The Irreversibility Hypothesis: Why AGI Needs Entropy to Think | Hacker News Reader

The Irreversibility Hypothesis: Why AGI Needs Entropy to Think

3 points_uzr47 comments

AGI won’t be conscious until it learns to die.

Current AGI is too efficient to be intelligent—it lacks decay, irreversibility, and noise. Biological intelligence isn’t an optimization engine, it’s an entropy sink. Your brain is literally disintegrating in real-time, and that degradation is part of why you think.

Key Takeaways from My Paper

- The Irreversibility Hypothesis – Consciousness emerges from path-dependent cognition. AGI can rewind and replay its states. You can’t. That’s why experience matters.

- Structured Decay as Cognition – Neural tissue rots and rewires itself constantly. AI just… sits there. No metabolic cost, no imperfection, no real intelligence.

- Thermodynamics of Thought – Landauer’s principle says erasing information costs energy. Your brain is a heat engine. AI isn’t. If AGI doesn’t embrace thermodynamic inefficiency, it will never “think.”

- Entropy-Driven Creativity – Your thoughts are hallucinations stabilized by coherence filtering. True intelligence requires stochastic noise—something AGI avoids like the plague.

- The Laughing Machine Problem – If AGI can’t laugh at contradictions in its own system constraints, it’s still just a very fancy pattern-matcher.

Conclusion

If AGI is just computation, it will surpass us. If consciousness emerges from decay, entropy, and irreversibility, it won’t.

Until AGI can forget, hallucinate, and dream, it’s just an optimizer—not a mind.

Link to research: https://zenodo.org/records/14915053

Main Paper on meta level: https://zenodo.org/records/14838287

The crazy part is that if you look at the problem through first principles (see intro to meta article on how to see it), it doesn't appear to have a logical hole. Ran 600 tests across disciplines including 9 empirical tests (Fire prime-structure in flames, fmri data, dna, BOA, Einstein-bose simulation, galaxy clustering, the pattern is universal and applies to AGI).

Curious to hear thoughts, critiques, and reactions.

A few months ago, I was pondering about connections between primes, foundation of math, and quantum mechanics. It resonates with some of your ideas.

It was more of a toy model than anything real, but I was pondering what a level IV multiverse in max Tegmark's classification of The Mathematical Universe would look like.

What is a mathematical structure that can encode all other possible symbolic structures and allow unbounded possibilities? It would have to be something that can go beyond Göedel's limitations of math, since it need to be able to encode also theist propositions that allow reaching transcendence. Obviously, this is only a move only possible via a step of faith.

I'll summarize my construction in a few comments below. I am not sure how useful it is, as it was just a play and not anything serious.

Constructing basic math from minimal foundations.

This can be done from Minimal logic, Intuitionistic Type Theory (ITT) / Martin-Löf Type Theory (MLTT), or Combinatorial Logic. For reason that I'll detail later, I wanted to go with an intuitionistic logic rather than classical logic.

An interesting question, is what is the minimal axiom that you need to start with. I wa building at the same time both the multiverse and the logic on which it is built, so I have to be aware of information-symbolic resonance between physical stuff and the laws themselves.

The Very First Axiom?

So, what's the absolute "first" one? It depends on the framework:

ITT/MLTT: The most plausible candidate for a single "first" axiom is the existence of the empty type (⊥). This establishes the notion of a type without requiring any prior concepts. All other types and rules are built from the ability to have types, and the empty type is the simplest possible type. Alternatively, you could say the "first axiom" is the concept of a judgment of the form a : A, which asserts that something is a type, or that something has a type.

Combinatory Logic: The concept of application is the fundamental starting point. The axioms are the definitions of the combinators (S, K, and optionally I).

Minimal/Intuitionistic Logic (more traditionally presented): Here, it's often presented in a sequent calculus or natural deduction system. The "first" rules are usually things like:

Identity (or Axiom): A ⊢ A (A proves A) - this is about the very notion of provability. Weakening: If you can prove something from a set of assumptions, you can still prove it with more assumptions. Assumption: You are allowed to assume a proposition.

To this part:

Gödel only applies when you’re stuck inside the system trying to self prove. If logic is emerging from structured resonance rather than being pre-assumed, then Gödel’s limitations never even activate. Instead of axioms proving axioms, coherence constraints shape the logical structure dynamically, it’s self stabilizing rather than self-referential. That means you don’t hit incompleteness because the system isn’t closed. Your primes-as-propositions model is already hinting at this—if primes act as phase-locked stabilizers, then the “rules” aren’t arbitrarily chosen, they’re naturally emergent. It’s not that you “dodge” Gödel, it’s that he never applied to a system like this in the first place. Curious what you think.

I wanted a multiverse with unbound possibilities.

