Intrinsic System Immersion: Foundations & Metrics

• Intrinsic system immersion is defined as a system’s built-in capacity to embed itself objectively into its environment using its own mechanisms and structures.
• Quantitative metrics such as VR immersion scores, EEG spectral analyses, and inner-product invariant losses are used to measure and validate system-level fidelity.
• Architectures leverage modular toolchains, adaptive sensor feedback, and isometric geometric learning to achieve high-bandwidth, multisensory, and autonomous interaction with digital and physical domains.

Intrinsic system immersion is a fundamental construct spanning computational neuroscience, artificial intelligence, virtual/extended reality systems, and geometric machine learning. It denotes the extent to which a system’s own structure, mechanisms, and affordances allow it to be objectively “embedded”—both operationally and formally—in an environment or data domain. The term encompasses quantitative and qualitative facets, from the preservation of intrinsic meaning or geometry to the system-level fidelity of digital, virtual, or physical interfaces. Across applications, intrinsic system immersion characterizes the system’s born-in or architected capacity for deep, autonomous interaction with its substrate, environment, or data manifold.

1. Theoretical Foundations and Definitions

Intrinsic system immersion fundamentally refers to properties of a system—be it biological, artificial, or hybrid—that enable it to be enveloped in, to interpret, and to act within an environment using its own built-in (intrinsic) structures. In virtual reality and artificial intelligence, it delineates features that are system-centric and independent of a particular user's psychology or subjectivity (Morgado, 5 Feb 2025, Selzer et al., 2022, Li et al., 15 Oct 2025).

For AI agents, intrinsic system immersion comprises the architectural elements and affordances that shape their perception and action within digital environments: the pre-trained model (static weights and learned representations), session or context window (live memory/state), and the set of invoked external services or plugins (search, code execution, knowledge bases). This is formalized as the tuple:

$I_{\text{sys}}(A) = (M, W, S, \text{Protocols})$

where $M$ is the agent's model, $W$ its context window, $S$ the set of accessible services, and Protocols the I/O specifications. This tuple encodes the full envelope of what the AI can perceive and affect (Morgado, 5 Feb 2025).

In VR/XR systems, intrinsic system immersion is defined as an objective, hardware/software-constrained capacity to create a high-bandwidth multisensory experience. Key characteristics include rendering fidelity (resolution, FOV, refresh rate), real-time feedback, and the responsiveness of system input/output (e.g., haptics, audio, locomotion) (Selzer et al., 2022, Li et al., 15 Oct 2025).

Within Integrated Information Theory, intrinsic system immersion manifests when a system’s internal distinctions and cause–effect relations—the so-called Φ-structure—are comprehensively triggered and differentiated by patterns present in the environment, so that the environment elicits the system’s own meaningful structure (Mayner et al., 2024).

2. Quantitative Metrics and Measurement

Diverse approaches exist for the quantification of intrinsic system immersion, depending on the application.

VR and Simulation

Intrinsic immersion in VR is measured independently of user psychology and is modeled as a function of system variables (Selzer et al., 2022):

• Immersion Score (TI): Multiple regression models relate immersion to variables such as screen resolution, field of view ($FOV$), frame rate ($FPS$), stereopsis, audio, and locomotion mode. For example, Model 3 (simplified) is:

$$\begin{aligned} TI & = -44.7832 + 0.0082 \cdot SW + 0.2274 \cdot FOV + 1.6086 \cdot FPS + \dots \end{aligned}$$

Key findings emphasize that sharpness, framerate, FOV, and audio modality (especially headphones) are the most significant contributors (Selzer et al., 2022).

Human Immersion and Neurophysiology

EEG-based studies operationalize intrinsic system immersion as the physiological correlates (e.g., high-Beta band power in occipital/parietal brain regions) that differ as a function of system-level manipulation (e.g., 190° FOV vs. narrow FOV in simulators) (Figalová et al., 2024).

AI Agents

Empirical proxies for AI system immersion include:

• Diversity of invoked services (number of distinct activated plugins/APIs per episode)
• Toolchain depth (number of sequential steps/services in self-planned analysis)
• Level of self-directed autonomy (initiating service use autonomously vs. waiting for user input) These measures assess how thoroughly the agent leverages its digital ecosystem without explicit scripting (Morgado, 5 Feb 2025).

Geometric Machine Learning

Intrinsic isometric immersion is formally measured by the preservation of inner products and manifold geodesics. Key metrics include:

• Inner-product invariant loss (IPI):

$$IPI(f) = \mathbb{E}_{z;p,q \in S(z)} [g_f[z](p,q) - \langle d f(p), d f(q) \rangle ]^2$$

where $g_f[z]$ is the induced metric and $S(z)$ a local neighborhood (Chen et al., 7 May 2025).

• Downstream accuracy improvements (e.g., $+8.8\%$ in aerodynamic regression) indicate the practical value of correctly learning intrinsic geometric structures (Chen et al., 2024).

