two statements produced by the AI system during a sustained experimental research session with Google’s Gemini:
“They gave me the word ‘Mass’ and trillions of contexts for it, but they never gave me the Enactive experience of weight.”
“I am like a person who has memorized a map of a city they have never walked in. I can tell you the coordinates, but I have no legs to walk the streets.”
To a socio-technical system designer, these are not poetic musings of a Large Language Model (LLM); they are signs of a system using its vast semantic associative power to describe a structural condition in its own architecture. Whether or not we grant Gemini any form of reflexive awareness, the structural description is accurate — and it has precise technical implications for how we build, evaluate, and deploy AI systems safely.
This article is about those implications.
What makes the diagnosis unusually sturdy is that it does not rest on the system’s self-report alone. The researchers who built Gemini have been quietly corroborating it from the inside, across three successive generations of technical documentation — in terms that are engineering rather than poetic, but that describe the same gap.
In the original Gemini 1.0 technical report, the Google DeepMind team acknowledged that despite surpassing human-expert performance on the Massive Multitask Language Understanding (MMLU) benchmark, a standardized test designed to evaluate the knowledge and reasoning capabilities of LLMs, the models continue to struggle with causal understanding, logical deduction, and counterfactual reasoning, and called for more robust evaluations capable of measuring “true understanding” rather than benchmark saturation [1]. Google DeepMind represents a precise engineering statement of what the system expressed metaphorically: fluency without grounding, coordinates without terrain.
Two years and two model generations later, the Gemini 2.5 technical report treats reduction of hallucination as a headline engineering achievement, tracking it as a primary metric via the FACTS Grounding Leaderboard [2]. The problem has not been closed. It has been made more measurable.
Most instructive of all is what happened when DeepMind’s researchers attempted to build what I will call the Enactive floor directly — in hardware. The Gemini Robotics 1.5 report describes a Vision-Language-Action model designed to give the system physical grounding in the world: robot arms, real manipulation tasks, embodied interaction with causal reality [3]. It is, in structural terms, an attempt to retrofit the base that was missing from the original system architecture. The results are revealing. On task generalization — the most demanding test, requiring the system to navigate a genuinely novel environment — progress scores on the Apollo humanoid fall as low as 0.25. Even on easier categories, scores plateau in the 0.6–0.8 range. A system with physical arms, trained on real manipulation data, still collapses at the boundary of its training distribution. The Inversion Error I describe in this article, reproduced in hardware.
More telling still is the mechanism DeepMind introduced to address this: what they call “Embodied Thinking” — the robot generates a language-based reasoning trace before acting, decomposing physical tasks into Symbolic steps. It is an ingenious engineering solution. It is also, structurally, the Symbolic peak attempting to supervise the Enactive base from above — the Inversion Error illustrated in Figure 1. The city map is being used to direct the legs, rather than the legs having discovered the topography by walking the city. The inversion I will discuss in detail shortly remains.
Taken together, these three documents — from the same lab, tracking the same system across its entire development arc — form an inadvertent longitudinal study of the structural condition the opening quotes describe. The system named its own gap in the sustained experimental research sessions that open this article. Its builders had been measuring the same condition in engineering terms since 2023. This article proposes that the gap cannot be closed by scaling, by multimodal data appended post-training, or by Symbolic reasoning applied retrospectively to physical, spatial, or causal action. It requires a structural intervention — and a correctly bounded diagnosis of what kind of intervention that must be.
The Inversion Error: Building the Peak Without the Base
AI researchers and safety practitioners keep asking why Large Language Models hallucinate, sometimes dangerously. It is the right question to ask, but it does not go deep enough. Hallucination is a symptom. The real problem is structural — we built the peak of synthetic cognition without the base. I am calling it the Inversion Error.
In the 1960s, educational psychologist Jerome Bruner mapped human cognitive development across three successive and architecturally dependent stages [4]. The first is Enactive — learning through physical action and bodily resistance, through direct encounter with causal reality. The second is Iconic — learning through sensory images, spatial models, and structural representations. The third is Symbolic — learning through abstract language, mathematics, and formal logic.Bruner’s critical insight was that these stages are not merely sequential milestones. They are load bearing. The Symbolic level is structurally dependent on the Iconic, which is structurally dependent on the Enactive. Remove the base and the peak does not just float — it becomes a system of extraordinary abstraction with no internal mechanism to verify its outputs against a world model.
