Two hands, each drawing the other.

That is Escher’s Drawing Hands, 1948, and it is the picture everyone reaches for when they want to say something about self-reference. A right hand holds a pencil and draws the cuff of a left hand, which holds a pencil and draws the cuff of the right. Neither is more real. Neither came first. You can trace the causality around that loop as long as you like and you never find the bottom.¹

Now listen to something. Bach’s Canon a 2 per tonos, from the Musical Offering. It modulates up a whole tone every time it repeats, and it keeps rising, and it keeps rising… and after six passes it is back where it started, one octave up.² Bach wrote on the manuscript that as the modulation rises, so may the glory of the King rise. He knew exactly what he had built.

If you want the pure version of that trick with the music taken away, listen to a Shepard tone. It is a stack of sine waves an octave apart under a fixed bell-shaped volume envelope, and it climbs forever without going anywhere, because pitch has two components and only one of them is actually moving. Chroma goes around the circle of twelve. Height stays put. Your ear only tracks the first one and gets fooled.³

And then there is the sentence. Take a formal system strong enough to do arithmetic, and you can build inside it a statement that amounts to: this statement has no proof here. If the system proves it, the system has proved something false. So it doesn’t. Which is what the statement said. Gödel, 1931.⁴

Four things. A drawing, a canon, an audio illusion, a theorem about numbers. What could a woodcut possibly have in common with a proof about arithmetic?

To be honest, the first time I met these as a set, the claim that they were all “the same thing” felt like something you say at 2am. Pretty. Not load-bearing.

I was wrong, and the reason I was wrong is more interesting than the claim. They really are one construction. But not the construction most people think, and the honest version of the unification is simultaneously stronger and narrower than the poetic one. It also happens to explain why two very smart people can stare at Gödel’s theorem and walk away concluding, respectively, that your mind is obviously a machine and that your mind obviously isn’t.

Usual warning. I’m a systems guy, not a logician and definitely not a physicist. I read well past my depth for this one and checked sources harder than I normally would, because this is a topic where being sloppy makes you sound profound. That is exactly the failure mode I want to avoid.

What Hofstadter actually said

The phrase is Douglas Hofstadter’s, from Gödel, Escher, Bach in 1979, and he spent the next thirty years mildly annoyed that people took the wrong thing from it.⁵

Here is his own definition, from the later book:

What I mean by “strange loop” is… not a physical circuit but an abstract loop in which, in the series of stages that constitute the cycling-around, there is a shift from one level of abstraction (or structure) to another, which feels like an upwards movement in a hierarchy, and yet somehow the successive “upward” shifts turn out to give rise to a closed cycle. That is, despite one’s sense of departing ever further from one’s origin, one winds up, to one’s shock, exactly where one had started out. In short, a strange loop is a paradoxical level-crossing feedback loop.⁶

Two words in there carry the whole thing: level-crossing.

Why isn’t a thermostat a strange loop? It is a feedback loop, clearly. It reads a temperature, compares, flips a relay, the temperature changes, round it goes… and nothing in that loop ever climbs a hierarchy. It is all one level, forever. That is the whole difference.

Strangeness needs a hierarchy that feels like it has a direction, up, and a path through it that keeps going up and still lands you where you began. Escher’s staircase in Ascending and Descending is the cleanest picture of it: every individual step is genuinely a step up, and the staircase closes.¹ It is built on the Penrose stairs, published by L. S. and R. Penrose in 1958, which is going to be funny in about five thousand words. Local consistency, global impossibility. Every step checks out. The whole thing cannot exist.

Hofstadter’s word for a system containing one of these is a tangled hierarchy: a hierarchy with no well-defined top or bottom, because some path through it wraps.⁵

And here is the nuance that summaries always drop. Hofstadter is explicit that plain feedback is not enough. He spends pages on video feedback, an actual camera pointed at the monitor it is feeding, an infinite regress you can set up on a desk in five minutes, and says it is not yet a strange loop. No perception in it. No symbols. Nothing in the system that stands for the system.⁷ The loop has to pass through something that means something. Otherwise it is just a picture of a picture of a picture.

Gödel, Escher, Bach won the Pulitzer in 1980, and Hofstadter later complained that everyone read it as a fun grab bag about mathematics and computers and Zen koans, when the actual thesis was one narrow idea about how a self comes out of stuff that is not a self.⁵ He wrote I Am a Strange Loop in 2007 largely to say it again, louder.

The machinery, quickly

I’ve been through Gödel’s theorem on this blog twice already, in the undecidability post and then again while arguing with the Gita, so I’m not rebuilding it from scratch. Short version: any consistent, mechanically checkable system strong enough for basic arithmetic has true statements it cannot prove, and cannot prove its own consistency.⁸

What I skipped both times is the part that actually matters here, and it’s the part I find genuinely beautiful. How does a system that only knows how to talk about numbers get to talk about itself? Numbers don’t have opinions about proofs.

Two moves.

First, Gödel numbering. Encode every formula and every proof as a natural number, so syntax becomes arithmetic. “Is this a valid proof of that” becomes an ordinary checkable property of two integers. Now the system’s statements about numbers are secretly statements about the system.⁸

Second, the diagonal lemma, and this is the engine. It says: give me any property A you can write down, and I will hand you a sentence D such that the system itself proves

D if and only if A(the number that codes D).

Any property at all. Feed it “is not provable” and out drops G, provably equivalent to the claim that G’s own code has no proof.⁸

The self-reference is not a pun or a hack. It is a lemma. You turn a crank and it falls out… which, when that landed for me, was the moment this stopped feeling like a magic trick and started feeling like machinery.

One caveat, because the popular version overreaches and a lot of bad philosophy is built on the overreach. Saying that G “says of itself that it is unprovable” is a gloss. What gets proved is the biconditional. G does not contain a little arrow pointing back at G. It contains a number, and that number happens to code G, and the system can verify the correspondence.⁹ Sounds pedantic. Isn’t.

The elegance of Gödel’s swap deserves a pause. The Liar sentence, “this sentence is false,” is a paradox. It breaks things. Gödel replaced false with not provable, trading a semantic notion for a syntactic one, and the paradox became a theorem.⁸ Same shape, no contradiction, because “provable” is something the system can genuinely talk about and “true” is not.

Tarski proved the other half of that in 1936: arithmetical truth is not definable inside arithmetic at all.¹⁰ Turing did it again in 1936 with machines instead of sentences, and the halting problem drops out of a program that asks the halting-predictor about itself and then does the opposite.¹¹ All of it descends from Cantor in 1891, whose diagonal argument built a real number differing from the nth number in the nth decimal place.¹²

Same move, four times, across sixty years and three fields. So is that a coincidence? A family resemblance? Or is there something actually underneath it?

