Naturalizing Mental Content

To naturalize content is to explain representation using relations acceptable within a naturalistic account of organisms and physical systems: causation, information, biological function, learning, and use. The aim is not merely to correlate a neural state with what subjects say it means. It is to explain why the state represents one condition rather than another, can occur when that condition is absent, and is correct or incorrect.

The central obstacle is misrepresentation. Causal covariance readily explains why a state carries information in normal conditions; it tends to redefine abnormal tokens as information about their actual causes rather than errors. Teleological theories add a function that fixes what the system is supposed to indicate. They gain error at the cost of disputed functions, indeterminacy, and difficult cases outside evolution.

Causal covariance

The simplest proposal is that state $R$ represents condition $C$ when instances of $C$ reliably cause instances of $R$:

$$C⟶R.$$

A retinal or neural state covaries with light of a wavelength; a frog's detector fires for small moving dark objects; a thermometer's state varies with temperature. Causal connection makes representation world-involving rather than assigned by an external interpreter.

Covariance is too abundant. Tree rings covary with age, smoke with fire, and shadows with objects. Not every correlation is a mental representation. The theory needs to distinguish a representation from a natural sign and identify which among many correlated conditions is its content.

Information

In an information-theoretic sense, $R$ carries information about $C$ when occurrence of $R$ reduces uncertainty about $C$. Under ideal deterministic conditions, $R$ may carry the information that $C$ occurred. This notion is objective and applies far beyond minds.

Fred Dretske proposed that information can ground content when a system learns or is recruited to use an indicator in controlling behaviour. Raw information becomes representation through a functional role.

The disjunction problem appears when a frog snaps at a pellet. If its detector is caused by flies in normal cases and pellets in abnormal cases, does it mean fly, small dark moving object, fly-or-pellet, or food-like target? Pure covariance favours the disjunction containing every cause and thereby removes error.

Information is also factive in some formulations: a signal cannot carry the information that $C$ when $C$ is false. Representation must be capable of false content, so informational content cannot simply be identical with semantic content.

Asymmetric dependence

Fodor's asymmetric dependence theory attempts to solve disjunction. A token caused by a pellet counts as a false fly representation because the pellet-to-state connection depends on the normal fly-to-state connection, while the fly connection does not depend on the pellet connection. If the system did not respond to flies, it would not mistake pellets for them.

Formally, the law

$$Pellet\to R$$

depends on

$$Fly\to R,$$

but not conversely. The asymmetry fixes fly rather than fly-or-pellet as content.

Counterexamples arise when causal dependencies are symmetric, historically contingent, or unclear. A detector may have been trained on artificial stimuli and only later encounter the natural kind; novel errors may not depend counterfactually on a normal connection. The theory also handles primitive concepts more readily than structured thoughts and abstract contents.

Teleosemantics

Teleosemantics grounds content in biological function. A state represents what it was selected to indicate for the sake of a consumer mechanism. The frog's state means prey, on one version, because ancestral instances were recruited to guide feeding and contributed to fitness when prey was present.

Function separates actual cause from proper content. A pellet can trigger the state, but the state malfunctions because it fails to indicate the condition its system was selected to exploit. Error becomes possible through a gap between current tokening and proper function.

The account needs both producer and consumer. What matters may not be what normally causes the signal but what downstream systems use it for. A bee dance can represent food location because other bees consume the signal to navigate; causal covariance alone underdetermines which correlated feature matters.

Etiological and systemic function

An etiological function is fixed by selection history: a trait's function is the effect for which ancestors of the trait were selected. This grounds normativity in success and failure relative to history.

A systemic function is the contribution a component makes to a current organised capacity. It can apply to learned, engineered, or novel systems without an evolutionary history, but risks making function observer-relative: a component contributes to many capacities depending on which system and goal are selected.

The contrast matters for artificial and socially learned representations. A newly built system has design and training history but no direct biological selection. One can extend teleology to learning, reinforcement, and design; doing so may make content derived from creators or users rather than original to the system.

Consumer-based content

Ruth Millikan emphasises how a representation is used. A producer state has content because consumer mechanisms perform their proper functions successfully when the world matches a mapping from representation to condition.

