Your VR Assistant May Soon Admit When It's Making Things Up
New research shows how VR assistants can visibly signal when they're inventing facts.
Researchers compared three ways to present hallucination-awareness information—uncertainty and provenance cues—from speech-based embodied conversational agents in virtual reality, against a no-cue baseline: embodied cues using gestures and posture, icon cues using visual indicators, and text cues using color-coded text with inline citations. In a within-subjects study with 24 participants, all three designs supported users in identifying hallucinations. Embodied cues were associated with higher trust and immersion, text cues offered clearer interpretability, and icon cues preserved relatively good interpretability while causing less disruption to immersion compared with embodied cues and text cues.
- Three cue styles were tested — body language, warning icons and colour-coded text — and all three helped people catch AI mistakes.
- Gesture-based cues made the virtual assistant feel most trustworthy and immersive, while icons were the least distracting.
- Only 24 volunteers in headsets took part, so real-world results with millions of users could look different.
Why It Matters
If virtual assistants visibly show doubt, you'll be less likely to act on confident-sounding nonsense.