Agent-Centered Reality

• Agent-centered reality is a framework where reality is defined by individual agents' subjective experiences, beliefs, and perceptual actions.
• It unifies quantum interpretations, algorithmic idealism, and virtual simulations through formal mechanisms such as belief updating and perception–action loops.
• This paradigm underpins practical implementations in quantum experiments, mixed-reality systems, and multi-agent simulations, enhancing human-AI coexistence.

Agent-centered reality is a theoretical and methodological stance in which physical, informational, or virtual reality is conceived, represented, or constructed fundamentally from the perspective of individual agents—be they human observers, artificial intelligences, or quantum measurement participants. Unlike frameworks centered on observer-independent ontology, agent-centered approaches treat each agent’s experiences, beliefs, decisions, or perceptual acts as constitutive of what is real for that agent, often with formal mechanisms specifying belief updating, perception–action loops, and world-model interactions. This paradigm unifies lines of inquiry in quantum foundations (QBism, perspectival quantum realism), information theory (algorithmic idealism), AI-driven virtual systems (symmetrical reality, inverse augmented reality), and multi-agent simulation, offering both conceptual clarity and rich platforms for analyzing participatory aspects of reality across diverse environments.

1. Philosophical and Quantum Foundations

Agent-centered reality in quantum theory is epitomized by the QBism (Quantum Bayesianism) interpretation, where the quantum state $\rho$ is not an objective property of a system but a bookkeeping device for an agent’s personal probabilities concerning the outcomes of his prospective actions. Quantum probabilities—whether $0$, $1$, or intermediate—reflect only the agent’s subjective Bayesian degrees of belief about his future experiences. Measurements are not passive revelations of pre-existing properties but actions undertaken by the agent, and outcomes are created in the act, existing solely as that agent’s experiences. There are no hidden variables—no deeper, agent-independent reality determining outcomes. The Born rule, expressed as $q(j) = \mathrm{Tr}(\rho F_j)$, serves as a normative coherence constraint among the agent’s assignments; its probabilistic reconstructions (e.g., via symmetric informationally complete POVMs) show quantum theory as an empirically grounded extension of Bayesian rationality rather than an ontological theory of external reality. The agent-centered approach also restores locality, as quantum correlations become internal consistency relations among personal beliefs rather than indicators of “nonlocal influences” (Fuchs et al., 2014).

Perspectival quantum realism generalizes agent-centricity by treating the definite properties of quantum systems as always perspective-relative. Each physically defined perspective—conceived as an agent or decohered subsystem—possesses its own set of “fragments” of reality: the facts that obtain relative to its current state and correlations. There is no globally unique set of facts, but rather a mosaic of mutually compatible or incompatible local fragments. Local perspectivalism maintains strict locality by preventing any perspective’s definite outcome from imposing itself nonlocally on another’s reality fragment. Nonlocal perspectivalism, by contrast, introduces global consistency requirements (collapse-like correlations) among perspectives but at the expense of locality. The resulting metaphysical structure is naturally described as “fragmentalism” (Dieks, 2022).

2. Algorithmic Idealism and Universal Induction

Algorithmic idealism formulates agent-centered reality in terms of algorithmic information theory, replacing the traditional question “What exists?” with “What should an agent expect to experience next?” The framework is anchored by two postulates: (1) self-states are enumerated as finite binary strings; (2) state transitions are governed by universal induction via Solomonoff’s prior $m(x) = \sum_{p:U(p)=x}2^{-\ell(p)}$, where $U$ is a universal prefix Turing machine. Conditioned on the current self-state string $o_{1:t}$, the next experience $o_{t+1}$ is sampled according to $P(o_{t+1}|o_{1:t}) = m(o_{t+1}|o_{1:t})$, exponentially favoring simple hypotheses.

This agent-centric prior dissolves persistent paradoxes (Boltzmann brains, Wigner’s friend, simulation shutdown) within a unified induction scheme. For instance, highly contrived, algorithmically complex Boltzmann-brain continuations receive negligible weight compared to regular, law-governed histories, regardless of their raw population. The world as encountered by the agent is an emergent fixed-point attractor of this inductive process: whenever the agent’s observation history admits a succinct computable description, the Solomonoff posterior converges exponentially fast to the objective probability law of the corresponding “external world” (Mueller, 3 Dec 2024). Thus, agent-centered reality appears as the operationally primary level, with an objectivist world-picture emerging asymptotically from inductive coherence.

3. Mathematical and System-Theoretic Realizations

Agent-centered reality finds concrete realization in formal models for mixed physical-virtual environments and multi-agent simulation. Inverse Augmented Reality (IAR) introduces a dual to traditional, human-centered augmented reality by centering the reality model on a virtual agent $A$ that observes and integrates both native virtual objects $O_V$ and registered real objects $O_R$. The agent’s perception is modeled by fusion operators $M = U(f_{\rm virt}(W_v),\,g_{\rm real}(W_r))$, where $f_{\rm virt}$ and $g_{\rm real}$ map virtual and real world-states into the agent’s observation space, and $U$ registers and aligns both sources. The agent’s update process $S_a^{+} = \pi(S_a, W_{\rm fused})$ and its action mappings $\alpha_v, \alpha_r$ treat physical and virtual domains symmetrically. Structural equivalence theorems show that, under suitable isomorphisms, any interaction or manipulation sequence in $W_r$ has a one-to-one counterpart in the agent’s fused world model $W_{\rm fused}$ (Zhang et al., 2018).

