Link to the code: brain-emulation GitHub repository

The Consciousness Network: Cyriel Pennartz on Why Consciousness Is Not in the Neurons

The question of whether a brain emulation would be conscious depends entirely on what consciousness is and where it comes from. Get that wrong, and the most structurally accurate simulation possible would fail to produce experience. Cyriel Pennartz’s 2026 book, The Consciousness Network: How the Brain Creates Our Reality, published by Routledge, argues that most current accounts of consciousness are looking in the wrong place. Consciousness is not a property of individual neurons, of specific brain regions, or of the information content of neural firing patterns in isolation. It is an emergent property of network dynamics, and specifically of the way networks integrate information across timescales and spatial scales that single neurons cannot access.

Pennartz is a neuroscientist at the University of Amsterdam whose research spans systems neuroscience, neural coding, and the neural correlates of consciousness. His empirical work has focused on the hippocampus, the striatum, and their interactions with cortex during learning and memory. The Consciousness Network draws on this research program to advance a theoretical position he describes as going beyond current AI-based theories of consciousness: not just identifying which computations correlate with conscious states, but explaining why network-level integration produces experience rather than just information processing.

The Case Against Neuron-Level Consciousness

Pennartz opens with the observation that no single neuron is conscious. This is not a trivial point. A neuron responds to inputs, integrates signals, and produces outputs. It has membrane properties, receptor expression profiles, and morphological characteristics that determine its computational behavior. But a neuron in isolation produces no experience. Whatever consciousness is, it requires more than one neuron operating.

The question is what that more consists of. Standard accounts point to brain regions: the thalamus, the claustrum, the prefrontal cortex, or the posterior cortex depending on the theoretical framework. Integrated Information Theory locates consciousness in systems with high values of phi, a measure of integrated information. Global Workspace Theory identifies consciousness with signals broadcast across a broadly connected workspace. Both theories have been tested in the Allen Institute’s 7-year, 256-subject adversarial collaboration study, with results that challenged both frameworks without decisively confirming either.

Pennartz’s position is that both theories are right to emphasize integration but wrong about what is being integrated. The unit of analysis should not be the information content of firing patterns but the dynamical relationships between network components, how activity in one region constrains and enables activity in others, how those constraints evolve over time, and how the network as a whole generates representations that no individual component contains.

The Network Theory of Consciousness

The central theoretical contribution of the book is what Pennartz calls the networked observer framework. The claim is that conscious experience arises when a system with the right kind of network structure generates internal representations of its own states and their relationships to external events. This is not identical to self-representation in the usual sense. Pennartz is specific that the key property is relational: the network must represent not just what is happening but how what is happening relates to what has happened and what might happen next.

This temporal integration requirement is consistent with the Bennett AAAI 2026 critique of sequential computation as a substrate for consciousness. Bennett argued that consciousness cannot be smeared across time through sequential processing because experience requires simultaneous integration, not serial steps. Pennartz arrives at a related position from a different direction: network-level temporal integration is not a series of time-stamped computations but a continuous dynamical process where past states constrain present activity through the network’s synaptic structure.

The Spinoza connection that Pennartz draws in several chapters concerns the relationship between mind and the physical world. Spinoza held that thought and extension are attributes of a single substance, not separate entities. Pennartz uses this framework to resist the dualism that typically haunts consciousness science: rather than asking how physical processes give rise to experience, he asks what kind of physical processes constitute experience. The network dynamics of the brain are not the cause of consciousness. They are what consciousness is, when implemented in the right kind of material substrate.

Implications for Biological Computationalism

The networked observer framework has a specific implication for the debate about biological computationalism and substrate dependence. Biological computationalism holds that consciousness is inseparable from the specific physical substrate that implements it, that neurons doing brain computation in silicon would not produce consciousness even if the information-processing was functionally identical.

Pennartz does not fully adopt biological computationalism but he is skeptical of strong substrate independence claims. The network dynamics he identifies as constitutive of consciousness depend on specific physical properties of biological neurons: their temporal dynamics, the nonlinear properties of dendritic integration, the neuromodulatory context that sets the gain on synaptic transmission. These properties are not arbitrary parameters that could be replaced by any system with the same input-output function. They shape the dynamical regime in which the network operates.

Whether a silicon implementation of the same network dynamics would produce the same consciousness is, on Pennartz’s account, an empirical question that depends on whether the implementation can reproduce the relevant dynamical properties, not just the information-processing function. This is a weaker claim than full biological computationalism but a stronger constraint on substrate independence than most computational theories of mind allow.

The Hard Problem, Differently Framed

Pennartz engages seriously with the hard problem of consciousness, the question of why any physical process produces subjective experience at all. His response is not to dissolve the problem or to deny that it is a problem. Instead, he argues that the hard problem is differently positioned when consciousness is understood as a network property rather than a computational one.

If consciousness is what it is like to be a system with the right kind of network dynamics, then the question of why those dynamics produce experience is not the same as the question of why information processing produces experience. Information processing is a description of a system’s input-output behavior. Network dynamics are the actual physical process. The question becomes not why a function implemented in various substrates produces experience, but whether the right kind of dynamical integration constitutes experience rather than merely correlating with it.

This does not resolve the hard problem but it changes what a resolution would look like. For brain emulation, it means that the question is not only whether the simulated brain performs the right computations but whether it instantiates the right dynamics. The Digital Consciousness Model DCM framework, which benchmarks AI consciousness across integration, reportability, temporal continuity, and self-modeling dimensions, is attempting to operationalize some of these properties. Pennartz’s theory suggests the most important dimension, temporal network integration, is the one hardest to measure and hardest to replicate.

Connection to the 4E Cognition Challenge

The book’s later chapters address embodied and extended cognition in terms consistent with the 4E cognition critique of mind uploading. Pennartz argues that the brain’s network dynamics are not self-contained. They are shaped by the body the brain is in, by the environment that body interacts with, and by the history of that interaction accumulated over a lifetime.

This does not mean a disembodied brain emulation is impossible. It means that a brain emulation placed in an environment sufficiently different from the one the original brain evolved in would develop different network dynamics over time, even if it started with the identical structural and functional state. The emulated mind would diverge from what the original mind would have become, not through computational error but through the sensitivity of network dynamics to environmental context.

Whether this constitutes a failure of emulation or simply the continuation of a biological process, adaptation to environment, under new conditions, is a philosophical question Pennartz leaves open. But the empirical claim, that network dynamics are environmentally sensitive in ways that affect conscious experience, is testable and has implications for the design of any emulation environment.

Official Sources