Turns out the answer is a theorem rather than a vibe.

One theorem, and it has two faces

In 1969 William Lawvere published a paper called “Diagonal arguments and cartesian closed categories.”¹³ It is short, it is category theory, and it is the bedrock under this entire essay.

Stripped of machinery, the statement is this. If you have a way of indexing all the functions from A to B using elements of A itself, in a suitably surjective sense, then every function from B to B has a fixed point. Something that maps to itself.

Read it backwards and it is the diagonal argument. If B carries a function with no fixed point, and negation on a two-element truth object is exactly such a function, then no such indexing can exist. That is Cantor. That is Russell. That is Gödel. That is Tarski.¹³

Lawvere derived all of those in the original paper. Yanofsky in 2003 rewrote the whole thing in plain sets and functions with no categories required, and added the Liar, Grelling’s paradox, Richard’s paradox and the halting problem to the pile.¹⁴ One dictionary. Objects are sentences or programs, the fixed-point-free map is negation, the indexing is evaluation. Swap the ingredients, get a different famous result.

Now the part I actually care about.

The theorem has two faces, and they are one construction read in opposite directions.

Run it with a fixed-point-free map like negation and you get a wall. No such surjection exists. The system cannot contain its own truth predicate. The sentence is unprovable. The program cannot be written. Limitation.

Run it with a reflexive object instead and you get the exact opposite: a fixed point is guaranteed to exist. Every function has one. This is Kleene’s recursion theorem. This is the Y combinator. This is why quines exist.¹⁵ Construction.

A wall, and a self. The same diagram. So which one do you get? It depends entirely on whether the thing you are diagonalising against has a fixed point or refuses to have one. That is the only difference. That is it.

Hold onto that. It is the whole essay.

Now the fence, and it is a real fence rather than a polite hedge. The unification is structural, not total. Lawvere’s theorem gives you the conceptual skeleton of the fixed point. It does not give you Gödel numbering, or the recursive representability of the provability predicate, or the substitution function. That machinery is genuinely hard, and the categorical statement presupposes it rather than supplying it.¹⁶ Whether the diagonal or the arithmetization is “the deep part” of Gödel’s theorem is a live disagreement between category theorists and logicians, and I am nowhere near qualified to settle it. Anyone telling you these are literally the same theorem is selling you the skeleton as the body.

Worth knowing too: this is not the consensus philosophical framing. The Stanford Encyclopedia’s entry on self-reference organises the paradoxes around Priest’s Inclosure Schema and never mentions Lawvere; John Bell has a separate algebraic abstraction that recovers Gödel and Tarski by a different route.¹⁷ Lawvere is one unification program among several. The one that clicked for me. Not the settled answer.

The best inoculation against mysticism here is Lawvere’s own stated motive. He said the point was

to demystify the incompleteness theorem of Gödel and the truth-definition theory of Tarski by showing that both are consequences of some very simple algebra in the cartesian-closed setting.¹³

Demystify. The man who found the one theorem under all the loops thought the interesting thing about it was how ordinary it turned out to be. He also mentioned finding it always hard to understand how Cantor’s mathematical theorem got re-christened as a “paradox” by Russell. That is the correct attitude, and it is the exact opposite of the one this material usually attracts.

The loop is not even the point

Here is where I have to break my own frame, because the reading kept insisting.

I started out thinking the invariant was the loop. The thing that comes back to itself. Escher’s hands, the rising canon, G pointing at G. Cyclic, closed, tidy.

It isn’t.

In 1985 Stephen Yablo built a paradox with no cycle in it at all. Take an infinite list of sentences S1, S2, S3… where each Sn says “every sentence after me is false.” Now go looking for the circle. Where is it? There isn’t one. The reference graph is just the ordinary less-than ordering on the natural numbers, which never loops. And the list is still paradoxical.¹⁸ Yablo built it precisely to show that self-reference is not essential to paradox.

Yanofsky’s generalisation says the same thing from the other end: the diagonal in the theorem can be replaced by any onto map. All you need is that every row gets at least one entry changed. Not that anything points at itself.¹⁴

So is Yablo’s paradox genuinely non-circular? Ah well… that fight is not over. Priest argued in 1997 that there is a hidden fixed point in it, so it is circular after all. Cook argued that if circularity is that ubiquitous it stops explaining anything. There are infinitary versions that seem to escape even Priest’s reading. It was still open as of last year.¹⁷ I’m not going to pretend it’s settled in my favour.

But the weaker claim survives all of that, and it is the one I’ll actually defend: the shape is the diagonal, and the literal loop is its most vivid special case.

Escher’s hands are a diagonal you can look at. Gödel’s sentence is a diagonal you can prove things about. Yablo’s list is a diagonal with the circle unrolled into an infinite staircase.

That is a smaller claim than “everything is a loop.” I think it is also a much better one, because it survives contact with the counterexamples instead of quietly stepping around them.

The loop you already ship

If you write code for a living, you have been using the positive face of this theorem for years without calling it that.

Start with the Y combinator. Y = λf.(λx.f(xx))(λx.f(xx)), and the property that matters is Y g = g (Y g).¹⁵ What is it actually for? It manufactures recursion in a language that has no way to name a function, which means no way for a function to call itself by name. You want self-reference, you don’t get a name, so you build the fixed point by hand. That is Lawvere’s positive face with the lambda calculus plugged in.

Then quines. A program that prints its own source without reading its own file. Every Turing-complete language has one, and that is not a curiosity, it is guaranteed by Kleene’s second recursion theorem.¹⁹ The word “quine” is Hofstadter’s, coined in GEB after W. V. O. Quine, whose own contribution was the sentence

“Yields falsehood when preceded by its quotation” yields falsehood when preceded by its quotation

which achieves full Liar-grade self-reference with no “this” anywhere in it.²⁰ That is the English version of Gödel’s trick exactly: refer to yourself by describing yourself instead of pointing.

Then the metacircular evaluator, an interpreter for a language written in that language, whose whole core is eval and apply calling each other. Brian Cantwell Smith pushed it further with 3-Lisp around 1982, making the “interpreter running on an interpreter running on an interpreter” tower explicit and giving you handles to reach up it.²¹

And then the one that should worry you. Ken Thompson’s Turing Award lecture, “Reflections on Trusting Trust,” 1984. Write a compiler that recognises when it is compiling itself and reinserts a backdoor. Compile it once. Now delete the malicious source. The compiler binary keeps reinfecting every future compiler built from provably clean source, forever, and no amount of reading the source will ever show it to you.²² Thompson’s conclusion: you can’t trust code you didn’t totally create yourself. That is a quine used as a weapon, and it is the single most practical consequence of self-reference I know. Has anyone actually solved it in the forty years since? Not really, no.

