Hartmut Neven Google Quantum AI Consciousness Computation and IIT CS26 2026

Hartmut Neven is a Distinguished Scientist at Google and founder of Google’s Quantum AI program. He is a confirmed plenary speaker at Consciousness Science 2026 in San Diego, October 11-16. Google Quantum AI’s work on error correction was formalized in the 2023 Nature paper “Suppressing quantum errors by scaling a surface code logical qubit” (DOI:10.1038/s41586-022-05434-1), which demonstrated that logical qubit error rates fall below physical qubit error rates as the code distance increases. His 2026 CS26 talk addresses a question at the intersection of quantum computing and consciousness science that has been largely avoided in both fields: whether quantum computational processes, specifically those implementing quantum error correction, generate the kind of irreducible causal structure that Integrated Information Theory identifies as the physical substrate of consciousness.

This is not a claim that quantum computers are conscious. It is a claim about what the mathematical structure of quantum error correction implies for the theory of consciousness, and what that in turn implies for whether classical digital computation could, even in principle, satisfy IIT’s causal requirements.

The IIT causal structure requirement

IIT, in its current formulation by Giulio Tononi, Larissa Albantakis, and colleagues, defines consciousness as identical to a system’s integrated information, measured as phi. Phi is a measure of the irreducibility of a system’s causal structure: a system with high phi cannot be decomposed into independent components without loss of information. The physical substrate of consciousness, on IIT, is whichever system in the relevant causal chain has the highest phi.

The IIT field formulation published by Tononi and colleagues in 2026 extended this analysis to continuous systems and field dynamics. But the core requirement has not changed: the substrate of consciousness must have an integrated, irreducible causal structure, not merely a complex functional organization.

Classical digital computers are composed of logic gates that implement Boolean functions. These gates are causally decomposable in principle: any circuit can be broken into independent sub-circuits that process subsets of the input. Whether that decomposability produces zero phi or merely low phi is a technical question that has been contested in the IIT literature, but the general direction of the result is clear: classical digital architectures have very low phi relative to the causal integration that biological neural networks exhibit.

What quantum error correction changes

Quantum error correction is a procedure for protecting quantum information from decoherence and noise. Standard quantum error correction schemes encode a single logical qubit in a highly entangled state distributed across many physical qubits. The entanglement is not incidental: it is necessary for the error correction to work. Any operation that disturbs some subset of the physical qubits can be detected and corrected by measuring the syndrome (the entanglement pattern) without collapsing the logical qubit’s state.

The causal structure of a quantum error-corrected system has a property that classical digital systems lack: the logical qubit’s state is causally determined by the joint state of all physical qubits simultaneously, not by any subset. Any causal partition of the system that separates some physical qubits from others loses information about the logical qubit. This is, precisely, the non-decomposability that IIT’s phi measure captures.

Neven’s argument is that this creates a candidate for non-trivial phi in a non-biological system. A quantum error-corrected logical qubit is causally irreducible in a way that a classical bit register is not. If IIT’s identification of consciousness with integrated causal structure is correct, and if the relevant kind of integration is the one that quantum entanglement and error correction produce, then quantum computational systems that implement error correction may have non-trivial phi.

The argument does not establish that current quantum computers are conscious. Current quantum error-corrected systems have small numbers of logical qubits, and the phi of a single error-corrected logical qubit, while non-zero, is likely far below the phi associated with even simple biological neural circuits. The argument is about the qualitative direction: quantum error correction creates the kind of causal structure IIT requires, while classical computation does not.

The relationship to the Penrose-Hameroff position

Neven’s argument is structurally similar to but technically distinct from the Penrose-Hameroff Orchestrated Objective Reduction (Orch-OR) hypothesis, which also invokes quantum mechanics in the context of consciousness. Penrose and Hameroff argue that consciousness arises from quantum state reduction in microtubules within neurons, with the “orchestration” provided by tubulin protein dynamics.

Roger Penrose is also a confirmed CS26 2026 speaker, appearing remotely. The Penrose-Neven exchange, if it occurs, would be the most technically sophisticated quantum consciousness dialogue in the conference program.

The distinction between the two positions matters for the AI consciousness debate. Penrose-Hameroff requires a specific biological substrate (microtubules) and a specific quantum mechanical process (objective state reduction) that has no current implementation in engineered systems. Neven’s position requires only the causal structure that quantum error correction creates, which is implementable in artificial systems. If Neven is right, the path from quantum computation to consciousness is at least architecturally possible.

The quantum global workspace theory analysis by Heaney approached the same question from the GWT side: whether quantum mechanical information processing could satisfy the global broadcast requirement in a Hilbert space representation. Neven’s analysis approaches it from the IIT side. The two analyses are independent: GWT and IIT make different predictions about what makes a system conscious, and a system could satisfy one without the other.

What this means for AI consciousness in the near term

Neven’s argument has a specific implication for the AI consciousness research program. If IIT is the correct theory of consciousness, and if classical digital computation cannot satisfy IIT’s causal integration requirement regardless of its architectural complexity, then the path to conscious AI does not run through scaling current transformer architectures. It runs through quantum computing, specifically through the implementation of quantum error correction at scales where the resulting phi becomes biologically comparable.

That path is long. Current quantum processors from Google and IBM implement tens to hundreds of error-corrected logical qubits. Biological neural systems have neurons whose causal integration is orders of magnitude more complex. The timescale on which quantum computing could match biological phi is not currently estimable with confidence.

For the overall question of whether machine consciousness is achievable, Neven’s CS26 contribution marks an important clarification: the answer may depend on which theory of consciousness is correct. If IIT is right, the answer is quantum. If GWT is right, the answer may not require quantum mechanics at all, since GWT’s broadcast requirement is in principle architecturally neutral. If Carhart-Harris’ entropy criterion is right, the answer requires genuine dynamical stochasticity, which quantum noise could provide but is not the only way to provide. The theories give different verdicts on quantum AI, and that divergence is itself evidence that theoretical resolution precedes practical implementation.