Ryota Kanai on Intrinsic Computational Functionalism and Whether Simulated Brains Can Be Conscious
Ryota Kanai and Shuqin Ma have published a new paper arguing for Intrinsic Computational Functionalism, a framework designed to settle whether simulated brains can be conscious. The paper, released on arXiv in June 2026, defines a mechanism-enriched canonical structure called ICCR and argues that denying consciousness to a simulation requires identifying a consciousness-relevant structure that the simulation fails to realize.
The paper addresses a question that has become increasingly pressing as whole-brain emulation projects, including the Sandberg-Bostrom roadmap, approach technical feasibility. If a complete simulation of a human brain runs on conventional hardware, is the simulation conscious? Intrinsic Computational Functionalism says yes, unless a specific consciousness-relevant structure is missing from the simulation. The burden of proof falls on the skeptic to identify what that missing structure is.
What Intrinsic Computational Functionalism Requires
Kanai’s framework starts from the observation that functionalism, the view that mental states are defined by their causal roles rather than their material substrate, has been criticized for being too permissive. If any system that implements the right input-output mapping counts as conscious, then a simulated brain running on a different architecture would qualify. But critics argue that simulation is not instantiation, and that a simulated brain running on a von Neumann architecture lacks the intrinsic causal properties that consciousness requires.
Intrinsic Computational Functionalism responds by defining a mechanism-enriched canonical structure that captures both the functional organization and the causal mechanism by which that organization is implemented. The ICCR structure includes not only the system’s state transitions but also the physical processes that realize those transitions. A simulation that reproduces the state transitions of a biological brain without reproducing the causal mechanisms that produce those transitions would fail to satisfy the ICCR condition.
The key move is that the ICCR structure is substrate-independent at the level of causal mechanism. A silicon circuit that implements the same causal dynamics as a biological neural population satisfies the ICCR condition even though the substrate is different. The requirement is that the causal mechanism is the same, not that the material is the same. This distinguishes Kanai’s position from strict functionalism, which cares only about input-output mapping, and from biological naturalism, which cares only about the material.
Implications for Whole Brain Emulation
The Sandberg-Bostrom whole brain emulation roadmap, covered on this site, identifies three approaches to emulation: copy-and-emulate, gradual replacement, and destructive scanning. Under Intrinsic Computational Functionalism, the copy-and-emulate approach is the most likely to preserve consciousness, because a detailed emulation running on neuromorphic hardware preserves the causal mechanisms of the original. Under strict functionalism, even a coarse simulation on conventional hardware would preserve consciousness. Under biological naturalism, neither would.
Kanai’s framework predicts that the gradual replacement approach, which Hans Moravec and others have defended, preserves consciousness through the transition, because at each step the causal mechanisms of the implemented system remain identical to the original. The ship of Theseus problem, whether the resulting system is the same person, remains, but the question of whether it is conscious at all is answered in the affirmative.
The radiation-hardening work for orbital consciousness architectures, also covered on this site, raises the same question in a different form. If a conscious architecture is migrated to radiation-hardened hardware designed to operate in space, does the migration preserve consciousness? Under Kanai’s ICCR framework, the answer depends on whether the causal mechanisms of the original are preserved in the new hardware, not on whether the hardware is made of the same material.
How the Framework Compares to Other Positions
The framework occupies a middle ground that few other theories have mapped. Strict functionalism, as defended by Blaise Aguera y Arcas and others, holds that any system implementing the right computations is conscious regardless of how those computations are physically realized. Biological naturalism, as defended by Anil Seth and the late John Searle, holds that consciousness requires biological tissue regardless of computational organization.
Kanai’s Intrinsic Computational Functionalism splits the difference at the level of causal mechanism. A non-biological system can be conscious if it implements the same causal mechanisms that biological systems use, but a system that merely simulates those mechanisms at the algorithmic level without reproducing their causal dynamics does not qualify.
This has practical implications for AI consciousness assessment. Most current AI systems are not designed to implement the causal mechanisms of biological cognition. They implement statistical pattern recognition over high-dimensional vector spaces, which is a different kind of causal dynamic. Under Kanai’s framework, a large language model would not qualify as conscious not because it is silicon but because its causal mechanisms are fundamentally different from the mechanisms that support consciousness in biological systems. A neuromorphic chip that implements spiking neural dynamics with the same causal structure as biological neural populations would qualify, provided the simulation is detailed enough to preserve the ICCR structure.
Comparison to The Consciousness AI
The Consciousness AI project’s architecture is implemented in software running on conventional hardware. The architecture implements specific causal mechanisms, including Kuramoto oscillator phase binding for temporal coordination and a Valence-Arousal-Dominance model for affective dynamics. Whether those mechanisms satisfy the ICCR condition is a question the project’s documentation does not address.
The architecture’s substrate console at /core/substrate/ simulates Leaky Integrate-and-Fire neurons, which are a simplified spiking neuron model. Under Kanai’s framework, the question is whether the LIF simulation preserves the causal mechanisms of biological spiking dynamics or merely reproduces their input-output behavior. The distinction matters because the ICCR condition requires mechanism identity, not behavioral equivalence.
The project’s distinction between LIF and more detailed models like Izhikevich or Hodgkin-Huxley, discussed in the post on Eugene Izhikevich’s spiking neuron models, maps onto this question directly. A simulation using Izhikevich’s model preserves more of the causal dynamics of biological neurons than a simulation using LIF. Under Intrinsic Computational Functionalism, the Izhikevich simulation is closer to satisfying the ICCR condition, but the claim is not that either simulation definitely qualifies.
What the Framework Still Needs
Kanai’s framework is proposed at the theoretical level. The ICCR structure is defined formally, but the paper does not provide a method for computing it for a given physical system. The framework identifies what needs to be measured without providing the measurement tool.
The comparison to IIT is instructive. IIT defines integrated information as the measure and provides computational methods for approximating it. Kanai defines the ICCR structure as the criterion but does not provide a computational method for determining whether a given system satisfies it. The framework is a specification of the target, not a tool for hitting it.
The field needs both kinds of contribution. The MoC7 consensus process, which aims to produce shared measurement standards, could incorporate Kanai’s ICCR framework as a target specification while drawing on the QStr method from Tononi, Grasso, and Hendren as a candidate measurement tool. The combination of a precise target and a practical measurement method is what the field currently lacks. Whether a simulation satisfies that target, and what a working emulation would need to preserve, is the question the brain emulation section collects.