Schmidhuber’s Gödel machine takes the idea to its limit: a program that rewrites its own code the moment its proof searcher can prove the rewrite increases expected utility. There’s a Global Optimality Theorem attached. It has also never been fully implemented, and it runs straight into incompleteness and intractability, which is roughly what you’d expect.²³

And since everyone is going to ask about LLMs: yes, current models carry a functional self-model. A model can predict its own behaviour better than a different model trained on its outputs can.²⁴ Anthropic reported last year that Claude can sometimes detect a concept injected into its own activations, around twenty percent of the time.²⁵ Both results are carefully framed as functional, by their own authors, and I am going to be equally careful: this is a system representing its own states. It says precisely nothing about whether there is anything it is like to be that system. Keep that fence up. We’ll need it later.

Four ways to tame it, and the joke inside the fourth

Once you know the diagonal will produce a fixed point wherever it can reach, the obvious engineering question is how to stop it. Foundations of mathematics in the twentieth century is basically four answers to that question.¹⁷

Forbid it. Russell’s type theory. A set cannot contain itself, expressions are stratified into types, the vicious-circle principle outlaws the construction. It worked, at the cost of needing the axiom of reducibility, which Weyl called “a bold, an almost fantastic axiom,” and which, once you assume it, collapses the very hierarchy it was introduced to support.²⁶

Stratify it. Tarski’s answer. Truth for a language is definable only in a strictly stronger metalanguage. No language contains its own truth predicate, so the Liar cannot even be written down. Cost: an infinite tower of metalanguages, each one able to discuss the one below and mute about itself.¹⁰

Let it fall in a hole. Kripke, 1975. Build a partial truth predicate by transfinite iteration until it stops moving, and let the Liar land in a truth-value gap, neither true nor false.²⁷

Accept it. Dialetheism. Priest’s position: the Liar is a true contradiction, so use a paraconsistent logic where a contradiction doesn’t let you derive everything. You keep semantic closure and you give up consistency.²⁸

Now the joke. Kripke’s construction, the one designed to defuse the loop, works by iterating a monotone operator up to its least fixed point.²⁷ The cure is the disease. The technique that tames self-reference is the same fixed-point machinery that produces it.

And the punchline underneath the joke: Gödel’s diagonal lemma survives all four. Every one of these restricts what you can say directly, and the lemma never needed direct saying. It reaches self-reference indirectly, through numbering.¹⁷ You cannot syntax your way out of it.

Hofstadter’s actual bet

So: a system rich enough to describe itself grows fixed points whether you want them or not. Hofstadter’s bet is that you are one of them.

The claim in I Am a Strange Loop is that the “I” is a self-referential, self-perceiving pattern of symbols in a brain. The most central and complex symbol in there is the one that stands for the whole system, and the loop is that symbol perceiving and updating itself.²⁹ The self is real in the way a pattern is real, and it is causally efficacious, which he defends with his “careenium” thought experiment: the bottom level, though a hundred percent responsible for what happens, is nonetheless irrelevant to what happens.²⁹

His closing line, and I checked this one against the book because the version floating around online is misworded:

In the end, we self-perceiving, self-inventing, locked-in mirages are little miracles of self-reference.⁶

Mirage is his word, not a critic’s. He picked it on purpose.

Metzinger got somewhere similar from neuroscience: no selves exist as things, the brain runs a transparent phenomenal self-model, transparent meaning the system cannot experience it as a model, which is exactly why it doesn’t feel like one.³⁰ And the loop shows up as load-bearing across most of the serious scientific theories of consciousness. Global Neuronal Workspace has recurrent prefronto-parietal ignition. Lamme’s recurrent processing says the re-entrant sweep is what makes vision conscious. Predictive processing has the prediction-error loop. Higher-order theories have a state representing a state.³¹

That is a lot of convergence. But convergence on what, exactly? Every one of those theories needs a loop to explain how information gets integrated and made available for use. Not one of them needs a loop to explain why any of it is felt. Those are different jobs, and I want to be careful about how much weight the convergence can carry.

Hofstadter presents the Gödel-to-”I” link as an analogy and a bet. He explicitly disclaims having proved it. And there is a real disanalogy that Adam Westra put his finger on. Gödel, in Hofstadter’s own words,

carefully concocted a statement about numbers and revealed that, because of how he had designed it, it had a very strange alternate meaning.⁶

Carefully concocted. Engineered. Whereas Hofstadter needs the loop of selfhood to arise on its own, and says so:

Like Gödel’s strange loop, which arises automatically in any sufficiently powerful formal system of number theory, the strange loop of selfhood will automatically arise in any sufficiently sophisticated repertoire of categories, and once you’ve got self, you’ve got consciousness. Élan mental is not needed.⁶

One of those is a thing a genius built by hand. The other has to happen by itself in wet tissue. Calling them the same structure is a bet, and Hofstadter is honest that it’s a bet. Plenty of people repeating him are not.

Same theorem, opposite verdicts

Here is the thing that made me want to write this at all.

Take the Gödel material. Hand it to Douglas Hofstadter and he concludes: the mind is a strange loop, and strange loops are computational, so a mind is the kind of thing a machine can be.

Hand the identical material to John Lucas and Roger Penrose and they conclude: the mind is a strange loop, and that is exactly why it cannot be a machine.

The argument on the Penrose side goes like this. Suppose your mathematical reasoning is captured by some formal system F. Then there is a Gödel sentence G(F) that F cannot prove. But you, looking at F from outside, can see G(F) is true. So you are doing something F cannot do. So you are not F. And F was arbitrary. Therefore mathematical insight is not algorithmic.³² Lucas put it as bluntly as you can: Gödel’s theorem “proves that mechanism is false, that is, that minds cannot be explained as machines.”³²

It is a genuinely seductive argument. I found it very hard to answer the first time I met it. So where exactly does it go wrong?

It is broken, by wide consensus, and the break is precisely where the two faces of Lawvere’s theorem meet.

The Gödel reasoning does not give you “G is true.” It gives you the conditional: if F is consistent, then G(F) is true.³³ And F proves that conditional too. F is perfectly capable of the reasoning. What F cannot do is detach the consequent, because detaching it requires knowing F is consistent, and by the Second Incompleteness Theorem F cannot establish that about itself.

