Higher-Level Causation

Mental explanations describe organisms at a higher level than cellular and molecular physics. The fact that a route is believed unsafe explains why a traveller turns back; the exact microphysical state varies across people and occasions. Higher-level causation asks whether such variables are genuine causes or merely useful summaries of lower-level causal processes.

The issue is general. Temperature, genes, software states, ecosystems, and economic variables are physically realised yet figure in causal explanation. These analogies show that derivation does not entail causal irrelevance. They do not automatically vindicate mental causes, especially irreducible phenomenal properties. A successful account must state what makes a higher-level variable difference-making, mechanistically grounded, and appropriately proportional to its effect.

Levels without layers of stuff

Explanatory levels need not be ontological floors made of different substances. They can be different ways of grouping states according to stable patterns and interventions. A gas's temperature is realised by molecular motion; a program state by electronic activity; a belief by neural and bodily organisation.

The relation is many-one:

$$P_1,P_2,\dots ,P_n\mapsto M,$$

where different physical states realise one higher-level kind $M$. The higher level abstracts from differences irrelevant to the regularity under study.

This abstraction can increase explanatory power. Knowing every molecular coordinate is less useful for predicting pressure than knowing temperature. The question is whether that usefulness corresponds to causation or only to limited human computation.

Difference-making

Counterfactual accounts say $C$ causes $E$ when, under suitable conditions, had $C$ not occurred, $E$ would not have occurred. Higher-level mental variables can satisfy this: had the subject not believed the bridge unsafe, they would not have turned around.

Simple counterfactual dependence fails under preemption and redundancy, and the counterfactual "same physical state but different belief" may be impossible if belief supervenes on the physical. The relevant comparison changes the realiser as needed to change the belief while holding other causally relevant factors fixed.

This is not cheating. Interventions on temperature also alter molecular states. A variable can be causally useful even when intervention must proceed through its base.

Interventionism

On an interventionist account, $M$ causes $B$ when ideal changes to $M$ change $B$ while blocking alternative routes and confounders. A mental intervention might supply new evidence, alter a goal, or induce a perceptual state through different physical means.

Multiple realisation helps. If changing belief content across varied neural realisers produces a stable change in action, the belief variable captures a robust causal pattern not tied to one implementation.

Interventions must be coherent. Wide content cannot always be changed surgically while holding the subject and environment fixed. Rational relations also link beliefs: changing one attitude may require wider adjustment. Interventionism therefore works best with carefully specified contrast classes rather than arbitrary mental switches.

Proportionality

Yablo argues causes should be proportional to effects: specific enough to explain the effect, but not burdened with irrelevant detail. A pigeon's peck is caused by the stimulus being red rather than its exact scarlet shade if any red shade would have produced the response.

Likewise, the exact neural microstate may be too specific. The belief that food is in the left box predicts search across many realisers. The mental cause matches the range of conditions under which the effect occurs.

Proportionality answers causal exclusion by denying that the lowest-level sufficient condition is always the best cause. It is a criterion of explanatory relevance, though critics question whether it establishes a distinct metaphysical causal power.

Mechanistic realization

A higher-level cause should be connected to a mechanism. A software bug causes failure because it is implemented in instructions that control execution; a belief causes action through perceptual, memory, valuation, and motor processes.

Mechanistic explanation links levels through decomposition:

graph TD
  M[Belief-level variable] --> A[Action-level outcome]
  M -. realised by .-> N1[Neural population state]
  N1 --> N2[Decision and motor processes]
  N2 -. realises .-> A

The upper arrow summarises a stable dependency implemented by the lower path. It is not an extra force. The model also permits failure: if no mechanism links the attributed content to action, the attribution may be predictive shorthand rather than a cause.

Causal realization

Some theories identify a higher-level property's powers with a subset of its realiser's powers. The belief is causally efficacious because the realising neural state possesses the organised powers definitive of that belief. No power is duplicated.

This position blurs non-reduction. If mental kinds are nothing beyond functional causal profiles, the account resembles reductive functionalism. Defenders distinguish identity with one physical type from realization by variable physical types. Reduction across levels need not be type identity.

The debate is partly over what autonomy requires. Distinct fundamental powers are not needed for explanatory and causal reality; they may be required for robust property dualism.

Exclusion revisited

The exclusion argument assumes causes at different levels compete for one causal slot. Higher-level theories offer three denials:

1. same-instance: mental and physical causes are one process under different descriptions;
2. realization: the higher-level causal relation is constituted by lower-level relations rather than added to them;
3. contrast relativity: mental and physical explanations answer different questions and select different relevant alternatives.

These responses work poorly for a mental property claimed to possess a novel power beyond its realiser. They work well for functional, representational, and organisational properties. The causal prospects of "the mental" therefore vary by mental kind.

Software

Software is a common analogy. A program causes a machine to print an error even though every transition is electronic. The software description supports debugging, counterfactuals, and interventions across different hardware.

The analogy helps with cognition and content where implementation structure is explicit. It is less decisive for phenomenal consciousness. Software properties are defined by functional organisation; whether qualia are so defined is exactly what absent-qualia arguments dispute.

There is also a danger of deriving metaphysics from explanatory convenience. A bug can be real as a pattern without being an additional force. That model supports non-reductive explanation, not irreducible mental powers.

Genes and biological functions

Genes cause phenotypic effects in environments through molecular mechanisms. "Gene for $T$" abstracts from complex pathways and can be causally useful when the contrast is specified. Biological function likewise depends on selection history and organisation.

Mental contents may resemble these relational properties. Their causal relevance can depend on environment, history, and consumer systems rather than one intrinsic molecule. This counters the objection that wide or normative properties cannot be causes.

The analogy also warns against unrestricted attribution. Gene-level claims become misleading when context and interaction are ignored. Mental causal claims require the same discipline about background conditions and level.

Temperature and multiple realization

Temperature is multiply realised by different molecular microstates and supports robust causal laws. It demonstrates that higher-level variables can be objective, measurable, and reducible without matching one microstate.

Mental states differ because their individuation may involve rational norms and phenomenal character. Still, the example blocks a simple inference from many realisers to non-physicality or causal impotence.

The right question is whether a mental variable participates in similarly stable laws, mechanisms, and interventions. Some perceptual and computational states may; broad folk categories may fragment; phenomenal qualities may require another account.

Explanatory pluralism

Different levels answer different why-questions:

• the neural explanation says how a movement was implemented;
• the computational explanation says which information transformation occurred;
• the reason explanation says why the action made sense to the agent;
• the social explanation says which norms and institutions shaped the choice.

These explanations can be compatible without all being equally causal in every context. Pluralism should not become indiscriminate. Explanatory variables need invariance, counterfactual support, and integration with mechanisms.

Assessment

Higher-level causation gives mental explanation a defensible place without adding a second causal stream. It succeeds most clearly where mental kinds are functional and mechanistically tractable. It does not settle whether irreducible phenomenal properties possess powers, and it does not show that every useful psychological category marks a cause.

Selected references

• Craver, Carl F. Explaining the Brain (2007).
• List, Christian and Peter Menzies. "Nonreductive Physicalism and the Limits of the Exclusion Principle" (2009).
• Woodward, James. Making Things Happen (2003).
• Yablo, Stephen. "Mental Causation" (1992).