Sense and Reference

In the [YOU] on AI Field Guide

The [YOU] on AI cycle is organized around the question of what these machines actually are—what is genuinely present in them and what is genuinely absent, stated precisely rather than impressionistically. Sense and reference is the conceptual instrument that makes that precision possible for the dimension of meaning. Where other critics of AI gesture at “understanding” or “consciousness” or “real intelligence,” Frege's distinction lets us name the specific thing that is missing: not understanding in some vague holistic sense, but reference specifically—the terminus of meaning in the world, the moment at which an expression connects to an actual object rather than to other expressions. A system that has mastered sense without reference has mastered half of meaning. That half is real and should be acknowledged as such. The other half is also real, and its absence explains the characteristic failure modes of these systems.

The fluency-authority decorrelation—the way a language model can write a confident sentence that states something false, or hallucinate a case citation with serene assurance—is exactly what Frege's framework predicts. A system that has no reference against which to check a sense has no world to consult, only the consensus of the text it was trained on. It optimizes for what humans tend to write, which is correlated with truth but is not the same thing, and when these diverge the system has no independent purchase. Its sense is real; its truth-conditions are not checked against the world. Frege spent his life insisting that truth and mere acceptance are different things. A machine trained on acceptance has, by its design, no independent grip on truth.

The engineering response to this—multimodal training that ties words to images, robotic systems that tie words to physical actions—is a deliberate attempt to supply the missing reference, to connect the sense-structure of language to something outside language. Whether this achieves reference in Frege's strict sense or merely enriches the web of senses with additional modalities is the live and genuinely unresolved question. Frege's own account complicates the skeptic's position: he never claimed reference was achieved by direct acquaintance, only that sense was the route by which a mind reaches an object, not the object itself. If reference is more mediated and social than the grounding skeptic allows, the machine's situation may be less categorically different from ours than it first appears.

Origin

The puzzle Frege solved had haunted the philosophy of language since antiquity but had never been given a technical treatment. Why is “the morning star is the evening star” informative, while “the morning star is the morning star” is empty—if both names refer to the same object? Earlier accounts, including those that treated meaning as simply co-reference, had no answer. Frege saw that the identity statement must be carrying information about the two different modes of presentation—the two different cognitive routes—by which the same object had been reached. The discovery that they converged on one object was the discovery that two senses, previously held independently, shared a reference. Meaning had two dimensions, not one.

He extended the distinction from names to complete sentences, and here the framework became philosophically momentous. A complete declarative sentence, Frege argued, has as its reference a truth-value—the True or the False—and as its sense the thought it expresses, the objective propositional content that is neither physical nor mental, available to any mind that grasps it. To fully understand a sentence is to grasp the condition under which it is true: to connect its sense to how the world would have to be. This sets a demanding bar that purely statistical systems cannot clear by their architecture alone, since no statistical summary of what humans write can substitute for the condition under which a proposition is true of the world.

Key Ideas

The asymmetry of sense and reference. Many senses can share one reference (all descriptions of Aristotle point to the same man), and a sense can exist without a reference (descriptions of fictional or impossible objects). The asymmetry explains the informativeness of identity statements and the coherence of fiction; it also explains how a language model can be semantically rich while referentially empty: having many senses internally organized does not entail reaching any object.

The symbol grounding problem as Frege's problem. The symbol grounding problem—how symbols inside a computation acquire genuine meaning rather than pointing only to other symbols—is Frege's sense-reference problem restated by cognitive science. Both identify the same structural predicament: an internal system of relations, however elaborate and coherent, does not by itself constitute world-contact. Reference must be added, and where it comes from—embodiment, perception, action, social anchoring—is the open empirical question.

Sense without reference as half of meaning. The honest assessment, on Frege's framework, is that large language models have achieved something genuinely unprecedented: a vast competence in the dimension of sense, far exceeding what any prior technology could manage. This is not nothing; it is half of meaning. The other half—reference, the terminus of meaning in the world, and truth, the terminus of the sentence in fact—is what the systems lack and what explains their characteristic failure at the boundary between what has been written and what is actually the case.