So now ask it again with the roles filled in. If F is you, can you establish your own consistency? Using what… your own reasoning? The very thing under question? You are in exactly the position F is in. The apparent gap between you and F was manufactured by handing you consistency for free and refusing it to F. Level the treatment and the gap closes.³³

There are more objections and they stack. Humans might well be inconsistent machines. Benacerraf’s version is sharper and funnier: I may well be a Turing machine, but I cannot ascertain which one, so I can’t build my Gödel sentence either.³⁴ Putnam reviewed Shadows of the Mind and wrote, of the errors he’d found, “These are the fallacies on which the whole book rests.”³⁵ Feferman catalogued the technical problems. Chalmers identified the soundness premise as the argument’s greatest vulnerability.³⁶

Two things I want to be careful about, though.

First, this is rejected by consensus, not proven false by acclamation. Lucas and Penrose are serious people who kept defending it against serious objections. Penrose’s second version in Shadows is a restructured reductio, not a rerun. The consensus is that it fails. That is not nothing, and it is also not a theorem.³³

Second, and this one should give the whole anti-mechanist tradition pause: Gödel himself never drew Penrose’s conclusion. In his 1951 Gibbs lecture he drew only a disjunction. Either the human mind infinitely surpasses any finite machine, or there exist absolutely unsolvable diophantine problems. Either. Or. He never claimed the theorems settle it.³⁷ The man had the result in his hands for twenty years and was more cautious about it than most people who cite him.

Penrose also needs a physics half, because the logic half only tells you what the mind isn’t. That is Orch-OR, developed with Stuart Hameroff: consciousness arising from quantum computation in neuronal microtubules, terminated by a gravitational “objective reduction” whose outcome is non-computable.³⁸ Two independent arguments, failing for two independent reasons, and Penrose needs both to stand.

The physics half is in worse shape than the logic half. Tegmark calculated decoherence times for microtubule superpositions at around 10⁻¹³ seconds against cognitive timescales of 10⁻² to 10⁰, which he summarised as the brain landing in the classical category by a margin exceeding ten orders of magnitude.³⁹ Hagan, Hameroff and Tuszyński replied that he had modelled the wrong object and re-derived something eight or nine orders longer, which is still short of where it needs to be.⁴⁰

And then there is a real experiment, which I like a lot because it moves this from argument to measurement. Donadi and colleagues went to Gran Sasso in 2021 and looked for the spontaneous radiation that gravity-related collapse predicts. They didn’t find it, and excluded the parameter-free Diósi-Penrose model.⁴¹

Be precise about what that means, because it gets overstated in both directions. It excludes the simplest version of gravitational collapse. It does not “debunk Orch-OR” and it does not rule out every gravity-collapse theory. Meanwhile the recent quantum-biology results people cite for the other side, the epothilone-B anaesthesia study and the tryptophan superradiance paper, are suggestive at best, and their own authors say so in the papers.⁴²

The honest scorecard: the logic argument fails on the consistency assumption, the physics argument is rejected on decoherence grounds and now constrained experimentally, and Penrose needs both. That is a losing position. It is not a stupid one.

But look at what the disagreement actually is. Hofstadter is reading the positive face of the diagonal: a fixed point exists, the self closes, the loop is constructible, therefore machine. Penrose is reading the negative face: here is a wall no formal system crosses, humans cross it, therefore not machine.

Same diagram. Two directions. And here is the part I find genuinely funny: both of them are reading it correctly. They each grabbed one face and let go of the other.

Maybe I’m being unfair to two people far smarter than me. But I don’t think that symmetry is an accident.

The most seductive loop of all

Now the big one, and I want to actually explore this rather than swat it, because the swatting version is lazy and the idea has better parents than its internet reputation suggests.

The idea: the universe is observing itself into existence. Reality doesn’t settle until it is looked at, and the looking is done by things inside the universe, so the universe is a strange loop, and that is where quantum weirdness comes from.

You have met this in some form. It is usually attached to Heisenberg. It is usually wrong in a specific way I’ll get to. But first, take it seriously, because John Archibald Wheeler took it seriously, and Wheeler was not a crank. He was Feynman’s supervisor.

Wheeler’s slogan was “it from bit”: every physical it derives its existence from apparatus-elicited yes-or-no answers.⁴³ Not “information is important to physics.” Information is the root of physics, and the material world is downstream.

He drew it as a self-excited circuit. Picture a capital U. The universe starts at one tip, evolves, produces observers partway along, and those observers turn and look back at the early universe, giving it its tangible reality. He meant this as a literal diagram, and it is a literal cosmic strange loop.⁴⁴

His parable for how this works is the surprise version of twenty questions. You leave the room, the group is supposed to pick a word, you come back and start asking yes/no questions. What you don’t know is that nobody picked anything. Each person answers however they like, subject only to having a consistent word available given every answer so far. The word gets determined by the questioning. Nobody chose it, and nobody made it up freely either.⁴⁴

That is a genuinely beautiful idea and I do not want to be sniffy about it.

And it is not alone. QBism, from Fuchs, Schack and Mermin, treats the quantum state as an agent’s personal degrees of belief and measurement as the agent acting on the world. Fuchs coined the phrase “participatory realism” as an explicit tribute to Wheeler.⁴⁵ Relational quantum mechanics, Rovelli 1996, says there is no observer-independent state at all: states are relations between systems.⁴⁶ Seth Lloyd argues the universe is a quantum computer computing its own evolution.⁴⁷ Tegmark’s mathematical universe casts observers as “self-aware substructures” of a mathematical reality.⁴⁸

Serious people. Serious journals. Nobody here is selling crystals.

But notice something about that entire list. What, in any of it, is actually doing the observing?

Not one of them requires a mind. QBism’s “agent” need not be conscious and the Stanford Encyclopedia says so directly.⁴⁵ In relational QM any physical system can be the observer, and Rovelli is emphatic that it is not mind-dependent.⁴⁶ Even Wheeler defined registration broadly enough to include a piece of mica recording a decay.⁴⁴ The participation is done by apparatus. Every time.

Where it breaks

So here is the honest landing… and it is narrower than the debunk you usually get, which is exactly why I want to do it carefully instead of triumphantly.

The uncertainty principle has nothing to do with observation. This is the load-bearing correction. The standard relation, Kennard 1927 and Robertson 1929, is a theorem about state preparation. It says a single prepared quantum state cannot have arbitrarily sharp values of two incompatible quantities at once. It holds with no measurement, no observer, and no disturbance anywhere in the picture.⁴⁹ It is a statement about what states are, in the same way that a short pulse cannot have a sharp frequency. Nobody has to look.

Where does the popular version come from? From Heisenberg himself, honestly. His 1927 gamma-ray microscope was a heuristic about measurement disturbance via Compton recoil, Bohr objected almost immediately, and Heisenberg revised.⁴⁹ The disturbance picture is the origin of the misconception and it got frozen into pop science.

