Every major debate about mind uploading eventually reaches the substrate question: can consciousness survive transfer to a silicon medium, or is the biological brain irreplaceable? A 2025 Royal Society paper by Nicolas Rouleau and Michael Levin takes a direct route through that question by reading the primary texts of the five most influential consciousness theories and asking what each actually specifies about substrate. The answer is that none of them does.

That finding has direct implications for whole brain emulation research. If consciousness theories are substrate-neutral by their own formal definitions, then objections to WBE on theoretical grounds require an additional premise that the theories themselves do not supply.

The Core Shift

Rouleau and Levin examined Integrated Information Theory (IIT), Global Workspace Theory (GWT), Higher-Order Theories (HOT), predictive processing frameworks, and Attention Schema Theory (AST). Each theory defines consciousness in terms of operations and information structures: IIT specifies irreducible cause-effect power; GWT specifies global broadcast accessibility; HOT specifies second-order representations; predictive processing specifies hierarchical error minimization; AST specifies a self-model of attention.

What none of them specifies is that these operations must be implemented in carbon-based biological tissue. The brain-centric framing in consciousness research reflects the historical accident that all confirmed conscious systems so far have been biological organisms. The theories themselves were not designed with that constraint.

Rouleau and Levin also examine empirical evidence from pre-neural biological systems: planaria (flatworms), slime molds, and collective cell behaviors. These systems demonstrate IIT-relevant integration, GWT-relevant information sharing, and predictive-processing-relevant anticipatory responses in the absence of nervous systems. The implication is that consciousness-relevant mechanisms predate neurons and can appear wherever the right information-processing conditions are met.

This connects directly to work on the digital sphinx cross-species connectome transfer, where C. elegans wiring controlled a virtual fly body, demonstrating functional substrate transfer across species. Whether consciousness transfers alongside function is precisely what the Rouleau-Levin framework addresses at the theoretical level.

Comparative Data

Theory	Formal Consciousness Requirement	Substrate Specified?	What Silicon Must Implement
Integrated Information Theory (IIT 4.0)	Maximal irreducible cause-effect power over a system’s own states	No	Genuine causal integration, not extrinsic computation by a simulator
Global Workspace Theory (GWT)	Global broadcast of information to distributed specialist processors	No	Architectural broadcast mechanism with modular specialization
Higher-Order Theories (HOT)	Second-order representations of first-order mental states	No	Self-referential representational capacity
Predictive Processing	Hierarchical generative model minimizing prediction error	No	Bidirectional prediction-and-correction signal flow across levels
Attention Schema Theory (AST)	Internal model of attention allocated to objects	No	Self-model of attentional state, distinct from the attention mechanism itself

No row in the substrate column reads “Yes.” The right column identifies what artificial systems must implement to satisfy each theory, and these are computational specifications, not material ones.

The distinction matters for WBE design. A system that implements genuine causal integration (IIT), a global broadcast bus (GWT), self-referential representations (HOT), hierarchical prediction (PP), and an attention self-model (AST) satisfies the formal criteria of all five major theories without requiring neurons.

Practical Impact

Rouleau and Levin’s framework shifts where the burden of proof sits in WBE debates. Objections to substrate-independent consciousness now require specifying which formal consciousness requirement is violated by the proposed substrate, rather than appealing to general brain-centrism.

The biological computationalism framework makes the strongest counter-argument: consciousness may require specific computational dynamics (hybrid discrete-continuous, metabolically grounded, scale-inseparable) that silicon does not currently implement. Biological computationalism accepts substrate relevance while rejecting strict biological essentialism. That position is compatible with Rouleau-Levin: both agree that theories do not require neurons; biological computationalism then adds that the computational type matters, not the atomic composition.

For the adversarial collaboration study on IIT versus GWT, where neither theory’s core predictions were fully confirmed in 256 biological subjects, Rouleau-Levin provides important context: if neither theory is fully validated in biological systems, the substrate debate cannot be resolved by appealing to those theories as settled foundations. WBE validation frameworks need to be designed around theory uncertainty.

Eon Systems’ fruit fly brain emulation produced full behavioral replication from connectome data. Whether that emulation also preserved any consciousness-relevant architecture is the question Rouleau-Levin equips researchers to ask: which of the five theoretical requirements does the implemented system satisfy, and which does it fail?

For researchers at theconsciousness.ai tracking the Artificial Consciousness Module project, the complete analysis of what each theory requires from non-biological substrates covers the Royal Society paper in detail, including empirical evidence from pre-neural biological systems that supports the substrate-agnostic reading.

Limitations and Open Questions

The Rouleau-Levin argument establishes theoretical permissibility, not feasibility. Showing that no major theory formally excludes silicon does not show that silicon implementations can achieve what biological tissue achieves. The gap between theoretical specification and physical implementation remains the central engineering challenge.

IIT’s cause-effect power requirement is the hardest to satisfy in digital systems. Simulation on a classical computer introduces an additional substrate layer (the computer’s own causal structure) that may not implement the intrinsic causal integration IIT requires. The IIT 4.0 measurement methodology addresses this directly by asking whether cause-effect power is preserved across implementation levels, rather than assuming it transfers automatically from algorithm to hardware.

Pre-neural biological evidence shows consciousness-relevant mechanisms in non-nervous systems, but does not show that those mechanisms produce phenomenal experience. The hard problem remains orthogonal to the substrate debate.

Rouleau and Levin do not address the measurement gap: even if silicon can in principle satisfy all five theoretical frameworks, current measurement tools for consciousness are calibrated against biological data. Applying them to radically different substrates without recalibration produces results with uncertain interpretation.

---

Official Sources

• Nicolas Rouleau & Michael Levin. “Brains and Where Else: Why Major Consciousness Theories Do Not Require Neural Tissue.” Philosophical Transactions of the Royal Society B, 2025. https://royalsocietypublishing.org
• theconsciousness.ai analysis: Brains and Where Else: Why Major Consciousness Theories Do Not Require Neural Tissue
• Tononi, G. et al. “Integrated information theory of consciousness: An updated account.” PLOS Computational Biology, 2016.
• Baars, B. J. “In the theater of consciousness: Global Workspace Theory, a rigorous scientific theory of consciousness.” Journal of Consciousness Studies, 1997.
• Sandberg, A. & Bostrom, N. “Whole Brain Emulation: A Roadmap.” Future of Humanity Institute Technical Report, 2008.