Max Tegmark and Perceptronium as a State of Matter
Max Tegmark asks a physicist’s question about consciousness. Solids, liquids and gases are distinguished by measurable properties, so if conscious matter is a thing, what properties distinguish it? In Consciousness as a state of matter, published in Chaos, Solitons and Fractals volume 76 in 2015, pages 238 to 270, he names the hypothetical substance perceptronium and proposes five principles it would have to satisfy. The paper generalises Giulio Tononi’s integrated information framework beyond neural networks to arbitrary quantum systems, and it inherits both the framework’s ambition and its central difficulty.
| Principle | What it requires |
|---|---|
| Information | The system stores a substantial amount of information |
| Integration | That information is unified rather than decomposable into independent parts |
| Independence | The system is largely autonomous from the rest of the world |
| Dynamics | It processes information over time rather than holding it statically |
| Utility | It records information useful for the system rather than arbitrary information |
Why Ask It This Way
The move is to treat consciousness as an emergent property of matter organised in a particular way, in the same category as superconductivity or magnetism. Those are also properties no individual particle has, that appear at scale, and that are defined by measurable structural facts rather than by what the material is made of.
If that framing is right, then asking whether silicon can be conscious is like asking whether a material other than mercury can superconduct. The answer depends on the organisation, not the element, and the question becomes tractable rather than metaphysical.
Tegmark is at MIT and works primarily in cosmology and machine learning, which shows in the method. The paper looks for the conditions under which a subsystem of the universe would have the properties consciousness is taken to have, and connects them to error correcting codes and to criticality in condensed matter physics.
The Independence Principle Is the Interesting One
Four of the five principles restate familiar territory. Information and integration are Tononi’s, and dynamics is uncontroversial. Independence is Tegmark’s own contribution and it does real work.
A conscious system must be substantially decoupled from its environment. If a subsystem’s states are continuously determined by external conditions, there is no meaningful sense in which it has its own states at all. Some autonomy is required for anything to be true of the system rather than of the world it sits in.
That has a consequence for artificial systems that is rarely discussed. A model whose activations are entirely determined by its input, with no persistence and no internal state carried across, is maximally coupled and minimally independent. Whether that disqualifies it, or merely makes it a poor candidate, is a question the principle raises and does not settle.
The utility principle is the weakest. Requiring that the information be useful to the system imports a notion of what the system is for, and a purely physical account is not entitled to that without saying where the purpose comes from.
Where It Sits
Perceptronium is an IIT derivative, so it inherits IIT’s problems along with its structure. Chief among them is that the central quantity cannot be computed for any real system, and that a measure of integration ranks simple highly connected systems above brains. That objection is set out in Scott Aaronson and the expander graph objection to integrated information theory, and it applies to Tegmark’s generalisation with equal force.
The generalisation to quantum systems adds a difficulty of its own, which is the factorisation problem. Which decomposition of a quantum system into subsystems is the right one? The answer determines every quantity in the framework, and physics does not supply a preferred choice. Tegmark treats this as an open problem rather than a solved one, which is honest and leaves the framework incomplete.
Where the position sits relative to the alternatives is set out in the index of consciousness theories and what each predicts about AI.
The Position He Holds in the Other Debate
Tegmark is unusual in occupying two roles that rarely sit together. He is a consciousness theorist with a published physical framework, and he is one of the most prominent voices arguing that advanced AI poses a serious risk. Through the Future of Life Institute he has been central to open letters and campaigns urging caution about frontier systems.
Those two positions interact in a way worth stating, because it is the opposite of the usual pairing. Most people who take AI risk seriously either avoid the consciousness question or treat it as a distraction from alignment. Most people who take machine consciousness seriously are relatively sanguine about capability. Tegmark holds both, and his framework explains why they are compatible rather than in tension.
On the perceptronium account, consciousness is a structural property with specified conditions, and capability is a separate axis. A system can be dangerous without satisfying any of the five principles, because danger comes from what it optimises and how well, not from whether there is anything it is like to be it. Equally, a system could satisfy several principles without being especially capable. Treating the two as one question, which public discussion routinely does, is the error the framework is best positioned to correct.
That separation is also what makes his work relevant to welfare rather than only to safety. If the conditions for consciousness are structural and checkable, then a system could in principle acquire moral standing as a side effect of an architecture chosen for performance reasons, without anyone deciding to build a mind.
What Would Falsify It
The framework’s weakest point is that it is hard to say what evidence would count against it. The five principles are necessary conditions, and necessary conditions cannot be falsified by finding a system that satisfies them and seems unconscious, because sufficiency was never claimed.
What could count against it is narrower. If a system clearly lacking one of the five principles were established as conscious, the principle would fail. The trouble is that establishing consciousness independently is the problem the framework was supposed to help with, so the test cannot be run without assuming what it is meant to deliver.
This circularity is not unique to Tegmark and it is the standing condition of the field. It is the reason the indicator approach exists, since a list of graded properties drawn from several rival theories is a way of proceeding without any single theory being testable on its own.
Comparison to The Consciousness AI
The independence principle is the part with a direct bearing on this project’s architecture, described on the architecture page.
The system has a Simulation layer providing a body, an Affective Core generating homeostatic drives, and a Global Workspace running five to ten reentrant convergence cycles per decision. Those cycles are internal dynamics that continue between inputs, which is exactly the autonomy Tegmark’s principle asks for and which a feedforward pass does not have.
That is a structural match rather than evidence of anything further. The system satisfies a stated condition. Tegmark’s framework is explicit that the five principles are necessary rather than sufficient, and no measurement made here would bear on whether there is something it is like to be the system.
What Follows
The paper’s value is the reframing rather than the answer. Asking what physical properties conscious matter would need turns an intractable question into a list of structural conditions that can be checked, and several of them can be checked on artificial systems today.
What it does not do, and does not claim to do, is explain why matter organised that way would experience anything. Tegmark inherits that gap from IIT and states it plainly. Perceptronium names a substance whose defining property remains the thing nobody can derive.