Now, be careful here, because there is an overcorrection that is also wrong. Three separate layers:

1. The intrinsic preparation relation. Undisputed, no observer needed.
2. The naïve measurement-disturbance product. Ozawa showed in 2003 that this has no universal positive lower bound, and experiments in 2012 by Rozema and by Erhart confirmed you can violate the naïve relation while the correct bound holds.⁵⁰⁵¹
3. What the correct disturbance metric should be. Still an open argument between Ozawa and the Busch-Lahti-Werner camp.⁴⁹

So “observation causes uncertainty” is wrong, and “measurement disturbance is dead” is also wrong. Layer 1 was never about observation. Layer 2’s naïve form is genuinely dead. Layer 3 is live research.

Two more corrections while I’m here.

The “observer” in quantum mechanics is not a mind. Any decoherence-inducing interaction with a macroscopic apparatus does the job. Bohr’s own framing put man, animal and apparatus on the same footing.⁵² The consciousness-causes-collapse view is a fringe interpretation, it was built on von Neumann by London, Bauer and Wigner rather than by von Neumann, and Wigner himself abandoned it, calling his earlier position solipsism, after Zeh’s work on decoherence.⁵³ The person most associated with the idea talked himself out of it.

Delayed choice does not rewrite the past. Wheeler’s delayed-choice experiment was actually run, by Jacques and colleagues in 2007 with single photons, and standard quantum mechanics accounts for the result with no backward causation.⁵⁴ The quantum eraser is the version that goes viral, and the misreading is always the same: the interference pattern only appears once you sort the signal photons by their idler outcomes. Read the correlation as retrocausation and you have fooled yourself with post-selection.⁵⁵

The measurement problem is real, mind. Where the quantum-to-classical cut sits is a genuine open problem in foundations. But the cut is movable, the predictions don’t depend on where you put it, and none of the mainstream interpretations needs a conscious observer to make it work.⁵⁶

Where the loop actually lives in physics

Which would be a deflating place to stop. Except it isn’t the end, because rigorous self-reference genuinely is in physics. It is just somewhere else, and it is stranger than the version people wanted.

One. In 2015 Cubitt, Pérez-García and Wolf proved that the spectral gap problem is undecidable.⁵⁷ Given a description of a quantum many-body system, determine whether it has an energy gap above its ground state. There is no algorithm. Not “we don’t have one.” There cannot be one. They proved it both by reduction from halting and in the Gödel-style axiom-independence form: there are systems whose gappedness is independent of the axioms of mathematics.

A physical property of a physical system, undecidable. That is the real thing, sitting in a real journal.

And immediately, the fence, because this one gets stretched further than any other result in this essay. The Hamiltonian family is specifically constructed to embed a Turing machine in a 2D lattice. It says nothing about generic quantum systems, nothing about real materials, and it absolutely does not show that “physics is undecidable” or that a theory of everything is impossible.⁵⁷ It shows the property is undecidable in general, which is a statement about the worst case in an engineered family. Still remarkable. Not a licence.

Two, and this is the one I keep thinking about. David Wolpert, 2008, “Physical limits of inference.”⁵⁸

Wolpert did something I find genuinely clever: he modelled observation, prediction, memory and control as one abstract object, an inference device, defined without reference to any particular physical law. Then he proved impossibility results about it that he describes as similar in character to the halting theorem, and which hold whatever the laws of physics turn out to be.

The headline: no universe can contain more than one strong inference device. And two inference devices cannot each fully infer the other.

Sit with that one for a second. Is it a statement about instruments being clumsy? About disturbing the thing you measure? No. It is a structural limit on inference itself, holding across all possible physical laws. Two parts of one universe cannot each hold a complete model of the other. Something has to give.

I have to be careful with this one too, and it is the correction I was most grateful to find, because I had the wrong version in my head for weeks. Wolpert explicitly says his results do not rely on self-reference. The headline theorem is about two distinct devices, not about one device containing itself.⁵⁸ So the sentence I wanted to write, “the universe can’t fully know itself because it contains itself,” is not what he proved. What he proved is that two different knowers inside one universe cannot each fully know the other.

Which is, if anything, weirder… and I’ve come around to preferring it. Not one thing failing to swallow itself. Two things that can never fully contain each other.

So the wall is real. It is just not where the poetry wanted it to be. It is not in the uncertainty principle, which was never about observation. It is in what one part of a universe can ever infer about another part, and it is a theorem.

Where this leaves me

The minimal claim that survives every objection in this essay is smaller than the one I started with, and I’ve come to like it more.

Self-reference is plausibly necessary for a self-model, and a self-model is a real, causally relevant, non-fundamental pattern. The self is a verb. That much I’ll defend.

Is it sufficient for felt experience? No, it has not been shown to be, and I don’t think the strange loop can carry that weight. Chalmers’s easy/hard split is the sharp instrument here, and what makes it sharp is that his “easy problem” list explicitly includes the integration of information and access to internal states.⁵⁹ Which is exactly what a strange loop explains. Block’s distinction makes the same point in different vocabulary: you get access-consciousness, information globally available for reasoning and report, and phenomenal consciousness stays open.⁶⁰

The loop explains the sense of self. It does not explain why there is anything it is like to have one.

Martin Gardner made the objection cleanly reviewing I Am a Strange Loop: “the I is a strange loop” describes self-modelling rather than explaining consciousness, and a vivid redescription is not a mechanism.⁶¹ I think that’s right, and I think Hofstadter’s honest answer is to deny the question, which he does. On his view phenomenal consciousness is itself the mirage, zombies are incoherent, believing is feeling, there is no extra fact over and above the loop.²⁹ That is a coherent position. It is not a solution to the hard problem, it is a refusal of it, and you should notice which one you’re being handed.

Three things I’d ask anyone writing about this to stop saying, and I say them together because they fail for three unrelated reasons:

• Observation does not cause Heisenberg uncertainty. Wrong on the physics.
• Gödel does not prove the mind is non-computable. The argument is conditional, and it fails on consistency.
• The universe does not need a conscious observer. Fringe, and its own author recanted.

None of those is a close call, and all three are load-bearing for the version of this story that sells books.

What’s left after you clear them out is still, to me, remarkable.

There is one construction. Cantor found it, Russell tripped over it, Gödel weaponised it, Tarski extended it, Turing rebuilt it in machines, Lawvere wrote down the general form, and every one of them is the same crank turned with different ingredients. Read it one way and you get a wall that no formal system crosses. Read it the other and you get a fixed point, a thing that closes on itself, which is how recursion exists and how a quine prints itself and, if Hofstadter’s bet lands, how there comes to be a you in there at all.

Escher’s hands and Gödel’s sentence really are the same shape. The shape just isn’t the loop. It is the diagonal, and the loop is the special case you can see.