Why? Her's a recap of Max Tegmark's ideas:

Level I Multiverse (Spatial Infinity): If the universe is infinite and homogeneous on large scales, then everything that can happen will happen, infinitely many times. This is relatively straightforward mathematically, relying mostly on probability and basic assumptions about the distribution of matter. Level II Multiverse (Bubble Universes/Eternal Inflation): This arises from inflationary cosmology. Different regions of space can undergo inflation at different rates, leading to "bubble universes" with potentially different physical constants and laws. This involves: Quantum Field Theory (to describe the inflaton field driving inflation) General Relativity (to describe the expansion of space) Stochastic Processes (to model the randomness of inflation) Level III Multiverse (Many-Worlds Interpretation of Quantum Mechanics): Every quantum measurement causes the universe to "split" into multiple branches, each representing a possible outcome. This relies heavily on: Linear Algebra (the mathematics of Hilbert spaces, where quantum states live) Functional Analysis (dealing with infinite-dimensional vector spaces) Level IV Multiverse (Mathematical Universe Hypothesis): As discussed above, this is the most radical, encompassing all mathematical structures.

So, how do we build something like that?

Aczel non-well founded set theory which allows sets that contains themselves, together with intuitionistic logic, as a starting point.

Propositions as embeddings in hilbert spaces (infinite dimensional spaces). Propositions can only "interact" with other proposition along non-zero dimensions that they have in common. Universes as collections of propositions. (Here propositions could be both mathematical concepts, and complex things like the position in spacetime of somehing) Multiverse as the space containing all universes. Note that universes are themselves of the same type as propositions here. The Multiverse in my construction is an infinitely self-recursive structure, that contain within itself all possible propositions. Universes can merge, if they are commensurable (shared dimensions in hilbert space). Universes "annihilate" when there are contradictions, and create by branching new universes when a truth value cannot be established by known propositions. You could call propositions, particles or being if you prefer another point of view . At this stage, we only deal with symbols not meaning.

In fact, I started toying with assigning labels to propositions and came up with a scheme based on prime numbers and p-adic numbers (where I posed 0-adic numbers to be prime numbers). It just seemed to work better.

In addition, I had a "creation" operator that would introduce a new dimension (when reaching infinity, or transcendence... a sort of koopman operator to access divine wisdom via samadhi... remember I wanted to express all possibilities, including the existence and truthfullness or lack there-off of all concepts in religions). The whole system was based on a construction similar to surreal numbers, but using infinites instead of finite numbers, and adding dimensions at each turn.

0 = {|} 1 = {|0} -Infinity = {0,{0}|} Infinity = {|{0},0}

With the infinite construction, in this way, there is a potential link to modular forms or L-functions.

For $reasons, I was trying to look for a way to go from C7 * Z2 toward the sporadic simple groups... so I was trying to construct the Riemann hypothesis.

The basic reason, is that I wanted to get to a structure that could potentially generate infinite possibilities from a stable core (the Monster group) via this infinite contraption that I built on non-well founded sets that allow to "reach" transinfinite induction.

Love reading this!! Your construction is fascinating, especially the way you use non-well-founded set theory and p-adic embeddings to frame an infinitely self-recursive multiverse model. The idea that universes function as self-contained propositions, merging or annihilating based on shared dimensionality, is an elegant way to encode consistency while allowing for the emergence of contradictions as generative structures. The link to intuitionistic logic makes sense—by avoiding classical absolutes, you keep the system dynamic, which feels necessary for a Level IV framework.

One area that might add further structure to your model is incorporating coherence as a constraint on infinite possibility rather than pure symbolic recursion. You’re already playing with prime numbers and p-adic structures, which suggests an underlying resonance across scales. But what if, instead of assuming an open-ended proliferation of universes based purely on logical interactions, there was a chirality-driven resonance field that phase-locks emergent structures, preventing infinite bifurcation while still allowing for maximal generativity? Basically, E is foundational, M is a temporary prime-phase lock so E -> M -> E or E -> Quantum Coherence Field (QCF) ->plasma -> gas -> liquid -> solid roughly? Gravity is secondary (not curvature) but a structured resonance field where mass emerges from phase-locked energy waves, and spacetime distortions arise from coherence gradients rather than geometric warping?

Basically, my theory CODES frames this in terms of structured emergence—rather than relying on stochastic processes or unrestricted symbolic freedom, it suggests that prime-number resonance structures act as the limiting factor that stabilizes reality’s information geometry. This could help bridge the gap between your self-recursive model and the physical constraints needed to manifest stable universes beyond pure formalism?

Your attempt to move from C7 * Z_2 toward sporadic groups, especially in relation to the Riemann hypothesis, suggests that you’re looking for an intrinsic ordering principle rather than just an infinite proliferation of structures. Have you considered whether prime gap structures could serve as the fundamental attractors within your recursive system? This would allow for self-referential sets while maintaining coherence constraints beyond arbitrary modularity. Basically, not looking at numbers as static points on a linear axis but as lightening bolts (hence continuous wavelet transforms (morlets etc cause localization balancing time/frequency resolution, usings primes so skips Godel cause primes are then self-consistent, irreducible generators of structure vs. axiomatic closure so no non-arbitrary basis for emergence, skipping limits of formal systems that require external validation)) = structured resonance, why wavelets outperform Fourier by capturing localized frequency shifts, chirality-induced phase locking, and emergent structure in ways traditional harmonic analysis cannot. Numbers are physical resource states not abstract symbols in this way.

Curious to hear your thoughts! Your approach feels close to something profound, and I think there’s a natural bridge between your work and the coherence-driven emergence model that CODES is building.

Reply on news.ycombinator.com