Prediction has always been the point where scientific models meet the world.
A model is built, refined, and tested against what happens next.
But this assumes something quite specific about what “what happens next” is:
that it can be expressed as a determinate outcome of a stable system.
Once we move to modelling relations instead of objects, that assumption quietly stops holding in its original form.
Prediction does not disappear.
It changes shape.
The classical idea of prediction
In the standard framework, prediction has a clear structure:
- a model defines state variables
- laws determine how those variables evolve
- initial conditions fix a trajectory
- the future is a continuation of that trajectory
Prediction is:
the extraction of a future value from a stable representational system
Success means:
- the predicted value matches observation
- within acceptable error bounds
The world is treated as:
something the model tracks from a fixed standpoint
What breaks in a configurational world
Once measurement is understood as configurational, and modelling as relational, this structure becomes unstable.
Because:
- there is no single privileged configuration
- outcomes depend on how systems are coupled
- different setups produce systematically different stabilisations
So the question becomes:
what exactly is being predicted?
Or something else entirely?
From values to distributions of stability
Under relational modelling, what is predicted is no longer a single outcome.
It is:
a structured pattern of possible outcomes across configurations
Prediction becomes about:
- where stability will emerge
- how it will vary with constraints
- what transformations preserve or disrupt it
Instead of:
this is what will happen
we get:
this is the range of stable relations that will appear under these conditions
Prediction as mapping, not point-estimation
The key shift is this:
prediction is no longer the identification of a single future pointbut the mapping of a space of possible stabilisations
This includes:
- regimes where outcomes converge
- regimes where they diverge systematically
- transitions between stable configurations
- sensitivity to specific constraints
The predictive object is no longer a point.
It is a structured field.
Returning to experiments
In a classical experiment:
- prediction is tested by repetition under controlled conditions
- convergence is the criterion of success
In a relational experiment:
- prediction is tested across variations in configuration
- structure of variation is the criterion of success
So success looks like:
- correct mapping of stability domains
- accurate identification of divergence patterns
- reliable transformation between regimes
Not:
one correct value
But:
a correct structure of relations across values
What happens to uncertainty
Uncertainty also changes meaning.
Traditionally:
- uncertainty measures deviation from a true value
Here:
- uncertainty measures sensitivity to configuration
It becomes:
a description of how outcomes depend on changes in conditions
Uncertainty is no longer just a margin around a value.
It is:
a map of structural dependence
The gravitational case revisited
Consider again gravitational measurements.
Under the classical view:
- different experiments should converge to a single value of G
- deviations indicate error
Under relational prediction:
- different experimental setups are expected to produce systematically related results
- deviations indicate structured dependence on configuration
Prediction now means:
specifying how gravitational interaction stabilises across different measurement regimes
We predict not “G,” but:
- the pattern of stability across configurations
- the transformation relations between experimental setups
- the conditions under which convergence appears or fails
Prediction becomes conditional structure
A key shift occurs:
Instead of:
if initial conditions, then outcome
we have:
if configuration, then structure of possible outcomes
This is a more complex but more faithful form of prediction.
Because it explicitly includes:
- apparatus
- coupling
- constraint
- regime
Prediction becomes:
a statement about how relations will organise themselves under specified conditions
Why this is still predictive
This is not a weakening of prediction.
It is a relocation of its target.
We are still making claims about the future.
But those claims are now about:
- patterns
- stability domains
- relational structures
and not only about:
- single numerical outcomes
Importantly, these predictions are:
- testable
- reproducible
- and highly constrained
They are not vague.
They are structured.
From certainty to structured expectation
The classical ideal of prediction often carries an implicit promise:
if the model is correct, the future is fixed
The relational model replaces this with:
if the configuration is specified, the space of possible futures is structured
We do not get less rigor.
We get:
a more explicit account of what rigor is doing
What becomes predictable
Under this framework, we can predict:
- where stability will occur
- how outcomes shift across regimes
- which configurations yield equivalence
- where small changes produce structural divergence
- how different experimental systems relate
This is a richer predictive space.
But it is also more honest about the role of conditions.
Closing
Prediction has not been abandoned.
It has been redefined.
No longer the extraction of a single expected outcome from a fixed system, prediction becomes:
the mapping of how stable relations emerge, transform, and dissolve across structured variation in conditions
This is more demanding.
But it is also more aligned with what experimental practice already reveals when we stop forcing convergence.
At this point, the series completes its internal arc:
- experiments generate structured variation
- misalignment becomes signal
- comparison becomes relational
- modelling becomes configurational
- prediction becomes structural mapping
The remaining question is no longer methodological.
It is epistemic in a deeper sense:
what kind of scientific understanding emerges when stability is no longer assumed, but actively produced, compared, and mapped across conditions?
No comments:
Post a Comment