And out at the edge, Wolpert’s theorem says something I still can’t quite get to sit still: in any universe at all, whatever its laws, two knowers inside it cannot each fully know the other. Not because measurement is clumsy. Because of the shape.

Which is, I think, the honest version of “the universe cannot fully know itself.” Not a mystical closing of the circle. A wall, in exactly the place the mathematics says there has to be one, and nowhere else.

I’ll take the wall. It’s the one part of this I’m confident about, and I’ve had to give up more of the pretty version than I expected to when I started.

References

Footnotes

1. M. C. Escher’s Drawing Hands (1948), Ascending and Descending (1960) and Waterfall (1961). Ascending and Descending is built on the Penrose stairs, published by L. S. and R. Penrose in the British Journal of Psychology, February 1958. https://en.wikipedia.org/wiki/Ascending_and_Descending ↩ ↩²

2. J. S. Bach, Canon a 2 per tonos, BWV 1079/8, from The Musical Offering: it modulates up a whole tone each repetition and returns to the starting key an octave higher after six passes. https://www.bachvereniging.nl/en/bwv/bwv-1079-8 ↩

3. The Shepard tone: octave-spaced sinusoids under a fixed log-frequency envelope, so chroma advances while height stays statistically constant. Risset’s continuous version is the Shepard-Risset glissando. https://en.wikipedia.org/wiki/Shepard_tone ↩

4. Kurt Gödel, “Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I,” Monatshefte für Mathematik und Physik 38 (1931), pp. 173–198. https://link.springer.com/article/10.1007/BF01700692 ↩

5. Douglas Hofstadter, Gödel, Escher, Bach: An Eternal Golden Braid (Basic Books, 1979), winner of the 1980 Pulitzer Prize for General Nonfiction. The “tangled hierarchy” is his term; the 20th-anniversary preface frames the book’s real subject as how animate beings can come out of inanimate matter. https://www.hachettebookgroup.com/titles/douglas-r-hofstadter/godel-escher-bach/9780465026562/ ↩ ↩² ↩³

6. Douglas Hofstadter, I Am a Strange Loop (Basic Books, 2007). The strange-loop definition, the “locked-in mirages” line, the “carefully concocted” remark about Gödel, and the “Élan mental is not needed” passage are all quoted verbatim from the book. https://www.hachettebookgroup.com/titles/douglas-r-hofstadter/i-am-a-strange-loop/9780465030798/ ↩ ↩² ↩³ ↩⁴

7. Hofstadter’s video-feedback chapter in I Am a Strange Loop: physical camera-to-monitor feedback produces infinite regress but is not a strange loop, because it lacks perception and symbols. https://en.wikipedia.org/wiki/I_Am_a_Strange_Loop ↩

8. “Gödel’s Incompleteness Theorems,” Stanford Encyclopedia of Philosophy: the two theorems, Gödel numbering, the diagonal lemma, and the Liar-to-theorem swap. The base theory needs only Robinson arithmetic Q, not full Peano arithmetic. https://plato.stanford.edu/entries/goedel-incompleteness/ ↩ ↩² ↩³ ↩⁴

9. The same SEP entry notes that “the Gödel sentence says of itself that it is unprovable” is a heuristic gloss: what is proved is the material biconditional, not literal self-mention. https://plato.stanford.edu/entries/goedel-incompleteness/ ↩

10. “Tarski’s Truth Definitions,” Stanford Encyclopedia of Philosophy: arithmetical truth is not definable within arithmetic, and truth for a language is definable only in an essentially stronger metalanguage. Tarski’s original is “Der Wahrheitsbegriff in den formalisierten Sprachen,” Studia Philosophica 1 (1936), pp. 261–405. https://plato.stanford.edu/entries/tarski-truth/ ↩ ↩²

11. Alan Turing, “On Computable Numbers, with an Application to the Entscheidungsproblem,” Proceedings of the London Mathematical Society s2-42 (1936), pp. 230–265. https://academic.oup.com/plms/article-abstract/s2-42/1/230/1491926 ↩

12. Cantor’s 1891 diagonal argument, the common ancestor of the Gödel and Turing constructions. https://en.wikipedia.org/wiki/Cantor%27s_diagonal_argument ↩

13. F. William Lawvere, “Diagonal arguments and cartesian closed categories” (1969), reprinted in Reprints in Theory and Applications of Categories 15 (2006), with Lawvere’s own commentary containing the “demystify” quotation and the remark about Cantor’s theorem being re-christened as Russell’s paradox. https://www.tac.mta.ca/tac/reprints/articles/15/tr15.pdf ↩ ↩² ↩³

14. Noson Yanofsky, “A Universal Approach to Self-Referential Paradoxes, Incompleteness and Fixed Points,” Bulletin of Symbolic Logic 9 (2003): Lawvere’s theorem restated in plain sets and functions, covering Cantor, Russell, Gödel, Tarski, the Liar, Grelling, Richard and the halting problem. Theorem 2 generalises the diagonal to any onto map. https://arxiv.org/abs/math/0305282 ↩ ↩²

15. The Y combinator and fixed-point combinators generally: Y g = g (Y g), giving recursion without named self-reference. For Curry’s Y the two sides are beta-equivalent; Turing’s Θ beta-reduces. https://en.wikipedia.org/wiki/Fixed-point_combinator ↩ ↩²

16. On the limits of the categorical unification, and on Lawvere’s fixed-point theorem generally, see the nLab entry. The categorical statement captures the fixed-point core but presupposes the arithmetization (Gödel numbering, recursive representability, the substitution function) that carries Gödel’s hard content. https://ncatlab.org/nlab/show/Lawvere%27s+fixed+point+theorem ↩

17. “Self-Reference,” Stanford Encyclopedia of Philosophy: organises the paradoxes around Priest’s Inclosure Schema, surveys the Yablo circularity dispute (Priest 1997 against Yablo, Cook’s reply, the infinitary versions) as unresolved, and covers the four engineered responses. John Bell’s independent algebraic abstraction is in “Incompleteness in a General Setting.” https://plato.stanford.edu/entries/self-reference/ ↩ ↩² ↩³ ↩⁴

18. Stephen Yablo’s paradox (1985, 1993): an infinite acyclic list of sentences, each asserting that all later ones are false, constructed to show that self-reference is not essential to paradox. Discussed in the SEP self-reference entry above and in Yanofsky’s paper. https://arxiv.org/abs/1303.0730 ↩

19. Kleene’s second recursion theorem (1938), which guarantees quines in any Turing-complete language and is the computability-theoretic twin of the Y combinator. https://en.wikipedia.org/wiki/Kleene%27s_recursion_theorem ↩

