William Cook
Abstract
Traditional views of intelligence often emphasize memory, processing speed, logic, or accumulated knowledge. This paper proposes an alternative framing: that intelligence may be more accurately understood as the capacity to form valid abstract relationships across distant conceptual domains. Under this model, abstraction is not intelligence itself, but a non-linear mechanism for generating relational mappings between otherwise disconnected forms of information.
The framework further proposes that cognition may operate through processes analogous to compression, decompression, and recompression of informational structures. Human understanding appears capable of reducing complex experiential reality into transferable symbolic, conceptual, or relational forms, then partially unpacking and restructuring those forms when encountering new contexts. Intuition, analogy, expertise, creativity, artistic interpretation, and systems thinking may all emerge from variations of this process.
Higher intelligence may therefore correlate not simply with the quantity of information possessed, but with the ability to efficiently compress, transfer, constrain, decompress, and recombine relational structures across multiple conceptual domains. This may help explain analogy formation, scientific insight, symbolic cognition, latent representation in artificial intelligence systems, and human pattern recognition.
Rather than presenting a complete theory of intelligence, this work is intended as an exploratory conceptual lens—one that may help reframe how cognition, abstraction, understanding, and relational transfer are interpreted across both biological and artificial systems.
⸻
1. Introduction
Most modern discussions of intelligence revolve around measurable outputs:
- memory retention,
- computational speed,
- problem-solving accuracy,
- linguistic ability,
- or mathematical performance.
While useful, these measurements may describe only the visible surface of cognition rather than its deeper organizing mechanism.
A person may possess enormous factual knowledge while lacking the ability to transfer understanding across domains. Conversely, another individual may possess less formal knowledge yet demonstrate remarkable insight through analogy, abstraction, and structural pattern recognition.
This paper explores the possibility that intelligence emerges not primarily from stored information itself, but from the capacity to form non-linear abstract connections between informational structures.
Understanding may therefore emerge not from information accumulation alone, but from the continual restructuring of compressed conceptual models across domains.
In this framework:
- abstraction acts as a relational bridge,
- compression reduces complexity into transferable structures,
- decompression reconstructs contextual meaning from compressed representations,
- recompression generates new relational insight,
- and intelligence governs the validity and usefulness of those operations.
⸻
2. Abstraction as Relational Compression
Abstraction reduces complexity into transferable forms.
For example:
- gravity explains falling apples, planetary motion, and galactic behavior through one underlying principle,
- electrical flow and water flow share structural similarities despite being physically different systems,
- biological evolution and machine learning both involve iterative optimization under constraint.
In each case, intelligence appears to recognize hidden structural relationships beneath surface differences.
This paper proposes that abstraction functions partly as a form of relational compression.
Under this view, cognition reduces large amounts of experiential, sensory, or conceptual complexity into smaller transferable structures capable of being reused across contexts. Human thought may then partially decompress those structures when confronting new situations, allowing relational comparison, prediction, analogy formation, and conceptual transfer.
For example:
- expertise may involve highly compressed experiential models,
- intuition may reflect rapid subconscious decompression of prior relational structures,
- creativity may emerge from recompressing previously disconnected abstractions into novel forms,
- and symbolic systems such as art, mathematics, language, and mythology may function as compressed carriers of meaning.
Thus:
Intelligence may correlate with the ability to detect, compress, transfer, decompress, constrain, and recombine relational structures across increasingly distant conceptual domains.
This differs fundamentally from memorization.
Memorization stores nodes.
Abstraction builds bridges.
Compression makes those bridges transferable.
⸻
3. Linear and Non-Linear Cognition
Linear cognition often progresses sequentially:
A → B → C
Non-linear abstraction may instead operate through distant relational jumps:
A ↔ Q ↔ economics ↔ neural systems ↔ ecosystems
Some of these jumps are invalid.
Others reveal previously hidden structures.
The difference between imagination and intelligence may therefore depend not on the existence of abstraction, but on the validity and usefulness of the connections produced.
This creates an important distinction:
Abstraction itself is not intelligence.
Rather:
abstraction may function as the mechanism through which intelligence explores relational possibility space.
⸻
4. Rosenblatt and Connectionist Foundations
Early neural network models developed by Frank Rosenblatt proposed systems that learned through weighted associations and reinforced pattern recognition.
Perceptrons could:
- strengthen useful associations,
- weaken poor ones,
- and classify information through repeated exposure.
However, these systems primarily operated within relatively local domains of recognition.
The present framework proposes a possible extension:
Intelligence may increase as systems become capable of forming abstract relational mappings across increasingly distant informational terrains.
Under this view:
- low-order cognition detects patterns,
- higher-order cognition detects structural similarities between patterns,
- and advanced intelligence transfers relational frameworks across domains.
This may partially explain why some individuals appear capable of synthesizing knowledge across science, philosophy, engineering, art, and social systems simultaneously.
⸻
5. Analogy and Structural Mapping
Human intelligence frequently operates through analogy.
An analogy is not merely similarity of appearance, but similarity of relationship.
