Beyond Symbolic Mind: Re-evaluating the Logical Model of Intelligence

Egil Diau
1National Taiwan University
Paper Zenodo Homepage
Project teaser

TL/DR:   Symbolic computation is often criticized for weak optimization and generalization, but its core failure is structural: it requires every intermediate condition to co-exist and every operation to be world-instantiable, so a single failed grounding collapses the entire chain. Human cognition never seeks such total realizability; it works through partial information and non-exhaustive specification, which is why symbolic inference is not how thought operates, but a formal ideal dependent on conditions the world does not provide.

Symbolic Inference Origins

Symbolic Inference: From Rationality to AI
Symbolic logic entered early AI not as one technique, but as the inherited model of rational thought from philosophy. It treated intelligence as explicit operators and fully specified chains of deduction rather than world-bound, partial cognition.

Symbolic AI Collapse

Symbolic AI: Cold Winter, Wrong Explanation
When symbolic systems failed under variation and could not generalize, the collapse was blamed on optimization and scaling. Yet the deeper limit is structural: long-form inference assumes jointly realizable conditions and world-instantiable operations the world rarely sustains.

Human Cognition Differentiation

Human Cognition: Partial, Not Symbolic
Human cognition succeeds not by completing chains, but by never requiring total realizability. It operates through partial information, underspecified states, and pattern-based inference—making it fundamentally distinct from symbolic logic rather than a weaker version of it.

The Two Faces of Symbolic Validity

Long-chain symbolic inference only works when two world-level requirements are met at once.

🧱

Jointly Realizable Conditions

Every condition must have a realizable joint event in the world. If the intersection cannot exist, the entire chain loses validity.

No realizable joint event → no valid inference

⚙️

World-Instantiable Operations

Each operator must map to a possible world transition. If no mechanism exists, the step is only syntactic.

No mechanism → no valid transition

The Core Bottleneck of Symbolic Inference

Symbolic inference requires extensive, fully specified conditions before it can operate. But once such information is in place, its growth becomes purely combinatorial: the system expands possible assignments without generating new structure or modifying the predicates themselves. More information does not produce more understanding—it produces more permutations of what is already assumed.

Where Chains Actually Break

Symbolic inference rarely fails by contradiction—it fails when the world cannot maintain the continuity that syntax presumes.

  • Multi-causal origins – A does not yield B under a single mechanism, but only through shifting or compound causes.
  • Multi-effect pathways – A branches into many outcomes, but symbolic form collapses them into one directed consequence.
  • Mechanism drift – A→B holds under one world structure, B→C under another; syntax composes them as if continuity holds.
  • No shared instantiation space – predicates are valid individually but have no realizable joint state in the world.
  • Operation without implementation – a transition is permitted in form but not supported by an actual world mechanism.

Abstract

Symbolic logic has long been treated as a model of intelligence, motivating rule-based inference and classical approaches to artificial intelligence. Yet systems built on explicit symbolic derivation have proven difficult to optimize, brittle under variation, and unable to generalize beyond tightly specified domains.

These limitations are not merely engineering obstacles. The validity of long-chain symbolic inference depends on conditions that the world rarely affords: each step requires jointly realizable states and transition relations that must remain constructible throughout the chain. When any such condition cannot be instantiated, collapse is systemic rather than local.

Human cognition, by contrast, operates under partial information, indeterminate predicates, and incomplete state specification, functioning without the requirement that all intermediate relations be jointly defined. This work re-evaluates the long-standing assumption that logical form constitutes the operative architecture of intelligence and clarifies why symbolic inference, though formally coherent, cannot serve as an operative account of cognition.

Why It Matters?

For a century, intelligence was equated with long-form symbolic reasoning—depth, precision, multi-step logic. Its image shaped philosophy, early AI, and even contemporary models of cognition. Yet almost no one examined its world-level requirements: that joint events must exist and operations must be realizable. Once those conditions are confronted, the framework collapses—not because logic is incorrect, but because the world cannot supply what it presupposes.

Cite This Work

@misc{diau_2025_17858758,
  author       = {Diau, Egil},
  title        = {Beyond Symbolic Mind: Re-evaluating the Logical
                   Model of Intelligence
                  },
  month        = dec,
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17858758},
  url          = {https://doi.org/10.5281/zenodo.17858758},
}