Teleosemantics explains representational content through selected function and producer-consumer organization. What a state represents depends partly on what states of that kind were stabilized to do, not simply on whatever happens to cause the state now.
The theory’s central achievement is to make error intelligible. Smoke correlates with fire and tree rings with age, but correlation alone does not say what a signal is supposed to get right. A producer-consumer history can privilege one correspondence rule over its rivals and explain how a state can be false even while remaining causally effective.
The Frog
A frog’s visual circuitry responds to small dark moving things. Does its state represent flies, edible moving prey, or small dark moving objects? Current cause cannot decide. Teleosemantics asks which mapping mattered to the stabilization and successful use of the producer-consumer organization.
The answer may remain coarse. If prey at this location and fly at this location explain the organization equally well, the grain may be indeterminate while both readings exclude a BB pellet. If nutritive moving prey and small dark moving thing remain tied, the extension is unsettled and so is the verdict that the BB was misrepresented. Relevant interventions help: hold darkness fixed while varying nutritiveness, then reverse the test. A mapping earns content when such differences predict consumer success and explain the organization independently of an observer’s preferred vocabulary.
The content belongs first to the producer-consumer mechanism. That does not turn the mechanism into a tiny frog or settle whether the whole animal perceives a distal object. Burge’s objectification threshold and teleosemantic content answer different questions.
Function, Content, and Level
An etiological function depends on history: doing F helps explain why a trait or state type was selected, learned, or differentially retained. A correspondence rule maps a family of producer states onto conditions that consumers use. The history makes that rule explanatory rather than optional.
Three questions must remain separate:
- Does the producer-consumer mapping fix a correctness condition?
- At what level should that content be attributed?
- What larger capacities does the system’s use of the content justify attributing?
Selection and learning may answer the first question at a subpersonal or mechanism level when the success condition itself fixes what the states concern. They can also merely internalize a semantic standard supplied by an interpreting practice. Wider coordination across memory, inference, learning, conflict resolution, and action may justify system-level belief or understanding. It does not create the component’s content.
Ownership Withdrawn as a Semantic Condition
The book previously added a proprietor requirement: an owner had to appropriate a historically shaped state before its content could become original. That requirement is withdrawn. It answered a question about whose state or capacity we should attribute to the larger system, not what makes the state accurate or inaccurate.
Self-maintenance remains important to theories of autonomy and agency. It may identify an individual with practical interests and explain persistence or welfare. It does not supply a missing semantic relation. Add a self-repair loop while holding the producer-consumer mapping fixed and the represented condition need not change.
Major Positions
Millikan. Proper functions belong to historically reproduced producer-consumer systems. Consumer use helps determine satisfaction conditions and makes misrepresentation possible.
Dretske. Learning can recruit an informational indicator for a use. The acquired function lets representational content diverge from present correlation.
Papineau and Neander. Their accounts develop the role of successful action, selected function, and natural information in fixing nonconceptual content.
Hutto and Myin. Radical enactivists deny that covariance and function yield truth-conditional content. The reply begins with a nonsemantic job description: one distal-condition-sensitive variable remains portable across consumers, recombines with changing inputs, and gets corrected at the mapping even when final payoff is rescued elsewhere. Ward’s decouplability test helps distinguish that organization from a one-link controller. Hutto and Myin can still call the richer system teleosemiotic rather than semantic. The book takes the history-privileged correspondence to constitute a satisfaction condition because it unifies portability, recombination, correction, and error under one rule. The disagreement concerns that semantic verdict, not a hidden proprietor.
Mallory. Training supplies a selection history for word2vec’s producer-consumer mechanisms. On his account, an embedding represents a fact about a word-type and its contexts while directing the downstream mechanism to produce a distribution over context words. Actual linguistic distributions, rather than the numerical code alone, supply the target against which training corrects the output. He argues that these states carry original content and explicitly treats linguistic relations as part of the world. This claim concerns mechanisms and their linguistic targets, not whole-model understanding or reference to the objects named by the words.
Swampman
Swampman at the instant of creation lacks the history teleosemantics uses to fix content. His self-maintaining organization cannot substitute for that missing history. A molecule-for-molecule duplicate can therefore lack teleosemantic content at t = 0 even though he immediately has the machinery needed to begin acquiring it.
A trained language model presents a different profile: it has a training history, but a text-prediction history may internalize semantic norms inherited from language rather than originate them. It may also lack perception, distal reference, integrated belief, or understanding. The contrast shows why causal history, semantic independence, autonomy, and system-level mentality must not be collapsed into one test.
My View
A family of states, a correspondence rule, and a producer-consumer history can establish content at the grain the explanatory organization supports. It counts as original to the extent that history and use fix accuracy conditions not exhausted by an interpreting practice. The mapping must explain stabilization and survive relevant interventions. Mallory’s account of content about actual word-types and distributions deserves that test; originality at this limited grain remains unsettled. The vocabulary, corpus, task, and loss have a human ancestry, but ancestry alone does not show that their interpreting practice exhausts the mechanism’s accuracy conditions. A coordinated reparameterization can preserve the word-to-context relation; changing the external pairing changes a candidate content-fixing relation. Neither maneuver alone refutes contextual content or settles its independence. No further proprietor relation turns content on.
Content at one mechanism remains compatible with ignorance at the level of the whole. The book’s later account must earn perception, belief, understanding, and phenomenal attribution through the organization and poising of content-bearing states rather than through ownership as a semantic ingredient.