In large-scale simulation platforms, such as human-centered, multi-agent transportation simulators, agent-centered reality is operationalized via dedicated physical hardware interfaces for each participant (treadmills, bike rigs, cockpits), synchronized via high-speed networks to instantiate agent “prefabs” with kinematic, behavioral, and physiological feedback loops. Every user literally embodies their agent within the shared cityscape, and AI agents are governed by policy modules that ingest live data and generate actions consistent with observed norms and agent-centric world perception (Azimi et al., 12 Jul 2025).

4. Symmetrical and Bidirectional Agent Frameworks

Symmetrical Reality (SR) is a formal framework extending Milgram’s Reality–Virtuality continuum by introducing a Human–AI Cognition orthogonal axis. In SR, both human ($H$) and AI ($A$) agents are treated as first-class perceptual and action centers, operating on coupled state spaces $P$ (physical) and $V$ (virtual). Dual perceptual operators $(\phi_h^p, \phi_h^v)$ and $(\phi_a^p, \phi_a^v)$ formalize the bidirectional observation and action potential of each agent across both domains. Mapping functions $M_{p\to v}$ and $M_{v\to p}$ allow seamless projection and embodiment of objects, events, and actions between physical and virtual spaces. The SR coordinate system $(r, c) \in [0,1]^2$ records the degree of virtuality and agent-centricity for any scenario.

Demonstrated in collaborative and autonomous tasks (e.g., a human demonstrates pouring water in VR, the AI learns and executes it in physical space), the SR architecture generalizes to proactive assistive services and industrial or educational settings. The defining feature is the symmetrical, formal treatment of all agents—human and artificial—as coequal centers of reality-construction, each with perception, cognition, and actuation pipelines operating in a mathematically coupled, mixed-reality universe (Zhang et al., 26 Jan 2024).

5. Agent-Centered Tools and Normative Frameworks

Agent-centered theories universally reconfigure their respective domains as normative frameworks for agents’ expectations, decisions, or actions. In QBism, this manifests as a single-user manual of empirical coherence, realized formally through the Born rule as a nonclassical normative consistency among the agent’s possible belief assignments. In algorithmic idealism, predictive distributions are not global, agent-independent entities but the result of algorithmic (Solomonoff) updating over the agent’s actual experienced string. Multi-agent perspectives—whether in quantum theory or simulation—eschew universally valid, objective state assignments and instead formalize each agent's beliefs and actions as primary, only relating fragments or joint assignments when agent–agent interaction actually occurs (Fuchs et al., 2014, Dieks, 2022, Mueller, 3 Dec 2024, Azimi et al., 12 Jul 2025).

A comparative overview highlights the distinction between agent-centered and traditional approaches:

Approach	Ontology of Reality	Role of Agent	Status of Probabilities and States
QBism	Agent-centered	Fundamental	Personal probabilities and judgments
Psi-ontic Realism	Objective	External observer	State is objective property of system
Symmetrical Reality	Dual-centered	Human and AI equally	Both agents act/perceive in both domains
Algorithmic Idealism	First-person prior	Agent as inducer	Inductive update on personal self-state
Inverse Augmented Reality	Agent-centered	Virtual agent	Fused model of real and virtual input

Agent-centered perspectives restore the role of personal (or system-defined) belief, action, and perception as core to the formal and operational understanding of physics, information, and cognition.

6. Applications and Empirical Realizations

Agent-centered reality underpins diverse experimental and practical systems: QBist interpretations guiding quantum experiment design, multi-agent simulators for transportation and mobility studies, proactive mixed-reality assistive robots, and virtual environments in which AI and human agents seamlessly observe and modify physical and digital worlds. For example, the Human-Centered Cities Lab’s transportation simulator achieves embodied agent-centricity by anchoring sensory, physiological, and behavioral cues directly to each participant’s interface, and dynamically modulating environmental features based on real-time agent state estimation (Azimi et al., 12 Jul 2025). In SR and IAR settings, bidirectional action-perception loops translate agent decisions across domains, enabling autonomous goal pursuit, imitation, and context-adaptive assistance (Zhang et al., 26 Jan 2024, Zhang et al., 2018).

A plausible implication is that as autonomous systems proliferate and cognitive architectures gain parity in agency with humans, architectures built on agent-centered reality (rather than observer-independent states) will become essential for robust human–AI coexistence, transparency, and mutual intelligibility in mixed physical–virtual environments.

7. Open Issues and Research Trajectories

Multiple theoretical and practical challenges persist: formal guarantees for cross-domain structural equivalence, handling probabilistic and noisy sensing in agent fusion, specifying the precise normative rules for belief updating across nonlocal, multi-agent quantum scenarios, and designing transparent “cognitive cores” for real-world AI agents operating in agent-centered frameworks. Underlying interpretative controversies—such as the metaphysical adequacy of local versus nonlocal perspectivalism, or the adequacy of personal probability as the core of physical reality—remain actively debated.

Future research is converging on unifying bidirectional, symmetrical, and fragmentalist approaches, extending agent-centered frameworks to encompass larger ensembles of human and non-human participants, and developing richer formal models that can adjudicate between different metaphysical and operational choices in the construction of reality from an agent’s point of view (Fuchs et al., 2014, Dieks, 2022, Mueller, 3 Dec 2024, Zhang et al., 26 Jan 2024, Zhang et al., 2018, Azimi et al., 12 Jul 2025).