For example:
- the atom and the solar system,
- electrical circuits and fluid systems,
- natural selection and optimization algorithms.
The mind appears capable of recognizing shared structures despite radically different physical manifestations.
This aligns closely with research into analogical cognition and structure-mapping theories, which suggest that relational mapping is central to advanced reasoning.
Under the present framework:
intelligence may scale with the distance and validity of transferable abstraction.
The greater the conceptual distance successfully bridged, the more intelligence appears involved.
5A. Compression, Intuition, and Recombination
Human cognition frequently appears capable of operating through compressed internal representations rather than exhaustive conscious analysis.
For example:
- an experienced electrician may immediately sense that “something feels wrong” before consciously identifying the precise issue,
- a chess master recognizes strategic board structures without calculating every possible move,
- an artist may reinterpret sensory experience into symbolic or emotional form, suggesting that artistic expression itself may function as a recompression of perceptual and emotional experience into symbolic form.
- And a scientist may suddenly recognize deep similarities between previously unrelated systems.
In each case, cognition appears to rely upon compressed relational models acquired through experience.
Under this framework:
intuition may partially reflect rapid subconscious decompression of previously compressed informational structures.
When encountering novel situations, the mind may:
- decompress fragments of prior relational models,
- compare them against incoming information,
- then recompress updated structures into revised understanding.
Creativity may partially emerge when previously separated compressed structures overlap and recombine into new configurations.
This may help explain:
- analogy formation,
- cross-domain insight,
- artistic reinterpretation,
- systems thinking,
- and certain forms of scientific discovery.
Importantly, this framework remains speculative and metaphorical rather than strictly neurological. The proposal is not that the brain literally functions as a digital compression algorithm, but that compression dynamics may provide a useful conceptual model for understanding abstraction-based cognition.
⸻
6. Constraint and Reality
One major danger of abstraction-based cognition is overextension.
Humans are capable of generating:
- false patterns,
- conspiratorial structures,
- mystical over-association,
- and internally coherent but externally invalid systems.
Therefore abstraction alone cannot define intelligence.
A valid model must include constraint.
Reality functions as a filtering mechanism against uncontrolled abstraction.
Thus:
intelligence may depend upon the ability to generate abstract relational models while remaining sufficiently constrained by predictive, explanatory, or empirical coherence.
This balance prevents abstraction from collapsing into fantasy.
Too little abstraction:
- produces rigid thinking.
Too much abstraction:
- produces untethered cognition.
Intelligence may exist in the tension between the two.
⸻
7. Artificial Intelligence and Latent Representation
Modern artificial neural networks increasingly demonstrate behaviors that resemble abstraction transfer.
Large-scale models develop latent relational spaces in which concepts become organized geometrically rather than merely symbolically.
For example:
- “king – man + woman ≈ queen”
is not direct memorization, but relational abstraction encoded within latent structure.
Similarly, transfer learning in machine systems demonstrates that knowledge acquired in one domain may be partially transferable to another.
These developments suggest that:
abstraction-based relational architectures may not be uniquely human, but fundamental to advanced cognition itself.
⸻
8. Silence as Signal
An important implication of this framework is that absence itself may carry informational value.
Silence:
- in conversation,
- in data,
- in astronomy,
- in intelligence analysis,
- or in behavioral systems,
may itself become meaningful.
This reflects a broader principle:
intelligence often detects structure not only in what is present, but in what is absent.
The clue may exist within deviation from expectation.
This idea parallels investigative reasoning, scientific anomaly detection, and predictive systems analysis.
In compression-based cognition, absence may become meaningful precisely because compressed predictive models generate expectations regarding what should be present.
⸻
8A. One major danger of compression-based cognition is overcompression.
Excessive compression may remove nuance, context, contradiction, or uncertainty, producing:
- stereotypes,
- ideological rigidity,
- propaganda,
- conspiratorial thinking,
- or emotionally persuasive but structurally weak abstractions.
Highly compressed symbolic structures often transmit efficiently through culture precisely because they reduce complexity. However, excessive reduction may distort reality rather than clarify it.
Thus intelligence may require not only abstraction, but adaptive decompression—the ability to reopen, inspect, refine, and revise compressed conceptual models when reality no longer supports them.
One major danger of compression-based cognition is overcompression.
Excessive compression may remove nuance, context, contradiction, or uncertainty, producing:
- stereotypes,
- ideological rigidity,
- propaganda,
- conspiratorial thinking,
- or emotionally persuasive but structurally weak abstractions.
9. Conclusion
This paper does not claim to solve intelligence.
Rather, it proposes a speculative reframing:
intelligence may be deeply connected to the ability to generate, transfer, constrain, and validate abstract relational structures across distant domains of information.
Under this framework:
- abstraction is not intelligence itself,
- but a non-linear mechanism for relational exploration,
- while compression and decompression govern the transfer and restructuring of conceptual understanding,
- and intelligence determines which abstractions survive contact with reality.
At minimum, this raises a potentially important question:
Are intelligent minds defined less by the quantity of information they contain, and more by their ability to compress, transfer, decompress, and validly recombine relational structures across seemingly unrelated domains?