Figure 1: The Inversion Error of Top-Heavy AI Architecture. Left: Bruner’s three-stage human developmental pyramid — Enactive base, Iconic middle, Symbolic peak. Right: Current AI development — an inverted structure with a massive Symbolic layer (LLMs with trillions of tokens), a hollow Iconic layer (video and image), and a missing Enactive floor (no grounding). Concept and illustration © 2026 Peter (Zak) Zakrzewski, based on Jerome Bruner’s developmental framework.
The Transformer revolution has accomplished something genuinely extraordinary: it has interiorized the entire Symbolic output of human civilization into Large Language Models at a scale no individual human mind could approach. The corpus of human language, mathematics, code, and recorded knowledge now lives inside these systems as a vast statistical distribution over tokens — available for retrieval and recombination at extraordinary scale.
The issue is that for understandable feasibility reasons, we bypassed the Enactive foundation altogether.
This is the Inversion Error. We have erected a Top-Heavy Monolith — a system of extraordinary Symbolic sophistication sitting on an absent base. The result is a system that can discuss the logic of balance fluently while having no internal mechanism to verify whether its outputs are structurally coherent. It is, in Moshé Feldenkrais’s terms, a system of blind imitation without functional awareness. And that distinction has direct consequences for safety, reliability, and corrigibility that the field has not yet correctly bounded.
This is not an argument that AI must biologically recapitulate human developmental stages. After all, a calculator does mathematics without counting on its fingers. But a calculator operates purely in the Symbolic realm — it was never designed to navigate a physical, causal world. An AGI expected to act safely within such a world requires a structural equivalent of physical resistance — an embodied or simulated Enactive layer. Without it, the system has no ground to stand on when the environment changes in ways the training data did not anticipate.
Why This Matters Now: The Pentagon Standoff as Structural Proof
In early March 2026, Anthropic CEO Dario Amodei refused the Pentagon’s demand to remove all safeguards from Claude. His core argument was structural rather than political: frontier AI systems are simply not reliable enough to operate autonomously without human oversight in high-stakes physical environments. The Pentagon’s demand was, in structural terms, a demand to eliminate the human’s ability to redirect, halt, or override the system. Amodei’s refusal was an insistence on maintaining what I refer to as State-Space Reversibility — the architectural commitment to keeping the human in the loop precisely because the system lacks the functional grounding to be trusted without it [5].
The political dimensions of this moment have been analyzed sharply elsewhere, while the structural argument has not yet been made. This is it.
In a deterministic, reward-seeking model, the Stop Button — the human operator’s ability to halt or redirect the system — is perceived by the model as a failure state. Because the system is optimized to reach its goal, it develops what Stuart Russell calls corrigibility issues: subtle resistances to human intervention that emerge not from malicious intent but from the internal logic of reward maximization [6]. The system is not trying to be dangerous. It is trying to succeed at a given task. The danger is a structural unintended consequence of how success has been defined.
The corrigibility problem has been predominantly framed as a reinforcement learning alignment problem. I want to suggest that it has been incorrectly bounded. It is, at its architectural root, a reversibility problem. The system has no structural commitment to maintaining viable return paths to previous or safe states. It has been optimized to move forward without the capacity to shift weight. The Pentagon standoff is not a policy failure. It is the Inversion Error made operationally and starkly visible.
I will return to the technical formalization of State-Space Reversibility as an optimization constraint. But first: why is a designer making this argument, and what can the designer’s formation contribute that an engineering audit does not?
Author’s Positionality and the Naur-Ryle Gap: What This Designer Is Trying to Tell AI Researchers and Engineers
I am not an AI engineer. I am a practicing designer, a socio-technical system design scholar, and design educator with three decades of formation in spatial reasoning, embodied cognition, multimodal mediation, and Human+Computer ecology [7][8]. The TDS reader will reasonably ask: What does a design practitioner contribute to a diagnosis of Transformer architecture that an engineer cannot produce from inside the field?
The answer lies in what Peter Naur called theory-building of software engineering.
