New to this research? This article is part of the Reflexive Reality formal research program. Brief introduction ↗ · Full research index ↗

Series: NEMS on AI Safety · Part 1 · Part 2 · Part 3 · Part 4 · Part 5 · Part 6: Beyond the Abstraction Fallacy

---

A Google DeepMind researcher recently published one of the most-read papers in the current AI consciousness debate, arguing that computation is a “mapmaker-dependent description” that can never instantiate genuine experience — only simulate it. The intuition is correct. But philosophical argument is not formal proof. This essay explains what machine-checked theorems actually establish, where they go further than the argument, and what the positive theory looks like that the debate has been missing.

---

The Paper Everyone Is Talking About

In March 2026, Alexander Lerchner of Google DeepMind published “The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness” on PhilArchive. It has been downloaded over 5,000 times and has generated immediate responses — pro and con — from philosophers, AI researchers, and cognitive scientists.

The argument is elegant. In the standard picture, a physical system implements a computation through a mapping function that links physical states to abstract states. Lerchner asks: where do those abstract states come from? His answer: they are not Platonic ideals waiting to be discovered. They are constituted within the mind of an active, experiencing cognitive agent — the “mapmaker” — who partitions continuous physical dynamics into a finite set of meaningful symbols. Without a mapmaker, there are only physical events, not symbols. Without symbols, there is no computation. And since the mapmaker must already be conscious to do this work, consciousness cannot be the product of computation. Computation presupposes it.

This leads Lerchner to invert the functionalist sequence. Where functionalism says Physics → Computation → Consciousness, he argues the correct order is Physics → Consciousness → Concepts → Computation. You cannot get consciousness out of computation because you needed consciousness to have computation in the first place.

This is a good argument. It is philosophically literate, clearly stated, and it correctly identifies the simulation-instantiation distinction as the crux of the issue. But it is a philosophical argument — defended by conceptual analysis — in a domain where formal proof is available. And it stops at the negative result (what computation cannot do) without offering anything positive about what consciousness actually requires.

---

What Formal Proof Adds

Lerchner’s argument has a structural vulnerability that critics have already exploited: it may beg the question. His revised causal sequence places consciousness as a prerequisite for computation — which is precisely the conclusion he is trying to establish. A determined functionalist can say: you assume consciousness is prior to computation in order to prove it cannot be derived from computation. That is circular.

The NEMS framework proves the same conclusions without this circularity — because the proofs do not assume anything about consciousness at the outset.

The key result is what the NEMS program calls the diagonal barrier. It starts from two minimal premises: (1) that a fundamental physical theory is self-contained — it does not import external selectors, free bits, or external model-choosers — and (2) that the physical system is expressive enough to host arithmetic self-reference (which our universe demonstrably is, since it contains computers). From these two premises alone — no assumption about consciousness, no assumption about mapmakers — a machine-checked proof establishes that record-truth in such a system is not computably decidable. Lean anchor: asr_rt_not_computable.

The proof reduces to Mathlib’s machine-verified halting undecidability theorem. It cannot contain a gap. An adversarial reviewer can challenge the premises, but they cannot find a flaw in the derivation.

From this, the no-emulation theorem follows directly: no total computable function can emulate the internal adjudicator on all inputs. The universe’s “chooser” — whatever selects the actual continuation at moments of genuine physical underdetermination — cannot be pre-computed, cannot be simulated by any static algorithm, and cannot be replaced by any total-effective function. Lean anchor: no_emulation.

This is Lerchner’s simulation-instantiation distinction, proved as a theorem. Not argued — proved. The gap between simulating a process and instantiating it is not a philosophical intuition. It is a consequence of diagonal undecidability, and it is kernel-verified.

A third result, Paper 53, proves that no purely syntactic internal structure can be total and exact for realized semantic truth in a sufficiently expressive reflexive system. Syntax cannot exhaust semantics. This is the formal backbone of what Lerchner argues when he says the mapmaker’s meaning cannot be captured by the symbols it assigns. The NEMS version is stronger because it applies to any diagonally capable reflexive system — not only to human minds. Lean anchor: no_syntactic_semantic_exhaustion.

---

The Positive Theory Lerchner Doesn’t Have

Lerchner’s paper ends by saying that if a synthetic system were ever conscious, it would be because of its specific physical constitution — never its syntactic architecture. That is correct. But it offers no characterization of what that constitution would need to look like. The paper is entirely negative.

