The Theory
AGI ALS / CTR
The superproblem
Today's AI industry solves problems of optimization, automation, and recombination of humanity's existing knowledge: summarizing, searching, planning, reasoning within known domains. That is real value, and it will keep growing. But it is not the problem AGI is meant to solve.
The actual superproblem of AGI is constructing reality at the level of new ontological knowledge — not a new fact inside an accepted picture of the world, but a change to the picture itself: a new system of basic entities, new causal mechanisms, a new level of description. Science has seen such transitions before — geocentrism to heliocentrism, alchemy to the periodic table, vitalism to genetics and DNA, caloric theory to thermodynamics, classical mechanics to quantum physics and relativity. Each one didn't just add facts; it opened a new dimension of reality and launched an entire technological era. Historically, such transitions have occurred roughly once a century.
In practice, this defines a class of problems current AI cannot reach — not for lack of scale, but by kind:
- New explanatory frameworks in fundamental physics, biology, and brain science.
- New causal models of complex systems (organisms, aging, immunity, climate, macroeconomics, geopolitics, infrastructure) where old models break down.
- New principles for materials, energy, computing, and bioengineering rather than optimizing known ones.
- Work in open environments with no example corpus, no external success metric, where the problem itself must change as understanding develops.
Criterion
What intelligence needs for this task
New ontological knowledge is not produced by computation, search, or reasoning inside a fixed axiom set. It requires a class of cognitive functions historically discussed only philosophically or psychologically, never given a constructive form:
- Intuition
- Locating a productive direction before formal proof, when the space of options can't be enumerated and data is insufficient.
- Creativity
- Not a new combination of the known, but a move to a new organization of knowledge: new entities, a new coordinate system of description.
- Heuristics
- Generating a method of solving where no ready procedure exists, guided by an internal criterion of productivity.
- Causality
- Working with the mechanism that generates events (including the causality of new structures emerging), not with correlations or predictively useful dependencies.
- Subjectivity and internal goal-setting
- A functional basis for cognitive acts that comes from the system itself, not from an external reward function.
- Qualia / functional mentality
- A subjective functional relation to the cognitive act, without which "understanding" stays imitation.
- Ontological plasticity
- The ability to restructure the system of categories itself, not just the knowledge held within it.
These functions cannot exist as a set of independent modules — they must be organized into a single functional architecture of a living-system type. That unification is one of the key results of our first stage.
The Conceptual Limit — even GPT-6 Astra is not AGI
The limit of the current paradigm isn't computational resources, scale, or statistics — it's conceptual, and it shows most clearly in the paradigm's own best result.
Models in the GPT-6 Astra class are the most mature expression of today's concept of intelligence. Their result on the interactive ARC-AGI-3 benchmark — roughly two-thirds of human-level performance, versus under 1% for prior generations — is a genuine breakthrough. Astra can explore unfamiliar environments, form and test hypotheses, build a local model of the rules, use tools, and adjust its behavior from feedback. This is no longer "a statistical language model" — it's a highly capable interactive, reflective agent.
That is exactly why its limit is so telling. Even in this form, the system stays in one ontological mode, as above. Astra can brilliantly uncover the hidden rules inside a given world. But it cannot question the sufficiency of that world of representations itself. In our terms, this shows up as four barriers shared across the entire paradigm:
No subjectivity.
Goals are set externally — by the task, the benchmark, the reward function. The system is reactive; it has no internal functional basis for its own acts.
Imitation of understanding.
Heuristics reduce to search (search trees, sampling with verification); creativity reduces to interpolation inside learned representations. Neither produces a new level of representation.
Blindness to causality.
Causality is replaced by operationally useful dependencies; the causality of a new structure emerging is not distinguished from the causality of an existing structure's behavior.
Reduction of randomness.
Randomness is treated as distributional noise to be averaged away. But randomness as its own thing — a symmetry break in equally probable potentiality — is the actual mechanism by which anything new is born. Architectures that optimize past distributions cannot, by construction, produce the genuinely new.
All four barriers share one root: there is no mechanism in the architecture that can change the coordinate system of reflection itself — what can even become an object, a cause, a distinction, a meaning for the system. Parameters and content change; the way the world gets built does not.