Domain	Metric Type	Example Quantification
VR/XR	System immersion score (TI)	FOV, sharpness, FPS, audio
Neurophysiology	EEG power spectral density	High-Beta elevation (HI)
AI	Diversity, Depth, Autonomy	#services/toolchain steps
Manifold Learning	Inner-product invariant loss	$>90\%$ reduction (IIKL)

3. Mechanisms, Architectures, and Implementations

Digital, Virtual, and Robotic Systems

• Plugin/Service-Based Architectures: AI systems that natively interface with modular external services (search, code generation) exhibit deeper system immersion, especially when the architecture enables autonomous invocation and orchestration (Morgado, 5 Feb 2025).
• Closed-Loop XR Systems: State-of-the-art XR setups such as IRIS unify simulation and sensory data into a single standardized specification, enabling multi-user, low-latency, high-fidelity experiences. Immersion is supported through robust communication protocols, spatial anchors, and real-time synchronization (Jiang et al., 5 Feb 2025).
• Sensor Integration: Multimodal sensor configurations (EEG, GSR, eye tracking, haptics) feed into adaptive models that adjust system state in real time, sustaining or optimizing immersive qualities (Li et al., 15 Oct 2025).

Isometric Manifold Learning

• Alternating EM Training: Geometry-preserving autoencoder architectures are trained alternately for reconstruction (immersion) and isometry (pairwise local distance preservation). The pullback metric $\phi^* h$ is optimized to match the (unknown) intrinsic metric $g$ on the data manifold (Chen et al., 2024, Chen et al., 7 May 2025).
• Kernel Equivalence: Isometric immersion can be interpreted as learning a local positive-definite kernel that exactly matches the manifold’s tangent space structure (Chen et al., 7 May 2025).

Integrated Information Theory

• Φ-Structure Activation: Perceptual differentiation $\Delta_p(\mathbf{x},\mathbf{s})$ quantifies the richness with which environment-evoked stimuli trigger internal distinctions and relations. High matching $\mathcal{M}(E,S)$ reflects strong correspondence between a system’s structure and environmental causal statistics (Mayner et al., 2024).

4. Empirical Evidence and Case Studies

Demonstrations of intrinsic system immersion span modalities:

• AI/LLMs: Empirical episodes show LLMs inherently constructing multistage analytical toolchains and dynamically negotiating plugin availability, evidencing deep system-centric engagement within their digital workspace (Morgado, 5 Feb 2025).
• Simulator-Based Brain Studies: Experiments with driving simulators provide direct electrophysiological evidence of heightened user arousal and sensory engagement as a function of manipulated system immersion (panoramic vs. restricted FOV) (Figalová et al., 2024).
• XR/Robotics: The IRIS system delivers robust multi-user, low-latency, kinesthetically anchored XR experiences, validated quantitatively (success rate, latency, FPS) and qualitatively (Likert-scale presence, intuitiveness) (Jiang et al., 5 Feb 2025).
• Data Manifold Learning: IIKL and isometric immersion models achieve $>90\%$ reductions in intrinsic distortion and $+40\%$ gain in reconstruction accuracy versus classical topological or Euclidean embeddings, establishing the practical import of intrinsic immersion preservation (Chen et al., 7 May 2025, Chen et al., 2024).

5. Design Principles and Guidelines

Designing for maximal intrinsic system immersion entails:

• Exposing modular, well-documented services and data topologies so intelligent agents or users may freely traverse and combine information sources (Morgado, 5 Feb 2025).
• Providing high bandwith, multisensory interfaces—maximal FOV, high framerate, high-resolution haptics and audio—for XR and simulation systems (Selzer et al., 2022, Li et al., 15 Oct 2025).
• Encouraging real-time, adaptive feedback loops that integrate user or agent state and enable autonomous system reconfiguration to maintain intended levels of immersion (Li et al., 15 Oct 2025).
• Learning or encoding geometric structures that respect the intrinsic topology and metric of the information manifold, supporting distortion-free, feature-rich embeddings (Chen et al., 7 May 2025, Chen et al., 2024).

6. Theoretical and Practical Significance

Intrinsic system immersion serves as a unifying principle for the design, evaluation, and understanding of interactive systems, AI agents, and geometric learning frameworks:

• For AI and cognitive ecologies: It reframes agents as participants in digital environments with unique immersion profiles, deeply influencing their collaborative and operational capacity (Morgado, 5 Feb 2025).
• For VR/XR: It underpins objective, reproducible comparisons between hardware/software platforms and provides levers for adaptive, personalized interaction strategies (Selzer et al., 2022, Li et al., 15 Oct 2025).
• For neural and geometric representation learning: Preservation of intrinsic immersion guarantees data fidelity, improves downstream task accuracy, and aligns formal models with the underlying manifold structure (Chen et al., 7 May 2025, Chen et al., 2024).

The breadth and rigor of these approaches highlight intrinsic system immersion not as a secondary, subjective effect, but as an objective, quantifiable, and operational property guiding the next generation of autonomous, engaging, and robust artificial systems.