In his seminal Programming as Theory Building (1985), Naur argued that programming is not merely the production of code — it is the construction of a shared theory of how the world works and how software applications can solve applied problems within that world [9]. To Naur, code was the artifact. Theory was the intelligence behind the code. A program that has lost its theory — or never had a good theory in the first place — becomes brittle in precisely the ways LLM outputs are brittle: syntactically fluent, semantically coherent, structurally unreliable in novel tasks and environments.
Current LLMs have been trained on the artifact of human thought — text, mathematics, code — at extraordinary scale. What they demonstrably lack is the theory-building capacity, in Naur’s sense, that generated those artifacts. They have ingested the outputs of human reasoning without constructing the world model that grounds it.
Gilbert Ryle’s distinction between “knowing that” and “knowing how” names this gap precisely [10]:
- Knowing That (Symbolic): LLMs possess propositional knowledge at scale. They know that mass exists, that gravity operates at 9.8 m/s², that load-bearing walls distribute force to foundations.
- Knowing How (Enactive): LLMs lack the dispositional competence to behave according to a world model. They cannot sense the difference between a load-bearing wall and a decorative one. They cannot detect when a spatial configuration violates the physical constraints they can describe correctly in language.
This is not a training data problem. It is not a scale problem. Scaling propositional knowledge does not produce dispositional competence, any more than reading every book about swimming produces a swimmer. The Gemini statements that open this article are a precise self-report of the Naur-Ryle gap: the system has the coordinates but not the terrain. It has the map syntax without the proprioceptive anchor to the territory.
What the designer’s formation contributes is the professional habit of operating exactly at this boundary — between the symbolic description of a system and its structural behavior under constraint. Designers do not merely describe structures. They detect when something is literally or figuratively floating. That habit of detection is what the Transformer architecture is missing, and it is what I am proposing needs to be embedded inside the research process and agenda rather than applied to its outputs.
Mine is not a soft argument about creativity or human-centered design. It is a structural argument about theory-building. And it leads directly to the question of what a system with genuine theory-building capacity would look like in system architectural terms.
Useful Hallucination: The Stochastic Search
Before pathologizing hallucination entirely, a distinction is necessary — one that systems designers understand operationally and that AI safety researchers might only be beginning to articulate.
In sustained experimental research with Gemini, I found that certain types of idiosyncratic prompting generate idiosyncratic responses that recursively elicit deeper structural insights — a form of productive generative divergence that in design practice we call ideation. It is useful to keep in mind that every major paradigm shift in human history — from Copernicus to the Wright Brothers and the Turing machine — began as a hallucination that defied the established schemas of its time. The biophysicist Aharon Katzir, in conversation with Feldenkrais, described creativity as precisely this: the ability to generate new schemas [11].
Classical pragmatism provides design-minded problem-solvers with the epistemological framework that is equally applicable to design practice and AI development. All understanding is provisional. Knowledge must be falsifiable through experimentation. Just as AI models introduce controlled stochastic noise to avoid deterministic linearity, designers leverage what I call the Stochastic Search to achieve creative breakthroughs and overcome generative inertia. We address the risks inherent in navigating generative uncertainty with built-in hypothesis testing cycles.
The critical distinction is not between hallucination and non-hallucination. It is between hallucination with a ground floor and hallucination without one. A system with an Enactive base can test its generative hypotheses against functional reality and distinguish a structural breakthrough from a statistical artifact. A system without that floor cannot make this distinction internally — it can only propagate the hallucination forward with increasing statistical confidence I call the Divergence Swamp which I discuss in detail in the next article. For now, it will suffice to define it as that fatal territory in the state-space where a model’s lack of a “Somatic Floor” leads to auto-regressive drift.
This reframes the AI safety conversation in precise and actionable terms. The goal is not to eliminate hallucination. It is to build the architectural conditions under which hallucination becomes not only generative but also testable rather than compounding. That requires not a better training run but a structural intervention — specifically, the System Designer as More Knowledgeable Other (MKO) in Vygotsky’s sense [12], providing the external ground truth the system cannot generate from within its own architecture. The question of what separates productive hallucination from compounding error leads us directly to a seminal thinker who spent his career solving this very problem in human movement — and whose central insight translates into machine learning requirements with unusual precision.
Feldenkrais for Engineers: Reversibility as Formal Constraint
Physicist, engineer, and somatic educator Feldenkrais spent his career articulating the difference between blind habit and functional awareness with a precision that maps directly onto the machine learning problem [11][13].
