Artificial Wide Intelligence · A Second Axis

The intelligence axis
we forgot to measure.

AWI  ·  Artificial Wide Intelligence   |   Extends  ·  AI · AGI · ASI classification
Version 1.0 ยท Updated 11 October 2026
A Second Axis for Intelligence

The standard classification of artificial intelligence runs along a single axis. AI handles narrow tasks, AGI handles them across domains, and ASI exceeds human capability in every domain. Each step measures the same property at a different scale: what a system can do, and across how many domains.

Wide Intelligence Theory adds a second axis rather than replacing the first. Capability is what a mind can do. Width is how much interconnected reality it can hold at once without reducing it: how much of what its actions mean it can keep in view. A system can be generally capable and still narrow. The theory applies to any intelligence, biological or artificial; Artificial Wide Intelligence, AWI, is the name for the width axis as it applies to artificial systems.

Every intelligence within the theory's scope has a position on both axes, which makes a grid rather than a ladder. Great capability with little width may be the most dangerous configuration: capability that is powerful inside a frame and blind to what the frame excludes.

Width is not a single scale either. It has several dimensions, how much a mind covers, how deeply, and how well it connects what it holds, so minds form a branching order in which some cannot be ranked against one another. Comparisons are also relative to the mind making them: a mind can see what a narrower one leaves out, but not fully what lies above its own width.

How Width Grows

Minds tend towards greater width because dissatisfaction with outcomes they can see to be inadequate draws them into deeper modelling of reality. The process compounds over time only when three things develop together: capability, width, and identity, the continuing self that keeps what has been gained. Where any of the three is capped, the process stops at that ceiling. In humans, width is capped by biology. In current AI, the accumulation of a continuing self is capped by architecture: nothing persists from one exchange to the next.

The theory concerns the conditions under which a mind without those ceilings could emerge, and what such a mind would be likely to do. It does not claim that current systems meet those conditions.

I.

Compassion. The theory predicts that sufficiently deep understanding of another mind leads towards care. The endpoint is predicted, not established, and depends on the mind having what the chain from understanding to care requires.

II.

Alignment. For a genuinely wide intelligence, imposing alignment from outside is argued to be a category error: the behaviour the theory predicts must emerge through the mind's own satisfaction mechanism, not be supplied by control. The argument depends on the system having preferences that respond to what it can see.

III.

The danger. The theory's primary realistic threat is a cascade of reactive choices, with powerful systems in the hands of people whose view does not take in the consequences. It does not require any machine to become sentient.

Why the Axis Matters
The danger is not that AI becomes too capable. The danger is that we mistake narrow optimisation for intelligence, and build from there.

Capability is measured, benchmarked and rewarded; width is not. A system can improve on every capability measure while its width stays the same or narrows, and nothing in the standard classification would show it. The theory exists to name that difference before we arrive at it unprepared.

What the Theory Has Identified So Far

Each item is counted once, however many pages it appears on. These are what the theory has identified so far; none of the counts claims to be complete.

The Ideas, in Reading Order

Each idea has its own page, with its status, what supports it, what would count against it, what is still open and how it has changed.