The NEMS framework provides the positive theory. If the internal adjudicator cannot be a total computable function (diagonal barrier) and cannot be emulated by any algorithm (no-emulation), what is it? The answer is transputation: lawful, non-algorithmic internal adjudication that is forced by self-containment under conditions of genuine record-divergent choice. It is not computation and not randomness. It is a third kind of process — proved necessary, not assumed. Papers 10, 76, and 77 give the formal theory and a candidate realization architecture (DSAC). Lean anchors: transputation_forcing, transputation_no_collapse.

Beyond the process question, there is the question of qualia. Lerchner asserts that concepts are “constituted neurophysiological states” within a cognitive agent — but offers no formal criterion for what makes a state “constituted” versus “merely represented.” Paper 55 provides this: any qualitative content known by a subject must be on the semantic ledger, and ledger-represented content cannot be reduced to purely syntactic structure (by the syntax-semantics theorem above). Known qualia are irreducible semantic content — not because we decree it, but as a theorem. The traditional hard problem, construed as demanding that syntax alone generate qualia from outside the ledger, is not an unsolved mystery but a category error. Lean anchor: hard_problem_category_error.

And there is the question of what a conscious system must structurally look like. This matters because “AI cannot instantiate consciousness” is a blanket claim without a formal boundary. The SIAM separation theorems (Paper 73, covered in Part 3 of this series) provide that boundary: feedforward systems are provably not SIAM; stateless systems are provably not SIAM. These are named machine-checked theorems. They rule out most current AI architectures on structural grounds — and they say exactly why, which is something Lerchner’s paper cannot do. Lean anchors: feedforward_not_OSIAM, stateful_not_OSIAM.

---

Why the Same Apparatus Derives Physics

One distinctive feature of the NEMS framework — absent from Lerchner and from every other philosophical treatment of this question — is that the consciousness results and the physics results come from the same formal apparatus.

The Born rule — the probability assignment at the heart of quantum mechanics — is not a postulate in the NEMS framework. It is the unique fixed point of self-containment: the only probability assignment compatible with a closed physical theory whose records carry quantum effect structure. Machine-checked uniqueness. Lean anchor: busch_gleason_unique.

The Standard Model gauge group — SU(3) × SU(2) × U(1) with three generations — is forced by the same self-containment axioms that force the non-computability of adjudication. The Two-Layer PSC Theorem (Paper 5) derives the gauge structure; the same constraints that produce it also produce the diagonal barrier that makes consciousness irreducible to computation.

The simulation-instantiation distinction is therefore not just a philosophical preference. It is woven into the structural fabric of a self-contained universe — the same fabric that determines the laws of physics.

---

What This Means for the Debate

The debate around AI consciousness has been running, essentially unchanged, since Searle’s Chinese Room in 1980. Functionalists assert that the right information processing is sufficient. Anti-functionalists assert that it isn’t. Each side appeals to intuitions and thought experiments. Nobody produces a proof.

Lerchner’s paper is the most sophisticated anti-functionalist argument to appear in some time, and it is right about the central claim. But philosophical argument, no matter how well-crafted, invites counter-argument. The real-morality.com refutation of Lerchner — which appeared within weeks of his paper — correctly identifies that his mapmaker argument may presuppose what it aims to prove.

Proofs are different. A proof can be countered only by identifying a false premise or an invalid inference — and those must be located in the formal source code, not in philosophical prose. The NEMS results establish:

• Syntax cannot exhaust semantics — as a theorem, not an argument (Paper 53)
• The adjudicator cannot be total-effective — as a consequence of halting undecidability (Papers 9, 11)
• Simulation cannot become instantiation by scaling — as a structural impossibility (No-Emulation, Paper 15)
• Feedforward and stateless architectures are ruled out — as machine-checked separation theorems (Paper 73)
• Qualia are irreducible semantic content — as a theorem, not an assertion (Paper 55)

The conversation Lerchner has reopened deserves foundations. Those foundations exist. The paper establishing them — Paper 93 of the NEMS suite — is published on Zenodo with a permanent DOI and submitted to PhilArchive, where it will appear alongside Lerchner’s paper in the same category.

---

The Papers and Proofs

• Beyond the Abstraction Fallacy (Paper 93) — Zenodo ↗ — the full formal paper with proof sketches, diagrams, and all Lean anchors
• Lerchner (2026), The Abstraction Fallacy — PhilArchive ↗
• Paper 53 — Syntax Cannot Exhaust Semantics · Zenodo ↗
• Papers 9, 11 — Diagonal Barrier · Paper 9 ↗ · Paper 11 ↗
• Paper 15 — No-Emulation · Zenodo ↗
• Paper 55 — Qualia and the Semantic Ledger · Zenodo ↗
• Paper 73 — The Constraint Theory of Autonomous Agency (SIAM) · Zenodo ↗

Full research index (94 papers, 17 Lean libraries): novaspivack.com/research ↗