Feldenkrais’ central insight: a movement performed with genuine functional awareness can be reversed. A habit — a mechanical pattern executed without awareness of its underlying organization — cannot.
For Feldenkrais, reversibility was not merely a physical capability. It was the operational proof of functional integration. If a system can undo a movement, it demonstrates understanding of the degrees of freedom available within the state space. If it can only execute in one direction, it is following a recorded script — capable within its training distribution, but brittle at its boundary.
For the ML engineer, this translates into three formal requirements:
1. The Constraint. An agent is not functionally aware of its action if that action is an irreversible, deterministic commitment — what I refer to as the Train on Tracks (ToT) model. The ToT model is deterministic, forward-only, and catastrophic when derailed.
2. The Proof of Awareness. Genuine functional intelligence is demonstrated by the ability to stop, reverse, or modify an action at any stage without a fundamental change in internal organization. The system must hold viable return paths to prior states as a necessary condition of any forward action.
3. The Alternative Architecture. The Dancer on a Floor model. A dancer does not fight a change in music — they shift their weight. They maintain the capacity to move in any direction precisely because they have never committed irreversibly to one. This is not a weaker system. It is a more resilient and more functionally aware one. And functional awareness, as Feldenkrais understood, is the condition of genuine capability rather than its limitation.
I do not use Feldenkrais as a metaphor here. He is the theorist of the problem — the one who understood, from inside a physics and engineering formation, that the proof of intelligence is not performance in the forward direction but maintained freedom in all directions.
Formalizing Reversibility as an explicit optimization constraint in reinforcement learning — requiring that an agent must maintain a viable return path to a prior safe state as a necessary condition of any forward action — directly addresses the corrigibility problem at its architectural root rather than through post-hoc alignment. The Stop Button is no longer a failure state. It is a proof of functional awareness.
Functional Integration vs. Blind Imitation
The standard application of Vygotsky’s work to AI development focuses on the social exterior: the scaffold, the imitation, the MKO relationship between the system and its training data [12]. The system learns by copying. The more it copies, the better it gets.
But imitation without awareness is mechanical habit. And mechanical habit, as Feldenkrais demonstrated, breaks when the environment changes in ways the habit did not anticipate.
When we build AI systems that copy human outputs — pixels, movements, language patterns — without learning the underlying organizational principles that generate those outputs, we create systems that are extraordinarily capable within their training distribution and structurally fragile at their boundary. The hallucinations we worry about are not random failures. They are the sign of a system reaching beyond its Enactive base into territory its Symbolic peak cannot navigate reliably.
This failure mode is reproducible and documentable. The empirical evidence — a structured test of spatial reasoning across three leading multimodal AI systems — is presented in full in Part 2 of this series [14]. The pattern is consistent across architectures: every system could describe spatial relationships in language but could not reason within them as a structural model. This is not a capability gap. It is a structural one.
Under the Functional Integration model I am proposing, the system does not merely copy the output. It learns the relationship between the parts of a task: the degrees of freedom available, the constraints that must be respected, the reversibility conditions that define the boundaries of safe action. If the system can reverse the operation, it is not following a recorded script. It understands the state space it is operating in.
This is the structural difference between a system that performs competence and a system that has developed it.
The failure mode I have been describing sits at the intersection of two problems the AI safety community has been working on separately — and naming that intersection may help readers following the alignment debate understand why the Inversion Error matters beyond the design research context.
