Mind, Matter, and Meaning

Information and Meaning

Information and meaning overlap in ordinary language, but distinguishable notions do different work in communication theory and philosophy. A signal can reduce uncertainty without expressing a claim. A meaningful claim can be false. An account of semantic information must explain how these two possibilities fit together.

Statistical Information

Claude Shannon’s mathematical theory concerns the probabilities of messages and the capacity of channels to transmit them. Information, in this setting, measures uncertainty and its reduction. It does not measure understanding, importance, or truth. An unpredictable sequence can carry more Shannon information than a familiar sentence while making no linguistic sense.

A thermometer illustrates the distinction. Under suitable conditions its reading reduces uncertainty about temperature. The same reading can also mean that the temperature is twenty degrees, because a measurement practice assigns it that content. The statistical relationship and the semantic interpretation need not have the same explanation. Similarly, smoke may provide evidence of fire without asserting that there is a fire.

Shannon’s framework deliberately abstracts from semantic questions. This does not show that meaning is unreal or irrelevant; it identifies what a theory of transmission can investigate without already solving the problem of interpretation.

Semantic Content and Error

Semantic content concerns what a word, thought, picture, or other representation presents as the case. A belief that the kettle is boiling has a condition of truth even when the kettle is cold. A map can depict a bridge that has been demolished. The state remains meaningful while getting its subject wrong.

Terminology requires care. Some theories reserve semantic information for true, meaningful content; on that usage misinformation is not information. The broader expression semantic content can include false representations. Neither convention should be confused with the claim that a statistical signal itself has beliefs or understands what it indicates.

The central philosophical problem is therefore not just how states covary, but how one condition becomes the standard against which a state counts as correct. A detector activated by both flies and small dark pellets raises a familiar question: does it represent flies, small moving objects, or something more inclusive? Correlation alone may leave several answers available.

Dretske’s Informational Approach

In Knowledge and the Flow of Information (1981), Fred Dretske develops an account connecting information, knowledge, and representation. His notion of an information-bearing signal is stronger than a loose statistical association: relative to relevant background conditions, the signal indicates that a source has a particular property.

The difficulty is that genuine indication is factive, whereas beliefs can misrepresent. Dretske’s later work gives learning and acquired function a central role. A state recruited to indicate a condition can continue to represent that condition when it is produced abnormally. The history that established its use helps distinguish its content from every cause that can subsequently trigger it.

Teleosemantic theories pursue related explanations through biological selection and producer-consumer organization. A signal-producing mechanism and the mechanism that uses its output may have coordinated functions that help fix what successful representation requires. Such theories aim to explain error without treating semantic interpretation as an unexplained starting point.

Objections and Alternatives

Indeterminacy. A mechanism’s history may privilege more than one description. A signal used to secure nutrition might track food, a chemical feature, or a reliable visual cue. Functional theories reply that the consumer’s success conditions and the circumstances explaining selection constrain the answer. Critics ask whether these constraints yield enough determinacy for ordinary thought.

Novel content. People can think about situations never encountered, and representations can be recombined. A theory built on learned indicators must explain how its basic contents support this productivity. Advocates distinguish the acquisition of components from the rules governing their combinations; critics dispute whether the resulting account captures all semantic relations.

Interpretation dependence. A string of bits can receive indefinitely many interpretations. Some philosophers emphasize the interpreter’s practices; others argue that causal-functional organization can privilege content independently of a particular interpreter. The useful question is what makes a proposed interpretation explanatory, rather than merely available.

Inferentialist, use-based, and externalist approaches supply other resources. They stress relationships among commitments, public standards of use, or connections with environmental objects and social practices. These approaches can complement informational explanations, although they disagree about which relations are constitutive.

Artificial Systems

A language model’s sensitivity to textual distributions establishes neither a purely nonsemantic verdict nor unrestricted understanding. Textual distributions are themselves possible targets of representation. Whether particular internal states represent them depends on the adopted account of content and on evidence about training and use. Having content concerning words also differs from referring to a particular distant object or understanding an entire conversation.

Mechanism-level representation, linguistic competence, whole-system understanding, and conscious experience remain distinct questions. Information theory alone does not settle any of the stronger attributions. Nor does the fact that a system was engineered establish that every possible content attribution must depend entirely on a human interpreter.

Further Reading

  • Shannon, Claude E. “A Mathematical Theory of Communication.” Bell System Technical Journal 27 (1948): 379–423, 623–656.
  • Dretske, Fred I. Knowledge and the Flow of Information. MIT Press, 1981.
  • Dretske, Fred I. Explaining Behavior: Reasons in a World of Causes. MIT Press, 1988.