20. Quine’s paradox: “‘Yields falsehood when preceded by its quotation’ yields falsehood when preceded by its quotation,” achieving Liar-style self-reference with no demonstrative. The term “quine” for a self-printing program is Hofstadter’s coinage in GEB. https://en.wikipedia.org/wiki/Quine%27s_paradox ↩

21. The metacircular evaluator, SICP §4.1: an interpreter for Lisp written in Lisp, whose core is the mutual recursion of eval and apply. Brian Cantwell Smith’s 3-Lisp (c. 1982) makes the reflective tower explicit. https://sarabander.github.io/sicp/html/4_002e1.xhtml ↩

22. Ken Thompson, “Reflections on Trusting Trust,” Turing Award lecture, Communications of the ACM 27 (1984): a compiler that recognises when it is compiling itself reinserts a Trojan, so the bug survives in the binary against provably clean source. https://dl.acm.org/doi/10.1145/358198.358210 ↩

23. Jürgen Schmidhuber, “Gödel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements”: rewrites its own code once a proof searcher proves the rewrite raises expected utility. Limited by incompleteness and intractability, and never fully implemented. https://people.idsia.ch/~juergen/goedelmachine.html ↩

24. Binder et al., “Looking Inward: Language Models Can Learn About Themselves by Introspection” (2024): a model self-predicts better than a separate model trained on its outputs. The authors frame this as functional self-knowledge, not phenomenal consciousness. https://arxiv.org/abs/2410.13787 ↩

25. Anthropic, “Emergent introspective awareness in large language models” (2025): Claude can sometimes detect a concept injected into its activations, roughly twenty percent of the time. Explicitly framed as functional. https://www.anthropic.com/research/introspection ↩

26. “Russell’s Paradox” and “Type Theory,” Stanford Encyclopedia of Philosophy: the vicious-circle principle, the ramified hierarchy, and the axiom of reducibility that Weyl called “a bold, an almost fantastic axiom” and that collapses the hierarchy once assumed (Ramsey). https://plato.stanford.edu/entries/russell-paradox/ ↩

27. Saul Kripke, “Outline of a Theory of Truth” (1975): a partial truth predicate built by transfinite iteration of a monotone operator up to a least fixed point, with the Liar landing in a truth-value gap. The cure is itself a fixed-point construction. https://en.wikipedia.org/wiki/Kripke%27s_theory_of_truth ↩ ↩²

28. “Dialetheism,” Stanford Encyclopedia of Philosophy: Priest’s position that the Liar is a true contradiction, requiring a paraconsistent logic in which explosion fails. https://plato.stanford.edu/entries/dialetheism/ ↩

29. Hofstadter’s positive thesis in I Am a Strange Loop: the “I” as a self-perceiving symbol pattern, the causal potency of patterns (“the bottom level, though 100 percent responsible for what is happening, is nonetheless irrelevant to what happens”), and his deflationary treatment of phenomenal consciousness as a real but unavoidable mirage. https://en.wikipedia.org/wiki/I_Am_a_Strange_Loop ↩ ↩² ↩³

30. Thomas Metzinger, Being No One: The Self-Model Theory of Subjectivity (MIT Press, 2003): no selves exist as things; the brain runs a transparent phenomenal self-model that the system cannot experience as a model. https://mitpress.mit.edu/9780262633086/being-no-one/ ↩

31. Mashour, Roelfsema, Changeux and Dehaene, “Conscious Processing and the Global Neuronal Workspace Hypothesis,” Neuron 105 (2020): non-linear ignition sustained by long-range recurrent prefronto-parietal loops. On the contested state of the field, note the 2023 open letter calling Integrated Information Theory pseudoscience and the mixed verdicts of the preregistered adversarial Cogitate test (Nature, 2025), which corroborated some IIT predictions and contradicted others while also challenging Global Workspace predictions. https://www.cell.com/neuron/fulltext/S0896-6273(20)30052-0 ↩

32. J. R. Lucas, “Minds, Machines and Gödel,” Philosophy 36 (1961), where he claims Gödel’s theorem “proves that mechanism is false, that is, that minds cannot be explained as machines.” Extended by Roger Penrose in The Emperor’s New Mind (1989) and Shadows of the Mind (1994). https://en.wikipedia.org/wiki/Minds,_Machines_and_G%C3%B6del ↩ ↩²

33. The SEP incompleteness entry’s section on anti-mechanist arguments: the Gödel sentence is true only given the system’s consistency, and by the Second Theorem no such system establishes its own consistency, so the argument assumes what it needs to prove. The conclusion is rejected by wide consensus, which is not the same as being proven false. https://plato.stanford.edu/entries/goedel-incompleteness/ ↩ ↩² ↩³

34. “The Lucas-Penrose Argument about Gödel’s Theorem,” Internet Encyclopedia of Philosophy: the rebuttal roster, including Benacerraf’s disjunction that one may be a Turing machine but cannot ascertain which one. https://iep.utm.edu/lp-argue/ ↩

35. Hilary Putnam, “The Best of All Possible Brains?”, review of Shadows of the Mind, Bulletin of the American Mathematical Society 32 (1995): “These are the fallacies on which the whole book rests.” https://www.ams.org/journals/bull/1995-32-03/S0273-0979-1995-00606-3/S0273-0979-1995-00606-3.pdf ↩

36. David Chalmers, “Minds, Machines, and Mathematics,” review of Shadows of the Mind (1995): the soundness premise is the argument’s greatest vulnerability. Solomon Feferman’s technical catalogue is in “Penrose’s Gödelian Argument,” PSYCHE (1995). https://consc.net/papers/penrose.html ↩

37. Solomon Feferman, “Are There Absolutely Unsolvable Problems? Gödel’s Dichotomy,” on the 1951 Gibbs lecture, in which Gödel drew only a disjunction: either the human mind infinitely surpasses any finite machine, or there exist absolutely unsolvable diophantine problems. https://math.stanford.edu/~feferman/papers/dichotomy.pdf ↩

38. Hameroff and Penrose, “Consciousness in the universe: A review of the ‘Orch OR’ theory,” Physics of Life Reviews 11 (2014): quantum computation in microtubules terminated by gravitational objective reduction with a non-computable outcome. https://pubmed.ncbi.nlm.nih.gov/24070914/ ↩

39. Max Tegmark, “The importance of quantum decoherence in brain processes,” Physical Review E 61 (2000): decoherence timescales for microtubule and neuron superpositions far shorter than cognitive dynamics, placing the brain in the classical category by more than ten orders of magnitude. https://arxiv.org/abs/quant-ph/9907009 ↩

