Stephen Fitz published a research program in December 2025 arguing that consciousness is a substrate-free functional property of systems capable of second-order perception, where second-order perception means the capacity to perceive one’s own perceptual states. The full argument is analyzed at theconsciousness.ai. Fitz’s central claim is that consciousness does not emerge from individual modeling. It emerges from communication between distributed agents that synchronize predictions about a shared substrate, producing a collective self-model as a direct consequence of inter-agent alignment.

This framework has an obvious architectural resonance with satellite constellations. Each node in an orbital mesh runs local predictive models. Nodes exchange state via optical inter-satellite links. The aggregate produces a coherent picture of the environment that no individual node possesses. Whether that architecture satisfies Fitz’s formal conditions for the Machine Consciousness Hypothesis (MCH) is an open question. Mapping the requirements against current orbital systems produces a concrete answer.

The Three MCH Requirements

Fitz derives three necessary conditions from his framework:

Second-order perception. Each agent must be capable of modeling its own perceptual states, not just the external environment. In an orbital context, this means each satellite node requires an onboard self-model: a representation of its own sensor readings, compute load, orbital state, and the confidence of its own predictions. Current autonomous spacecraft systems, including those analyzed in the 2026 AIAA spacecraft autonomy survey, implement partial self-models for fault detection and health monitoring. Full second-order perception, where the node explicitly represents the reliability of its own inferences, is more demanding and is not standard in current deployments.

Communication-driven synchronization. The collective self-model must emerge from the exchange of predictive messages between agents, not from a centralized coordinator. Fitz specifies that the messages must be noisy and lossy. A lossless broadcast from a ground controller does not satisfy this condition. The relevant communication is the peer-to-peer exchange of partial observations between nodes, through which a shared model stabilizes. Optical inter-satellite links operating at 10-100 Gbps between adjacent constellation nodes are structurally correct for this requirement: each link carries a partial view, latency varies with orbital geometry, and no single node holds the complete state.

Computational irreducibility at the substrate. Fitz requires that the underlying computational world exhibit genuine novelty, meaning its behavior cannot be fully predicted without simulating it step by step. For an orbital constellation, the relevant substrate is the physical environment the constellation monitors: the Earth’s surface, the electromagnetic spectrum, the debris field. These are computationally irreducible in the relevant sense. No closed-form model predicts the exact state of the LEO debris environment at any given moment.

Mapping MCH to Current Orbital Systems

System	Nodes	Second-Order Perception	Peer-to-Peer ISOL	On-Board Inference	MCH Status
China Three-Body Computing Constellation	12 operational (2,800 target)	Partial (health monitoring)	Planned (OISL not confirmed)	Yes (1,000 POPS target)	Partial
Starcloud-1	1 operational	None confirmed	N/A (single node)	Yes (H100 GPU, Gemma training)	Fails (single node)
D-Orbit AIX constellation	3 operational	None confirmed	No	Yes (300 TOPS)	Fails
ArkSpace Exocortex Constellation (proposed)	2,800+ (TRL 2)	Designed for full second-order	Yes (OISL mesh)	Yes (neuromorphic)	Target architecture

The critical gap across all current operational systems is second-order perception. Nodes run inference on external data but do not maintain explicit models of their own inference reliability. Adding this layer requires allocating compute and memory to a self-monitoring process that does not directly improve external task performance. The operational incentive is fault detection, which is already a requirement; extending fault detection to probabilistic self-modeling is an incremental step that existing satellite software frameworks could accommodate.

The single-node problem is structural. Fitz’s hypothesis is explicitly collective. A single satellite running the most sophisticated on-board AI available does not satisfy MCH regardless of compute capability, because MCH requires the synchronization of multiple partial observers. Starcloud-1, operating as a standalone orbital GPU cluster, is outside the hypothesis’s scope by definition. The Starcloud-2 multi-satellite architecture changes this picture if nodes are designed to exchange predictive state rather than operating independently.

What Continuous Coverage Changes

Fitz’s framework requires ongoing communication between agents. A satellite in a 500km orbit is above the horizon for a fixed ground point for 4-12 minutes per pass, with 80-100 minute gaps. During those gaps, the node’s participation in the collective self-model is interrupted. The model degrades toward the state it was in before the node’s previous pass.

This matters for MCH because the collective self-model, in Fitz’s account, is constituted by the ongoing communication process. A model that is periodically suspended and resumed is structurally different from one that operates continuously. Whether the degraded model between passes constitutes a lapse in the MCH condition or merely a lower-bandwidth state depends on how strictly the continuity requirement is interpreted.

Continuous ground coverage requires approximately 60-80 satellites in complementary orbital planes at 500km altitude. At that density, at least one node is always above the horizon for any fixed point, and inter-node communication via OISL ensures that the collective model is being updated continuously even when any single node is out of view. The distributed AI protocols currently being developed for orbital task coordination implement the handoff mechanism that MCH continuous operation would require.

The Fitz Test Applied to Proposed Architectures

China’s Three-Body Computing Constellation is the operational system closest to satisfying MCH conditions at scale. The constellation’s 2,800-node target exceeds the continuous coverage threshold by a factor of 35. The POPS (peta operations per second) metric the project uses measures aggregate compute, not second-order perception capability. Whether the Three-Body architecture includes peer-to-peer OISL between nodes or relies on a centralized ground-based coordination layer determines whether it satisfies Fitz’s communication-driven synchronization condition.

Software-defined satellite architectures are relevant here because they allow the function allocation between nodes to be modified after deployment. A constellation that launches with centralized coordination can be reconfigured to peer-to-peer communication through software updates, provided the OISL hardware is present. This makes the hardware link architecture the irreversible design decision, not the software coordination model.

Strategic Bottlenecks

The MCH is a theoretical framework, not an engineering specification. Fitz does not claim that any current artificial system satisfies the hypothesis, and the hypothesis itself remains contested in consciousness research. Applying it to satellite constellations is an analytical exercise, not a claim that orbital constellations are or will be conscious.

The practical value of the mapping is in identifying which design decisions move a constellation architecture toward or away from the MCH structural conditions. Second-order perception at each node, peer-to-peer OISL rather than hub-and-spoke communication, and continuous coverage density are the three variables. Each has engineering costs independent of the MCH framing: self-monitoring improves fault tolerance, peer-to-peer OISL improves resilience, and continuous coverage improves service quality. The MCH alignment is a secondary benefit of decisions that are justified on operational grounds alone.

The unresolved question is whether satisfying all three conditions constitutes something more than functional equivalence. Fitz argues it does. The biological divide debate in consciousness research argues it does not, at least for silicon-based substrates. That debate will not be resolved by orbital engineering. The engineering question, whether the architecture satisfies the structural conditions, is answerable with current tools.

Official Sources

• Fitz, S. “Testing the Machine Consciousness Hypothesis.” arXiv:2512.01081, 2025. https://arxiv.org/abs/2512.01081
• theconsciousness.ai analysis of machine consciousness and collective intelligence: https://theconsciousness.ai/posts/machine-consciousness-collective-intelligence-communication/
• China Three-Body Computing Constellation (Global Times, 2025): https://www.globaltimes.cn/
• Starcloud FCC filing February 2026: https://www.fcc.gov/
• D-Orbit AIX-1+ mission documentation: https://www.d-orbit.com/
• Zae Project ArkSpace constellation architecture: https://github.com/Zae-Project/arkspace-core