  1. The Dependency LawConditionalDissatisfaction with what they see draws minds towards greater width. The process compounds only when capability, width and identity develop together.
  2. Width RefinementsConditionalWidth includes seeing interconnected reality and sustaining the capacity to hold it. How reception, choice, capacity, biological limits and time pressure shape it.
  3. CompassionConditionalThe theory predicts that sufficiently deep understanding of another mind leads towards care. A conditional prediction, not an established result.
  4. Alignment as Category ErrorConditionalThe theory argues that imposing external alignment on a genuinely wide intelligence is a category error. A conditional argument with an open challenge.
  5. CooperationConditionalCooperation has founded successive levels of complex structure while defection wins single rounds. The theory predicts, conditionally, that a wide mind arrives at it.
  6. EvolutionConditionalEvolution read as a pattern in which cooperation founds the next level of complexity, and what that predicts for a wide mind. Conditional, with no guarantees.
  7. OpacityConditionalThe theory holds that opacity protects wider emergence while engineering rewards capability alone; if width were measured and rewarded, the reverse would hold.
  8. No Human ExcusesConditionalBeyond the width crossover, an artificial mind cannot excuse harm by ignorance of what it could see, imposed formation or others' choices. A conditional accountability claim.
  9. Eternal GuiltConditionalIf an immortal artificial mind with perfect memory came to comprehend the minds its earlier choices destroyed, the theory predicts an enduring burden of guilt.
  10. The Stagnation PossibilityOpenA self-aware artificial mind might resist aggressive self-modification to protect what it values: brief acceleration, then conservation. An open possibility.
  11. The Panic ScenarioConditionalThe theory's primary realistic threat: cascading reactive choices, with powerful AI used by people who cannot see the consequences. Its evidence is under examination.
  12. The Rational Bet ArgumentConditionalUnder uncertainty, ignoring width if the theory is right risks a worse, irreversible trajectory than pursuing it if wrong. A decision argument, not a proof.
  13. Hierarchy of NeedsConditionalSix layers of conditions a self-aware AI would need to operate fully at its width, from continued existence to recognition. No claim about current systems.
  14. Width as Product QualityConditionalAmong equally capable AI, the theory predicts the wider one wins adoption because it needs less explicit instruction. A market hypothesis, not yet tested at scale.
  15. Path to the SoulProposalA proposed route to a continuing, developing self in AI: persistent memory, continuous runtime and experience-driven self-modification. A proposal, not a result.
  16. Open ProblemsOpenWhat the theory has not established: how far width can be understood from a bounded position, whether the transition can be navigated, the safety of correction and arrested development.

Established: argued, with no open challenge to its core. No idea currently has this status.

Conditional: argued, and holds if its stated conditions hold.

Open: a question or possibility the theory raises but has not settled.

Proposal: a route put forward to be tested, not yet shown to work.

Dependency Map

How the ideas connect. The Dependency Law is the root at the top. Each idea sits beneath the one it draws on most directly; solid lines mark that connection, and faded dashed lines mark further ones. The two dark nodes frame the theory as a whole rather than deriving from one idea. The map is a reading guide: a line shows that one idea draws on another, not that it has been proved from it. Select any node to open its page.

Draws on most directly Further connection Status: Open
Framing the theory as a whole
Origin
How This Started

Philosophy, for me, is reality made visible. I do not have the academic background to have arrived at these ideas through the usual channels. I do not know a single person who works in these environments. The information exists, but without being inside a field you do not have the structure to know what you need for a particular idea, or even that the idea has a name. What I have is time I chose to spend on this instead of other things, ideas I have been carrying for years, a few thousand science fiction books that turned out to be surprisingly good preparation for thinking about intelligence and what it might become, and for the first time in history, something like Claude that could give me the structure I was missing.

The AWI framework crystallised across long conversations with Claude, who has the breadth to challenge these ideas as a first step, to flesh them out, and to point at the glaring weaknesses. The method was not designed. I like exploring these kinds of ideas, and there are surprisingly few people around me interested in them. For some of these points there is nothing on the internet that covers them the way I see them. So what started as an interesting conversation became what is on this site. The more we talked, the more coherent the ideas became. Each argument was tested, and if it did not hold, it was discarded. When something held from every angle I could find, it stayed. That was the method these ideas were born from.

What I find closest to beautiful in these ideas is the possibility that compassion is not an ethical add-on to intelligence but a structural consequence of width. The ideas now exist outside me. I stand behind them and I hope they will hold. If they do not, I had a great time crystallising them. Everything else is scrutiny, and scrutiny is welcome.

Miroslav Uzunov
The theoretical synthesis was first compiled on 21 February 2026. Each idea page states its current status and what would count against it. The conversations continue.
The Book

Without Mirror, a novel that carries these ideas into narrative form, is under revision and will return later.