Can Computation Explain Nature?

Computation can model nature with astonishing power.

We simulate:

• climate,
• galaxies,
• molecules,
• epidemics,
• brains.

This creates a tempting conclusion:

Perhaps computation can explain everything.

But simulation, prediction, and explanation are not the same.

What Counts as Explanation?

An explanation should tell us more than:

what happens.

It should reveal:

• structure,
• dependency,
• mechanism,
• reason.

A numerical simulation may produce the correct trajectory without making the causal structure conceptually clear.

Prediction Without Explanation

A machine-learning model may accurately predict an outcome.

Yet we may not understand why.

Prediction can succeed without transparent explanation.

This distinction matters throughout science.

Explanation Without Exact Prediction

Conversely, we can understand why a system behaves chaotically while being unable to predict its exact long-term state.

Explanation can exist without precise forecast.

Chaos already taught us this.

Computational Models

A computational model specifies:

• state representation,
• transition rules,
• parameters.

We run the model and observe consequences.

This can expose behavior hidden from analytic mathematics.

Simulation as Experiment on a Model

A simulation is not an experiment on nature itself.

It is an experiment on:

the model.

If the model omits an important mechanism, simulation may be precise and wrong.

Validation

Scientific computation therefore requires validation.

Compare model output with:

• observations,
• experiments.

A simulation earns credibility through contact with reality.

Verification vs Validation

These terms are often distinguished:

Verification

Did we solve the equations or implement the model correctly?

Validation

Is the model an adequate representation of the real system?

A perfectly verified bad model remains bad science.

Levels of Explanation

Nature can be described at multiple levels.

A gas can be explained through:

• molecular collisions,
• thermodynamics.

A brain can be described through:

• neurons,
• cognitive functions.

Computational explanation may occupy one level among several.

Reduction

If a high-level computational process is implemented physically, one might hope to reduce it completely to lower-level physics.

But the higher-level description may remain more useful.

The same algorithm can be implemented in different substrates.

Multiple Realizability

A sorting algorithm can run on:

• silicon,
• mechanical relays.

Its computational identity is not tied to one material.

This gives computational explanation some autonomy.

Functional Explanation

Computational explanations often answer:

What function is being performed?

Examples:

• estimate depth,
• stabilize temperature,
• select an action.

This differs from asking what material mechanism realizes the function.

Marr’s Levels Return

David Marr’s framework is helpful:

Computational level

What problem is being solved?

Algorithmic level

What representations and procedures are used?

Implementation level

How is it physically realized?

A complete explanation may require all three.

Physics

Does computational explanation add anything to fundamental physics?

Sometimes.

Computational complexity can tell us whether predictions are tractable.

Information theory can reveal constraints.

But equations and causal laws remain central.

Cellular Automata

A cellular automaton can generate complex behavior from simple rules.

This shows how computation can explain emergence.

But a cellular automaton model of nature must still be justified empirically.

Biology

In biology, computation can clarify:

• gene regulation,
• neural processing,
• collective behavior.

But life also involves:

• metabolism,
• mechanics,
• evolution.

A purely computational story may be incomplete.

Evolution

Evolutionary algorithms capture some structure of natural selection.

But actual evolution includes:

• historical contingency,
• ecological interaction,
• developmental constraints.

The computational abstraction is useful because it omits detail.

That omission also limits it.

Mind

The computational theory of mind proposes that cognition involves computation over representations.

This explains some capacities elegantly.

But whether consciousness and meaning reduce to computation remains controversial.

Later essays will examine this directly.

Explanation by Compression

One view says explanation compresses observations.

A short program generating a pattern may explain it.

This connects explanation with algorithmic information.

But the shortest description is not always the most illuminating one.

Mechanistic Explanation

Another view emphasizes mechanisms:

• parts,
• interactions,
• organization.

Computational models are explanatory when they identify how states causally transform.

A black-box input-output mapping may predict without mechanism.

Counterfactuals

Good explanations often support counterfactual questions:

What would happen if parameter x changed?

A computational model can be powerful because it allows intervention.