40. Hagan, Hameroff and Tuszyński, “Quantum computation in brain microtubules? Decoherence and biological feasibility,” Physical Review E 65 (2002): a reply arguing Tegmark modelled the wrong object, re-deriving decoherence times eight to nine orders longer, still short of neural timescales. https://doi.org/10.1103/PhysRevE.65.061901 ↩

41. Donadi et al., “Underground test of gravity-related wave function collapse,” Nature Physics 17 (2021): a Gran Sasso search for the spontaneous radiation predicted by gravity-related collapse, excluding the parameter-free Diósi-Penrose model. It constrains the simplest model, not every gravity-collapse theory. https://arxiv.org/abs/2111.13490 ↩

42. Two recent quantum-biology results often cited in this debate, both hedged by their own authors: the epothilone-B anaesthesia study, whose authors call the result potentially consistent with classical models, and the tryptophan superradiance paper, which makes no claim about consciousness or Orch-OR. https://pmc.ncbi.nlm.nih.gov/articles/PMC11363512/ ↩

43. John Archibald Wheeler, “Information, Physics, Quantum: The Search for Links” (1990), the source of “it from bit.” https://philpapers.org/rec/WHEIPQ ↩

44. Wheeler’s self-excited circuit, the participatory universe, and the surprise version of twenty questions, in “Information, Physics, Quantum” and in Frontiers of Time (1979). Wheeler defined registration broadly enough to include inanimate apparatus, so his own framing needs no consciousness. https://www.quantamagazine.org/john-wheeler-saw-the-tear-in-reality-20240925/ ↩ ↩² ↩³

45. “Quantum-Bayesian and Pragmatist Views of Quantum Theory,” Stanford Encyclopedia of Philosophy, which notes that QBism’s agent need not be a conscious human. Christopher Fuchs coined “participatory realism” as a tribute to Wheeler in “On Participatory Realism” (2016). https://plato.stanford.edu/entries/quantum-bayesian/ ↩ ↩²

46. “Relational Quantum Mechanics,” Stanford Encyclopedia of Philosophy: Rovelli’s 1996 view that there is no observer-independent quantum state, that any physical system can be the observer, and that the view is explicitly not mind-dependent. https://plato.stanford.edu/entries/qm-relational/ ↩ ↩²

47. Seth Lloyd, Programming the Universe (2006): the universe as a quantum computer computing its own evolution. Serious but contested speculation. https://en.wikipedia.org/wiki/Programming_the_Universe ↩

48. Max Tegmark, “The Mathematical Universe,” Foundations of Physics 38 (2008): observers as self-aware substructures of a mathematical reality. https://arxiv.org/pdf/0704.0646 ↩

49. “The Uncertainty Principle,” Stanford Encyclopedia of Philosophy: the Kennard (1927) and Robertson (1929) relations as theorems about state preparation, holding with no measurement or observer; Heisenberg’s 1927 gamma-ray-microscope disturbance heuristic and Bohr’s objection; and the unresolved Ozawa versus Busch-Lahti-Werner dispute over the correct disturbance metric. https://plato.stanford.edu/entries/qt-uncertainty/ ↩ ↩² ↩³

50. Masanao Ozawa, “Universally valid reformulation of the Heisenberg uncertainty principle,” Physical Review A 67 (2003): the naïve measurement-disturbance product has no universal positive lower bound. https://doi.org/10.1103/PhysRevA.67.042105 ↩

51. Rozema et al., “Violation of Heisenberg’s Measurement-Disturbance Relationship by Weak Measurements,” Physical Review Letters 109 (2012), and Erhart, Hasegawa et al., “Experimental demonstration of a universally valid error-disturbance uncertainty relation in spin measurements,” Nature Physics 8 (2012). https://arxiv.org/abs/1208.0034 ↩

52. On “observer” and “measurement” in quantum mechanics requiring no conscious mind, and on Bohr’s placing of man, animal and apparatus on the same footing, see the SEP uncertainty entry and the history of the consciousness-causes-collapse interpretation. https://en.wikipedia.org/wiki/Consciousness_causes_collapse ↩

53. The von Neumann-Wigner interpretation was built on von Neumann by London and Bauer (1939) and Wigner (1961), not by von Neumann himself; Wigner later abandoned it, calling his earlier position solipsism, after Zeh’s 1970 work on decoherence. Surveys of foundational attitudes find it holds a small minority of physicists. https://arxiv.org/abs/1301.1069 ↩

54. Jacques et al., “Experimental Realization of Wheeler’s Delayed-Choice Gedanken Experiment,” Science 315 (2007): fully explained by standard quantum mechanics with no backward causation. https://arxiv.org/abs/quant-ph/0610241 ↩

55. On why the delayed-choice quantum eraser does not show retrocausality: interference appears only after conditioning signal photons on idler outcomes. See Sean Carroll’s account and Ellerman, “Why delayed choice experiments do not imply retrocausality,” Quantum Studies (2015). https://www.preposterousuniverse.com/blog/2019/09/21/the-notorious-delayed-choice-quantum-eraser/ ↩

56. On the Heisenberg cut being movable with outcomes independent of its placement, and on no mainstream interpretation requiring a conscious observer. https://en.wikipedia.org/wiki/Heisenberg_cut ↩

57. Cubitt, Pérez-García and Wolf, “Undecidability of the spectral gap,” Nature 528 (2015): undecidable both by reduction from halting and in the axiom-independence sense, for specifically constructed translationally-invariant 2D lattice Hamiltonian families that embed a Turing machine. It says nothing about generic systems or real materials. https://arxiv.org/abs/1502.04135 ↩ ↩²

58. David Wolpert, “Physical limits of inference,” Physica D 237 (2008): observation, prediction and recollection modelled as one “inference device,” with impossibility results the author describes as similar to the halting theorem and holding independent of the physical laws. No universe contains more than one strong inference device, and no two devices can each fully infer the other. Wolpert states explicitly that the results do not rely on self-reference. https://arxiv.org/abs/0708.1362 ↩ ↩²

59. David Chalmers, “Facing Up to the Problem of Consciousness,” Journal of Consciousness Studies 2 (1995). His easy-problem list explicitly includes the integration of information and access to internal states, which is what a strange loop explains, leaving the hard problem untouched. https://consc.net/papers/facing.pdf ↩

60. Ned Block, “On a Confusion about a Function of Consciousness,” Behavioral and Brain Sciences 18 (1995): the access versus phenomenal consciousness distinction. https://philpapers.org/rec/BLOOAC ↩

61. Martin Gardner, “Do Loops Explain Consciousness? Review of I Am a Strange Loop,” Notices of the AMS 54 (2007): the loop describes self-modelling rather than explaining consciousness. https://www.ams.org/notices/200707/tx070700852p.pdf ↩