This produces a telling difference at the edge of knowledge. For Astra, hitting a limit means "the solution hasn't been found in the current space yet" — so it searches, reasons, and checks harder. For AGI ALS/CTR, hitting a limit means "this representational space has stopped being sufficient" — and calls for a different kind of act: rebuilding the foundation itself. That is where the current paradigm ends and ours begins.
The Solution — AGI ALS / CTR
AGI ALS: general intelligence as living-system behavior
AGI ALS was the first framework to rigorously define general intelligence not as a computational agent or a statistical model of representations, but as a property of living-system behavior. Outside the behavior of a living system, general intelligence loses its meaning — the correct formulation of AGI is: intelligence of an "artificial" living system.
This gave an independent result in its own right:
- A concrete specification of the functional systems a full AGI needs — subjectivity, internal goal-setting, causality, qualia, mentality, intuition, heuristics, creativity, and — further out — more complex forms of mental functionality (altered states of consciousness, savant-type cognitive configurations, transcendent phenomena).
- A demonstration that these functions form a single hyper-networked functional architecture, not independent modules.
- Formalization through a theory of linguistic conceptography, developed into functional conceptography: cognitive functions are mapped onto a conceptographic system as a functional hierarchical structure, which lets cognitive processes be described as dynamic living systems rather than as manipulation of symbols, tokens, or embeddings.
- A working MVP of the first stage of the AGI ALS model (a hybrid architecture with an LLM loop), built on this basis.
Why ALS necessarily leads to CTR
Working with ALS surfaced a limit that no amount of added complexity could remove. It has two sides.
The first side — "unconscious intelligence." Modeling only the cognitive processes human consciousness can observe, analyze, and describe keeps you on the surface of intelligence. The key phenomena — the emergence of cognition itself, the birth of new meaning, the evolution of ontology, the internal causality of a cognitive act — unfold at a pre-reflective level for which "visible" consciousness cannot serve as a prototype.
The second side — the living. If intelligence is a functionality of a living system, we need to answer what the elementary constructive act of the living actually is — what distinguishes behavior from a physical process. Tracing this down from organism to cell to virus, and further to the limiting living structure — a single protein molecule at the boundary of the living and the non-living — shows that a classical framework cannot locate a behavioral act at the molecular level at all, while a quantum-mechanical and pre-quantum framework opens up practically unlimited possibility for organizing acts of behavior. The elementary act of the living leads directly into the construct of reality itself.
Both lines converge on one requirement: a theory of the fundamental mechanisms from which locality, distinguishability, reflection, and meaning all arise — what cognition is built from in the first place. That is the level of the Constructive Theory of Reality — CTR (Vasilenko).
Project logic: AI → AGI → AGI ALS → CTR, then the reverse pass: building on CTR's mechanisms, the model of a living system unfolds as AGI ALS/CTR — the complete functional system of AGI.
Reverse pass — built on CTR's mechanisms
AGI ALS / CTR
Complete functional system of AGI
The relationship between the two levels is asymmetric. CTR is not a layer on top of ALS; ALS only fully unfolds on CTR's mechanisms — the same way classical physics is a limiting case of quantum physics, not its foundation.
CTR: the Constructive Theory of Reality
CTR is a theory of constructive reality. To be precise: CTR's goal is not to interpret the physical universe, but to formulate the principles and mechanisms of reality's construct needed to model AGI. It touches ideas from Deutsch (constructor theory), Penrose, Wilczek, and Tegmark, but differs from them in a key way: the fundamental medium of reality is treated as a pre-object entity — a medium of potential reality with unlimited possibility, not tied to physical reductionism. CTR gives a single constructive mechanism — a formula and a set of construct principles — for the existence and evolution of reality across physical, informational, and mental entities.
Founding formula (simplified)
Actual reality isn't "given" — it constructively arises as the localization of potential variability, its reflection, its acceptance, and its resulting definiteness of meaning. Meaning here is not a human psychological function but an objective structural state of reality; human understanding is meaning of a second order.
CTR does not hand over ready-made "blueprints" of mental processes. But it isn't philosophy either — it's a system of concrete principles and mechanisms of reality on which mental functionality becomes engineerable. That is exactly why the four barriers above get, in CTR, a mechanism rather than a description.
This is also the answer to whether intuition, creativity, and heuristics share one problem. They do — the missing mechanism for changing the coordinate system of reflection itself is the shared root. And the solution is likewise shared: not three separate modules, but one constructive mechanism of reality from which these functions unfold as its particular regimes.