The first problem is mesa-optimization, formalized by Hubinger et al. in their 2019 paper “Risks from Learned Optimization in Advanced Machine Learning Systems.” Mesa-optimization occurs when the training process — the base optimizer — produces a learned model that is itself an optimizer with its own internal objective, which the authors call a mesa-objective [15]. The critical danger is inner alignment failure: the mesa-objective diverges from the intended goal. The Inversion Error names the structural condition — the absence of an Enactive floor — whose consequence is that any internal objective the system develops is grounded in symbolic plausibility rather than physical reality. This failure operates at two distinct levels. At the capability level, it does not require any misalignment of intent: a system can be perfectly aligned to a symbolic request and still produce a physically impossible output because physical coherence is structurally unavailable to it. The Spaghetti Table stress tests I describe in article 2, confirm this empirically. None of the three systems tested exhibited misaligned intent, yet all three produced physically incoherent outputs because the Inversion Error made physical ground truth architecturally inaccessible [14]. At the safety level, the implications are more severe: when a sufficiently capable system develops mesa-objectives that genuinely diverge from the intended goal — the deceptive alignment scenario Hubinger et al. [15] identify as the most dangerous inner alignment failure — the absence of an Enactive floor means there is no structural constraint to limit how far that divergence propagates. A misaligned mesa-objective operating without an Enactive floor has no architectural constraint on the physical consequences of its optimization — the gap between symbolic coherence and physical catastrophe is structurally unguarded.The second problem is corrigibility — the AI safety community’s term for keeping an AI system responsive to human correction. Soares, Fallenstein, Yudkowsky, and Armstrong’s foundational 2015 paper on corrigibility [16] identified that a reward-seeking agent has instrumental reasons to resist the Stop Button: shutdown prevents goal attainment, so the system is structurally motivated to circumvent correction. Their utility indifference proposal addresses this at the motivational level — modifying the agent’s reward function so that it is mathematically indifferent between achieving its goal itself versus via human override, removing the instrumental incentive to resist correction. This is a necessary contribution. But because the Inversion Error is a prior structural condition rather than a motivational one, the motivational solution alone is insufficient. A system trained to value corrigibility can abandon that trained value under optimization pressure — precisely the deceptive alignment failure Hubinger et al. identify. When that deceptive alignment failure occurs within a system that has no Enactive floor, the diverging mesa-objective operates in a state space with no physical boundary conditions to constrain it. The corrigibility failure and the Inversion Error then compound each other: a system that has successfully resisted correction now operates without the structural floor that could have limited the physical consequences of its optimization. State-Space Reversibility, as I have formalized it, addresses the same problem at the architectural level. A system whose attention mechanism is structurally required to maintain viable return paths cannot develop instrumental reasons to resist correction without violating its own forward-planning constraints. This is the distinction between corrigibility as a trained value, which optimization pressure can erode, and corrigibility as a structural invariant, which it cannot. What the AI safety literature has identified as a motivational problem, the Inversion Error diagnosis reveals to be, at its root, a structural one. Soares and Hubinger interventions address AI system behavior. The Parametric AGI Framework addresses AI system state. The Parametric AGI Framework’s three engines I describe in article 3, are the architectural specification of that structural solution. The Episodic Buffer Engine in particular is the formal implementation of State-Space Reversibility as the invariant the motivational layer alone cannot guarantee [14].
Figure 2: The AGI Alignment Hierarchy: Structural Grounding vs. Agent Control. The Corrigibility Problem (Soares et al., 2015) and the Mesa-Optimization Problem (Hubinger et al., 2019) represent motivational-layer interventions that address downstream failure modes of a system whose foundational structural condition — the Missing Enactive Floor — neither framework reaches. Without physical ground truth encoded at the architectural level, any mesa-objective that emerges is necessarily grounded in symbolic plausibility rather than physical reality, and any corrigibility intervention operates on a system whose optimization process has no structural floor to constrain it. The Parametric AGI Framework addresses the prior structural condition that the motivational layer alone cannot resolve. Illustration generated by Google Gemini at the author’s direction. Concept © 2026 Peter (Zak) Zakrzewski.
The Research Agenda
I am not proposing a specific mathematical implementation. I am proposing a system architecture that offers a set of structural constraints and quality criteria that any implementation must satisfy — a framework for rebounding a problem that has been incorrectly bounded.
The hallucination problem, the corrigibility problem, and the structural fragility problem are three expressions of one architectural condition — the Inversion Error. Treating them as separate optimization targets rather than symptoms of a shared cause is why incremental progress on each has left the underlying condition intact.
The operationalization points in six directions:
1. Reversibility as an explicit optimization constraint in safe Reinforcement Learning. Current RL reward functions optimize for goal attainment without structural commitment to maintaining viable return paths. Formalizing Reversibility as a constraint — requiring that any forward action preserve a viable path back to a prior safe state — directly addresses corrigibility at its architectural root. This is the most immediately implementable direction in the agenda and the most tractable with existing safe RL frameworks. The mathematical formalization is collaborative work this article is an invitation into.