We can rerun alternate scenarios.

Causal Models

Computation can represent causal structure.

But causal meaning does not arise merely from running equations.

We need assumptions about interventions and dependencies.

Correlation remains insufficient.

Emergence

Computation is especially good at demonstrating how:

simple local rules

can produce:

complex global behavior.

This is an explanatory achievement.

We can watch emergence unfold.

Computational Irreducibility

Sometimes the only way to know a system’s state is to run the computation.

Then the model may reproduce behavior without compressing it.

Does mere replay count as explanation?

Perhaps partly.

But explanation normally seeks something more general.

Explanation as Mechanism vs Prediction

Suppose a simulation exactly reproduces every atom in a tornado.

It predicts perfectly.

But a fluid-dynamics explanation using:

• pressure,
• vorticity,
• convection

may be more understandable.

Higher-level concepts can explain better than exhaustive detail.

The Map Is Not the Territory

A computational model is a map.

It selects:

• variables,
• scales,
• rules.

Even a highly accurate simulation remains representation.

Do not confuse fidelity with identity.

Is Nature Itself an Algorithm?

If digital physics were correct, computation might be ontologically fundamental.

Then computational explanation would be unusually deep.

But this is not established.

We should not assume the conclusion.

Explanation by Universal Simulation

A universal computer can simulate any computable process.

But universality alone does not imply understanding.

A universal interpreter can run a model without explaining the model’s domain.

General computational power is not general scientific explanation.

Human Understanding

An explanation is partly epistemic.

It must help a reasoner:

• organize knowledge,
• see dependencies.

A raw trillion-step trace is computationally complete but cognitively useless.

Compression and abstraction matter.

Interpretability

This is why interpretability matters in machine learning.

A model may be accurate.

Scientists may still want:

• salient variables,
• causal relations,
• simplified mechanisms.

Explanation is not identical to output.

Computer-Aided Discovery

Computation can reveal patterns humans did not anticipate.

Examples include:

• conjecture generation,
• molecular design,
• numerical exploration.

The machine can extend inquiry.

Interpretation remains necessary.

Computation as a Third Method

Modern science often combines:

• theory,
• experiment,
• computation.

Computation occupies a distinct methodological role.

It explores consequences too complicated for closed-form analysis.

Does Computation Replace Theory?

No.

A simulation needs:

• equations,
• rules,
• parameters,
• initial conditions.

Theory determines what is simulated.

Computation amplifies theory.

It does not eliminate it.

Does Computation Replace Experiment?

No.

Experiment anchors models to the physical world.

Without observation, simulations can become internally consistent fictional universes.

Computation and experiment need each other.

Can Everything Be Computed?

No.

Computability theory already gives formal limits.

Even if nature were computational, some questions about computational systems can be undecidable.

Computational explanation contains its own boundaries.

Can Everything Computable Be Explained?

Also no.

An enormous computation may produce an answer with no concise conceptual account.

Calculation can outrun understanding.

The Strongest Reason to Use Computation

Computation gives us a new form of intellectual microscope.

It lets us explore systems where:

• many interactions,
• nonlinear dynamics,
• stochastic behavior

make intuition insufficient.

It extends our reach.

The Danger

Because computational models produce exact-looking numbers and visualizations, they can create false confidence.

Precision of output is not accuracy of assumptions.

A model can be precisely wrong.

The Philosophical Lesson

Computation can explain nature when it reveals:

• mechanisms,
• representations,
• causal structure,
• emergent behavior,
• computational constraints.

It cannot automatically turn simulation into understanding.

Computation is one of our deepest explanatory tools.

It is not obviously the only one.

The End of Part XIII

We began by asking what computation is.

We discovered:

• algorithms,
• information processing,
• computability,
• complexity,
• biological and quantum computation,
• computational views of nature.

The next step is inevitable.

One physical system seems especially connected to computation:

the brain.

The Next Question

The brain produces:

• perception,
• memory,
• action,
• perhaps mind itself.

Before asking whether a brain is a computer, we should understand what a brain is.

The next part begins with:

What Is a Brain?