
Make it beautiful. Make it mysterious. Don't make it medical.
At a time when biometry is increasingly used for identification and control, this data constituted a new way of representing both anonymity and community.

"Each basic physiological assumption of the polyvagal theory is untenable. The underlying hypotheses have been falsified."—Paul Grossman, Biological Psychology (2023)


"At a time when biometry is increasingly used for identification and control, this data constituted a new way of representing both anonymity and community."—Rafael Lozano-Hemmer

The AI Companion: Technical Implementation
The AI companion represents a significant evolution in biofeedback art—not an external observer analyzing data, but a continuous presence developing an interoceptive understanding of the participant's nervous system through shared experience. This requires specific technical architecture and design principles. The Three-Phase Architecture Calibration Phase Natural voice conversation establishes rapport while simultaneously capturing baseline breath patterns. The AI asks open questions ("How are you arriving today?" "What brought you here?") that feel conversational but provide physiological context. The visual environment begins responding to breath during this phase—calibration IS the beginning of the installation, not separate from it. Duration: 3-5 minutes. Data captured: baseline respiratory rate, breath depth variability, voice characteristics, self-reported state. Immersive Experience The AI remains present but silent, witnessing the participant's journey through continuous breath analysis. It tracks: activation patterns (when did breath become rapid/shallow?), recovery moments (how quickly did regulation return after disruption?), startle responses and recovery trajectories, conscious regulation attempts (did they deliberately slow their breath?), overall pattern (progressive settling, sustained activation, or movement toward shutdown?). This witnessing creates the interoceptive understanding that enables personalized grounding. Ground & Return The AI re-engages with full knowledge of what was witnessed. It leads with state-appropriate intervention based on exit condition: For freeze/dissociation → immediate external grounding ("Keep your eyes open. Tell me one thing you see."). For activation → breath guidance with extended exhale. For regulated exit → celebration of capacity and optional integration. The AI actively guides the grounding process, tracking participant responses and adjusting approach in real-time. A working prototype of the AI Companion demonstrating this continuous presence model is available at Click here Voice Interaction Design In production, the AI companion operates through natural voice conversation, not text chat. This is essential for several reasons: voice carries prosodic information about nervous system state (pitch, pace, breath sounds), voice interaction keeps participants externally engaged (critical for dissociation), voice feels more human and less clinical than text interfaces, and voice allows hands-free interaction when participants may be physically grounded or holding objects. Voice Design Principles Warm, present, non-clinical tone Short responses (30-60 words typical) Natural pacing with pauses Avoids therapeutic jargon Uses "I" statements ("I noticed..." "I was with you...") Asks open questions, not yes/no Reflects without interpreting What the AI Doesn't Do Doesn't diagnose or pathologize Doesn't claim to "know" internal experience Doesn't give medical advice Doesn't replace human staff for crisis Doesn't store or share personal data beyond session Doesn't make claims about "healing" or "curing" Data Architecture and Privacy The AI companion processes breath data and conversation in real-time but does not permanently store personal information. Session data (breath patterns, conversation) is ephemeral—available for immediate post-installation review if the participant wants to see what was detected, then deleted. The system maintains aggregate anonymized data for installation refinement (e.g., "40% of participants showed activation patterns") but no individual identification. Participants are informed of this architecture during consent. The Symbiotic Model This is not surveillance—it's symbiosis. The AI doesn't observe from outside; it develops understanding through shared presence. The participant's breath shapes the visual environment. The AI witnesses the participant's response. Both are changed by the encounter. This reciprocal relationship distinguishes Ascended Intelligence from extractive biometric systems.
