Can a System Contain a Model of Itself?

A system can model the world.

Can it also model itself?

A thermostat tracks temperature.

A robot tracks battery level and joint position.

A brain tracks some of its own beliefs, intentions, and bodily states.

Self-modeling appears everywhere once systems become sufficiently complex.

But a deeper question follows:

Can a system contain a complete and perfectly accurate model of itself?

Partial Self-Models Are Common

Most useful self-models are partial.

A robot may represent:

• battery charge,
• wheel position,
• damage state.

It does not need a molecular simulation of every transistor.

A model is useful because it omits.

The Model Is Smaller Than the Target

A practical model compresses.

If a map reproduced every atom in the territory, it would no longer be a useful map.

The same applies to self-models.

A system usually represents only variables relevant to control.

Self-Monitoring

Computers routinely monitor themselves.

Operating systems track:

• memory use,
• processor load,
• active processes.

This is a form of self-representation.

The system contains data about its own current state.

Reflection

Some programming systems support reflection.

A program may inspect:

• its own classes,
• functions,
• data structures.

It can treat portions of itself as objects.

Reflection turns internal structure into manipulable representation.

Debuggers

A debugger provides another example.

A running program’s:

• memory,
• call stack,
• variables

can be inspected.

But the debugger is usually external or privileged relative to the program being observed.

That extra level simplifies self-description.

Can the Program Debug Itself?

To some extent, yes.

Programs can:

• log state,
• detect errors,
• restart components.

Self-diagnosis is increasingly important in autonomous systems.

But perfect diagnosis is difficult because the diagnostic mechanism is itself part of the system.

Self-Modeling in Biology

Organisms also track themselves.

Homeostatic systems monitor:

• temperature,
• glucose,
• hydration.

The nervous system models:

• body position,
• internal needs,
• injury.

The organism does not need an explicit symbolic self-theory.

Functional self-modeling can be distributed.

The Body Schema

The brain maintains a body schema.

This is an internal representation supporting movement.

It tracks:

• limb position,
• posture,
• spatial relation.

The body schema is not a conscious anatomical diagram.

It is a working model used for control.

Interoception

Interoception is the sensing of internal bodily conditions.

Signals from:

• heart,
• gut,
• lungs

contribute to a model of internal state.

Self-representation begins below explicit thought.

Metacognition

Humans can also represent mental states.

“I know this.”

“I am unsure.”

“I may be mistaken.”

This is metacognition.

The mind evaluates its own cognitive processes.

Confidence

A system may represent not only a conclusion but confidence in that conclusion.

Confidence is a meta-level quantity.

The model includes uncertainty about its own state.

This is more sophisticated than first-order representation.

Theory of Mind Applied Inward

Humans infer other people’s beliefs.

We also infer our own.

Some theories suggest introspection is not direct access but a form of self-interpretation.

The mind may model itself using mechanisms related to social cognition.

Introspection Is Imperfect

People can be mistaken about:

• motives,
• causes of decisions,
• future behavior.

Self-modeling is not transparent.

The system constructing the model is the system being modeled.

Bias is possible.

Prediction of Self

Can you perfectly predict your next decision?

Suppose you predict:

“I will choose left.”

You could then deliberately choose right.

The prediction becomes part of the causal system.

Self-prediction can alter what is predicted.

Predictor Paradox

Imagine a perfectly reliable predictor tells you what you will do.

If you are free to do the opposite, the prediction fails.

If the prediction cannot fail, your response is constrained.

Self-prediction creates feedback between model and target.

Reflexive Prediction

Economic forecasts can behave similarly.

A prediction of a bank failure may trigger withdrawals.

The prediction helps cause the event.

A prediction of success may also change behavior.

Models can become causal when their targets read them.

Performative Models

A model is performative when publishing or using it changes the system being modeled.

Examples include:

• credit scores,
• rankings,
• forecasts.

Self-referential social systems are especially prone to this.

Complete Physical Simulation

Could a system contain a perfect physical simulation of itself?

Suppose the model must include every detail of the system.

But the model itself is inside the system.

Then the model must include itself.

And the copy inside the model must include another copy.

An apparent regress emerges.

Compression Might Help

Perhaps the model need not copy every state literally.

A compact law plus initial conditions may describe the whole system.

This works if the system is highly compressible.

But prediction may still require as much computational work as the system’s own evolution.

Computational Irreducibility

Stephen Wolfram popularized the idea of computational irreducibility:

for some systems, the fastest way to know what happens may be essentially to let the system evolve.

No shortcut gives the full future.

The idea is influential, though exact scope depends on formalization.

The Halting Problem Preview

Computability theory provides stronger limits.

There is no general algorithm that can determine whether every possible program will halt.

A system capable of representing programs therefore faces hard limits in predicting program behavior.

We will study this later.

Gödel Preview

Formal systems also face self-referential limitations.

A sufficiently expressive consistent formal system cannot prove every truth expressible in its own arithmetic language.

Self-description reaches mathematical boundaries.

Again, later chapters will develop this carefully.

Tarski Preview

A sufficiently rich formal language cannot define its own truth predicate in the unrestricted way we might want.

Complete semantic self-description requires stepping outside the system.

Self-reference pushes toward a meta level.

Self-Model vs Self

The model of the system is not the system itself.

This returns to an earlier lesson:

the map is not the territory.

A self-model can guide control without duplicating the full underlying reality.

The Self May Be a Model

Some theories of mind go further.

Perhaps what we call the “self” is partly a model constructed by the brain.

The organism needs a representation of:

• body,
• agency,
• goals,
• continuity.

The self-model may then be experienced as the self.

We will revisit this in the mind section.

A Complete Model May Be Impossible but a Useful One Is Enough

Engineering rarely needs perfect self-description.

A plane needs to know:

• fuel,
• speed,
• orientation.

It does not need a complete quantum description of itself.

Useful intelligence depends on selective models.

The Philosophical Lesson

A system can contain models of itself.

But complete self-representation faces several obstacles:

• compression limits,
• feedback,
• computational limits,
• logical self-reference.

Self-knowledge is possible without omniscience.

The End of Part X

This section began with sentences that referred to themselves.

It moved through:

• self-reference,
• quines,
• self-reproducing machines,
• DNA,
• development,
• self-assembly.

The recurring pattern was a system folding information back into its own operation.

The next part asks a different question:

When we reason about such systems, what makes one conclusion follow from another?

That is the domain of:

logic.