subject experts approved, so every learner meets reviewed material and the facilitator knows what each path covers in the debrief. [Mater Education](https://peopletechrevolution.com/work/case-studies/mater-speaking-with-good-judgement) staff practised feedback conversations this way, in a branched recreation of real hospital rooms. Two reviewers compare response panels. ### Where a branch needs care The [CHI authors](https://doi.org/10.1145/3613905.3637128) call deeper scenarios the hardest area to improve, because every new way to respond means more branches to write. AI characters ## A partner who answers in its own words A generative AI character replies to whatever the learner says, so the conversation goes where the learner takes it. A colleague across a meeting table. ### What an AI character gives you More turns and more ways in, without writing each branch. PTR runs this as [AI roleplay](https://peopletechrevolution.com/ai/ai-roleplay-practice): a briefing, a live conversation, structured review against set criteria, then the same scenario again. Two designers set markers at a boundary. ### Where an AI character needs care The [CHI authors](https://doi.org/10.1145/3613905.3637128) call AI characters for training unproven, needing strong guardrails so replies meet the learning objective and never turn offensive in a sensitive conversation. [NIST’s Generative AI Profile](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf) names the same risks: confident answers that are false, and harmful content. Two people adjust controls around a model. ### How PTR sets the boundaries Each roleplay is grounded in the organisation’s own policy and practice and runs within configurable guardrails, with human oversight and a clear path back to a person. Choosing ## Match the method to the moment A branch where the words must be exact; an AI partner where practice needs volume. A visitor gestures at a service counter. ### When a branching scenario fits The wording matters, as in a disclosure, a formal notice or a clinical handover. Assessment needs every learner to meet the same situation. The topic is sensitive enough that every reply should be reviewed first. A practitioner raises a hand to begin. ### When an AI character fits Learners need many attempts at an open conversation, such as feedback, a complaint or a first meeting, on a laptop or phone between sessions, with a facilitator reviewing the attempts. Together ## Rehearse the moment, then keep practising A scripted scene in a [VR headset](https://peopletechrevolution.com/xr/virtual-reality) for the moment itself, an AI partner for repeated practice, and a facilitator for the [debrief](https://peopletechrevolution.com/training/vr-training-and-roleplay). Two people practise a service conversation. A support worker rehearses a coaching talk. A colleague across the table, arms folded. Questions ## What training leads ask first Harm, assessment and equipment, settled before a method is chosen. Two colleagues review blank tokens. ### Can an AI character say something harmful? It can. Guardrails lower the risk without removing it, so PTR grounds each character in approved material and keeps a person reviewing the practice. A wheelchair user rehearses with a colleague. ### Which suits assessment better? A branching scenario: its paths are written in advance, so each attempt can be assessed against criteria agreed before training starts. A service moment rehearsed at a cafe table. ### Do learners need a headset? [AI roleplay](https://peopletechrevolution.com/learn/ai/ai-training-roleplay) runs on a laptop or phone; immersive VR uses a headset. See [choosing the right technology](https://peopletechrevolution.com/learn/choosing-the-right-technology). - [AI roleplay training](https://peopletechrevolution.com/ai/ai-roleplay-practice) - [Immersive training and roleplay](https://peopletechrevolution.com/training/vr-training-and-roleplay) - [The Google CHI 2024 study](https://peopletechrevolution.com/work/research-evidence/vr-dei-training-google-chi-2024) - [What is virtual reality?](https://peopletechrevolution.com/learn/xr/what-is-virtual-reality) Choosing how your team will practise? [Open the contact form](https://peopletechrevolution.com/learn/xr/branching-scenarios-and-ai-characters#contact) Ask our CEO Simon Lowe’s Digital Twin a question ## Sources The paper and the risk framework, on their own sites. ### [Leveraging Virtual Reality to Enhance Diversity and Inclusion Training at Google](https://doi.org/10.1145/3613905.3637128) Bursztein, Brown, Sanderson and Kelley, CHI EA ’24 Scenario depth, and the proposed use of AI characters. ### [Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf) NIST AI 600-1, 2024 The risks a generative character must be guarded against.