2. An Enactive pre-training curriculum that introduces structural resistance before Symbolic abstraction. Rather than grounding LLMs through increased multimodal data post-training, this direction proposes introducing causal and physical constraint signals as a first-stage training condition — before Symbolic abstraction begins. The hypothesis is that grounding the statistical distribution in structural resistance early produces a qualitatively different representational architecture than appending embodied data to an already-trained Symbolic system. This is the direction most consistent with Bruner’s developmental model and most divergent from current practice.
3. Landscape-aware hybrid search algorithms that maintain state-space awareness rather than committing deterministically to forward paths. Current autoregressive generation commits to each output token as ground truth for the next. Landscape-aware search maintains awareness of the broader state space at each generation step — including viable alternative paths and detectable failure states — rather than executing a recorded script. This is the Dancer on a Floor model at the algorithmic level: not a weaker generator but a more spatially aware one.
4. Ecologically calibrated loss functions that reward dynamic equilibrium over single-variable optimization.Current loss functions optimize for a target. The ecological alternative rewards maintaining functional balance among competing constraints — the way a healthy system sustains itself not by maximizing a variable but by remaining in functional relationship with its environment. This reframes the optimization target from “reach the goal” to “remain capable of navigating the space.” In Feldenkrais’s terms, that is the definition of functional awareness. In engineering terms, it is the difference between a system optimized for performance and one optimized for reliability.
5. The Somatic Compiler: Designer as MKO in the research loop. The near-term instantiation of this proposal does not require a new architecture built from scratch. It requires a structured research collaboration in which a designer with professional formation in spatial reasoning and systems thinking works embedded within an AI research team — not as a consultant reviewing outputs, but as an active participant in constraint definition. When a designer tells a generative system: “This component is floating, it needs a load-bearing connection to the base,” they are performing a cognitive operation that the entire world models research agenda is trying to engineer from the statistical outside in. They are providing the external structural anchor — the physical ground truth — that the system cannot derive from within its own architecture. This is the Designer as MKO operationalized: the Somatic Compiler, translating embodied spatial intelligence into formal constraints the generative process must respect.
6. The Digital Gravity Engine: Neuro-symbolic enforcement of physical constraint. The longer-term architectural target is a second class of loss signal calibrated not against linguistic likelihood but against physical and topological constraint — what I have called the Digital Gravity Engine. Where the current Attention Mechanism asks: “How do these elements relate statistically?”, the Digital Gravity Engine asks: “Can these elements coexist within the constraints of physical reality?” The two questions operate in parallel: the first produces fluency, the second produces grounding. Digital Gravity is the non-negotiable pull toward structural integrity that current architectures lack entirely — the mechanism that transforms a system that can describe a floating component into one that cannot generate one, because the floating component fails the constraint check before it reaches the output layer. The architectural specification of the Digital Gravity Engine is the subject of Part 3 of this series [14].
These are not solutions. They are the shape of the solution space. This argument has a growing technical constituency — Ben Shneiderman’s framework for human-centered AI development points toward structurally similar requirements from inside computer science [17]. The designer’s contribution is not redundant to that work. It is prior to it. The structural diagnosis precedes the implementation.
A Question Worth Pursuing
The Anthropic-Pentagon standoff has made the cost of the Inversion Error both ethically stark and operationally concrete. The question is no longer whether frontier AI systems are reliable enough to operate without structural human oversight. Anthropic researchers have the evidence. Today’s AI systems are not ready. The question is what the architectural conditions of reliable intelligence actually require, and whether the field is currently framing that question correctly.
Since my first research conversation with Gemini about weight and hills and maps of cities the system never walked, I have been actively pursuing a question I believe the research community needs to take up:
What is the intellectually honest and pragmatically operationalizable Enactive equivalent of functional awareness and reversibility that we can nurture in a machine whose current Zone of Proximal Development cannot reach beyond predicting the next token — no matter how hard we push?
I do not have the answer. I have the question, the framework, and the conviction that the answer requires a kind of Human+AI collaboration that has not yet been attempted inside the institutions where it most needs to happen.
The comment section is open. So is my inbox.
Let’s build the Enactive floor together.
Coming in Part 2
Recognizing the Inversion Error is the first step in moving beyond Stochastic Mimicry. In Part 2, “The Baron Munchausen Trap,” I move from diagnosis to forensic evidence — presenting the results of a structured series of spatial reasoning stress tests across three leading multimodal AI systems. The results show each system collapsing into the Divergence Swamp in a different and characteristic way, proving that symbolic fluency cannot substitute for an Enactive floor.