This handles indeterminacy better than source-only theories. A signal may covary with both flies and retinal patterns, but the motor consumer needs prey location rather than retinal activation to succeed. Content follows the condition relevant to consumer function.

Consumers can themselves be described at several levels. A motor routine succeeds in capturing food, receiving reinforcement, or preserving fitness. Different success conditions generate different contents. Teleosemantics therefore narrows but may not eliminate indeterminacy.

Misrepresentation

A naturalistic theory should distinguish at least three failures:

• false token: the system represents $C$ when not-$C$;
• malfunction: the producer or consumer fails to perform its proper function;
• misleading but normal signal: the representational mechanism works as designed in an unusual world and produces a false content.

Error cannot require malfunction in every case. A perfectly functioning perceptual system can be fooled by an unusual but valid stimulus. Teleosemantics can say the system performs its representational function while the represented condition fails, provided content and success condition are independently fixed.

The account must also permit lucky truth: a malfunctioning process may accidentally produce a correct representation. Correctness and proper operation are related but not identical.

Indeterminacy

Natural relations usually fix content only approximately. Does a frog state represent flies, edible moving objects, nutrition, or a region of motion-feature space? Evolution may not discriminate among coextensive conditions in the ancestral environment.

One response accepts content indeterminacy. Mental content can be as determinate as behaviour and selection require without always selecting one proposition. Another appeals to richer inferential and consumer roles. A third treats public language as sharpening contents that underlying cognition leaves coarse.

Indeterminacy is not automatically fatal. Scientific quantities and biological kinds can have vague boundaries. It becomes problematic when a theory needs precise content to explain systematic inference or rational error.

Abstract and novel content

Causal and teleological accounts fit perception of ecologically important conditions better than thought about prime numbers, distant futures, or newly invented entities. No ancestral selection history targeted quarks or constitutional law.

Compositionality offers one route: primitive contents are naturally grounded, and structural rules combine them into novel thoughts. Abstract concepts may be fixed by inferential role within a public practice rather than direct causal tracking. This suggests a hybrid theory rather than one relation covering every content.

Novel concepts also arise through stipulation, measurement, and theory construction. Their content depends on how they connect to older concepts, instruments, and communal norms. Naturalisation must include social and cognitive history, not only direct environmental covariance.

Normativity

Content supports norms: beliefs can be true, inferences valid, concepts misapplied, and reasons good or bad. Can causal and evolutionary facts generate this semantic normativity, or do they merely explain why a system behaves as it does?

Teleosemantics derives a thin ought from proper function: the state is supposed to indicate what enabled the selected system to succeed. Critics argue that evolutionary success is not truth and that a trait can perform its selected function while producing irrational beliefs. Defenders distinguish basic representational correctness from the richer epistemic norms of reflective agents.

Inferential and social practices may supply the richer norms, but then naturalisation extends beyond individual biology. A complete theory may layer natural signs, biological representation, learned concepts, and public reasons rather than derive all normativity at once.

Representation without consciousness

Naturalistic accounts deliberately make representation independent of phenomenal consciousness. Retinal circuits, motor controllers, and unconscious predictions can carry content through their causal and functional roles. This explains why intentionality and consciousness come apart.

The advantage becomes a limitation if original intentionality is thought to require conscious understanding. A theory can explain correctness conditions from outside while leaving open whether anything means the content for the system. The later page on consciousness and content compares this phenomenal-intentionality challenge with the separatist view.

Assessment

Naturalisation succeeds most clearly for basic biological representation. Its harder task is scaling from tracking and control to fine-grained, compositional, socially shared, and normatively assessable thought. The likely alternatives are a layered account or a stronger claim that inferential or phenomenal role supplies what causal history leaves open.

Selected references

• Dretske, Fred. Knowledge and the Flow of Information (1981).
• Fodor, Jerry A. A Theory of Content and Other Essays (1990).
• Millikan, Ruth Garrett. Language, Thought, and Other Biological Categories (1984).
• Neander, Karen. A Mark of the Mental (2017).
• Papineau, David. Reality and Representation (1987).