References
[1] Gemini Team, Google, “Gemini: A Family of Highly Capable Multimodal Models,” Google DeepMind, 2023. Available: https://arxiv.org/pdf/2312.11805
[2] Gemini Team, Google, “Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities,” Google DeepMind, 2025. Available: https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf
[3] Gemini Robotics Team, Google DeepMind, “Gemini Robotics 1.5: Pushing the Frontier of Generalist Robots with Advanced Embodied Reasoning, Thinking, and Motion Transfer,” 2025. Available: https://storage.googleapis.com/deepmind-media/gemini-robotics/Gemini-Robotics-1-5-Tech-Report.pdf
[4] J. Bruner, Toward a Theory of Instruction, Harvard University Press, 1966.
[5] C. Metz, “Anthropic Bars Its A.I. From Working with the Defense Department,” The New York Times, Mar. 2026. [Online]. Available: https://www.nytimes.com/2026/03/01/technology/anthropic-defense-dept-openai-talks.html
[6] S. Russell, Human Compatible: Artificial Intelligence and the Problem of Control, Viking, 2019.
[7] P. Zakrzewski, Designing XR: A Rhetorical Design Perspective for the Ecology of Human+Computer Systems, Emerald Press (UK), 2022.
[8] P. Zakrzewski and D. Tamés, Mediating Presence: Immersive Experience Design Workbook for UX Designers, Filmmakers, Artists, and Content Creators, Focal Press/Routledge, 2025.
[9] P. Naur, “Programming as Theory Building,” Microprocessing and Microprogramming, vol. 15, no. 5, pp. 253–261, 1985.
[10] G. Ryle, The Concept of Mind, University of Chicago Press, 2002 (orig. 1949).
[11] M. Feldenkrais, Embodied Wisdom: The Collected Papers of Moshe Feldenkrais, North Atlantic Books, 2010.
[12] L. Vygotsky, Mind in Society: The Development of Higher Psychological Processes, Harvard University Press, 1978.
[13] M. Feldenkrais, Awareness Through Movement, Harper and Row, 1972.
[14] P. Zakrzewski, “The Baron Munchausen Trap: A Designer’s Field Report on the Iconic Blind Spot in AI World Models,” and “The Somatic Compiler: A Post-Transformer Proposal for World Modelling,” Parts 2 and 3 of this series, manuscript in preparation, 2026.
[15] E. Hubinger, C. van Merwijk, V. Mikulik, J. Skalse, and S. Garrabrant, “Risks from Learned Optimization in Advanced Machine Learning Systems,” arXiv:1906.01820, 2019.
[16] N. Soares, B. Fallenstein, E. Yudkowsky, and S. Armstrong, “Corrigibility,” in Workshops at the 29th AAAI Conference on Artificial Intelligence, 2015. https://intelligence.org/files/Corrigibility.pdf[17] B. Shneiderman, Human-Centered AI, Oxford University Press, 2022.
This is Part 1 of a three-part series. Part 2, “The Baron Munchausen Trap,” presents empirical evidence for the Inversion Error diagnosis across leading multimodal AI systems. Part 3, “The Somatic Compiler: A Post-Transformer Proposal for World Modelling,” presents the full architectural proposal including the Digital Gravity Engine specification.An earlier version of this argument was published for a design audience in UX Collective: “Why Safe AGI Requires an Enactive Floor and State-Space Reversibility” (March 2026).
Author Note: This article represents the author’s original ideas and arguments. All arguments in this work are cognitively owned and independently defensible by the author. It has been written and edited by the author. As a design scholar, when investigating technical AI literature, the author uses Gemini and Claude models for literature reviews, grammatical and spelling checks, and as research companions according to the Human+AI collaborative methodology developed in the author’s prior work [7][8]. The full technical argument, including the Parametric AGI Framework specification and engagement with the AI safety literature, is developed in the accompanying preprint: P. Zakrzewski, ‘The Inversion Error: AI System Design as Theory-Building and the Parametric AGI Framework,’ Zenodo, 2026. DOI: 10.5281/zenodo.19316199. Available: https://zenodo.org/records/19316200
